Cloud simulation automatic modeling method and device, equipment and storage medium

Abstract

The invention relates to the technical field of simulation modeling, and discloses a cloud simulation automatic modeling method, device and equipment and a storage medium, and the method comprises the steps: obtaining original signals from various sources, carrying out the preprocessing of the original signals, carrying out the feature extraction of the preprocessed original signals, and obtaining a target signal; carrying out modeling by adopting adversarial neural network GANs training based on the target signal to obtain a generative AI model; vVamp is adopted; the strategy A verifies each layer in the generative AI model, and continuously adjusts parameters of the generative AI model according to verification results; deploying the optimized generative AI model on a cloud platform, so that a user can access and run a simulation task through an API (Application Program Interface); according to the method, a multi-level automatic modeling and self-verification mechanism is adopted, automatic modeling and self-adaptive optimization are achieved from signals to each level of a system, and a simulation model and VVamp are efficiently generated; the strategy A provides a reliable verification mechanism for the model of each level, and the authenticity and the high efficiency of a simulation result are ensured.

Description

Technical Field

[0001] The present invention relates to the field of simulation modeling technology, and in particular to a cloud simulation automatic modeling method, device, equipment and storage medium. Background Art

[0002] In the traditional simulation modeling process, professional modelers are usually required to manually build the model, which not only takes a lot of time and energy, but also requires high professional knowledge and skills of the modelers. In addition, traditional modeling methods often have problems such as low modeling efficiency, low accuracy, and difficulty in adapting to complex systems. Summary of the invention

[0003] The purpose of the present invention is to solve the above-mentioned problem and to design a cloud simulation automatic modeling method, device, equipment and storage medium.

[0004] A first aspect of the present invention provides a cloud simulation automatic modeling method, the cloud simulation automatic modeling method comprising:

[0005] Acquire original signals from various sources, perform preprocessing operations on the original signals, perform feature extraction on the preprocessed original signals, and obtain target signals, wherein the preprocessing operations at least include cleaning, sorting, and normalization;

[0006] Based on the target signal, adversarial neural network GANs training is used to perform modeling to obtain a generative AI model, wherein the generative AI model at least includes a signal layer, an information layer, a unit layer, a module layer, an equipment layer, a system layer, a process layer, a network layer, and a system layer;

[0007] Using the VV&A strategy to verify each layer of the generative AI model, and continuously adjusting the parameters of the generative AI model based on the verification results;

[0008] The optimized generative AI model is deployed to the cloud platform so that users can access it through the API interface and run simulation tasks.

[0009] Optionally, in a first implementation of the first aspect of the present invention, obtaining original signals from various sources, performing preprocessing operations on the original signals, and performing feature extraction on the preprocessed original signals to obtain target signals includes:

[0010] Acquire original signals from various sources, wherein the sources of the original signals at least include sensors, actuators, processing units, radars, and communication equipment;

[0011] The original signal is subjected to noise removal and missing value filling processing, and the format conversion and normalization processing are performed on the original signal data after the cleaning processing to obtain a preliminary processed signal;

[0012] Decomposing the preliminary processed signal by using variational mode decomposition optimized by subtractive mean optimization algorithm, and screening IMF components by using kurtosis criterion;

[0013] The filtered IMF components are extracted to obtain time domain features, frequency domain features, and time-frequency domain features, and the extracted feature vectors are fused to obtain the target signal.

[0014] Optionally, in a second implementation of the first aspect of the present invention, the decomposing the preliminary processed signal by using variational mode decomposition optimized by a subtraction mean optimization algorithm comprises:

[0015] Get the preliminary processed signal, initialize the VMD parameters, calculate the new position of each search agent in the subtraction mean optimization algorithm,

[0016]

[0017] In the formula, x i Represents an individual, represents a random number between [0,1], N represents the total number of individuals, v represents the scaling factor, X j represents the position of the jth search agent, X i Indicates the location of the current search agent. Indicates the new position after update;

[0018] If the updated position is better, replace the original position, otherwise keep it as it is:

[0019]

[0020] In the formula, and F i Represents the search agent and X i The objective function value of

[0021] During the optimization process, the fitness function is calculated to decompose the signal according to the current VMD parameters to obtain the modal components, and the envelope entropy is used as the fitness function for optimizing the VMD parameters:

[0022]

[0023] Where P j Represents the normalized envelope entropy value of the modal signal, IMF EE (k) represents the envelope entropy of k modal signals;

[0024] Update the VMD parameters based on the calculated fitness function value, and determine whether the preset maximum number of iterations has been reached. If not, continue to use the subtraction mean optimization algorithm to optimize the VMD parameters. If yes, output the optimized VMD parameters.

[0025] The VMD parameters obtained by the subtraction mean optimization algorithm are used to perform variational mode decomposition on the preliminary processed signal to obtain various IMF components.

[0026] Optionally, in a third implementation manner of the first aspect of the present invention, the step of screening the IMF components by using the kurtosis criterion includes:

[0027] Calculate the kurtosis value of each IMF component obtained by variational mode decomposition optimized by the subtraction mean optimization algorithm and the correlation coefficient with the preliminary processed signal:

[0028]

[0029] In the formula, r represents the correlation coefficient, which is used to measure the linear correlation between two variables p and q, and its value range is [-1,1]. i p variable representing the i-th observation, q i The q variable representing the i-th observation, represents the mean value of the p variable, represents the mean value of the q variable, and n represents the total number of observations;

[0030] According to the calculated correlation coefficient, the components with large differences from the preliminary processed signal are screened out, and then according to the calculated kurtosis value, the components with the top three kurtosis values ​​are screened out as representatives;

[0031] The kurtosis and correlation coefficient are combined as the criterion for IMF component screening. According to the calculated kurtosis value and correlation coefficient, the IMF components with characteristic information are screened out.

[0032] Optionally, in a fourth implementation of the first aspect of the present invention, the modeling is performed based on the target signal using adversarial neural network GANs training to obtain a generative AI model, including:

[0033] The target signal is divided into a training set, a validation set and a test set, a confrontation neural network GANs including a generator and a discriminator is constructed, the parameters of the generator and the discriminator are initialized, and the generator and the discriminator are alternately trained using the training set;

[0034] In the generator of the adversarial neural network GANs, a depth-wise separable convolution is used and a residual connection is introduced. Suppose the input of one layer of the generator is e, and the output after transformation is F(e), then the output with the residual connection is d=F(e)+e;

[0035] Construct multiple discriminators of different scales in the adversarial neural network GANs, perform feature matching, and calculate the distance between the generated features and the real features in multiple intermediate feature layers;

[0036] Use the training set for multiple iterative training, use the validation set to evaluate the model performance, adjust the hyperparameters to optimize the model, and obtain a generative AI model.

[0037] Optionally, in a fifth implementation of the first aspect of the present invention, a gradient penalty term is added to constrain the behavior of the discriminator in the adversarial neural network GANs, assuming is the linear interpolation between the real feature x and the generated feature G(z), that is Where ∈~U(0,1);

[0038] The gradient penalty term is The loss function of the discriminator is:

[0039]

[0040] The generator loss function of the adversarial neural network GANs is:

[0041]

[0042] Where, L D represents the loss function of the discriminator, L G represents the loss function of the generator, represents the expectation of the discriminator's discrimination result on the real feature, x represents the real feature sample, P r represents the real feature distribution, D(x) represents the discriminant result of the real feature, represents the expectation of the discriminator on the discriminant result of the generated feature, z represents the input noise of the generator, P z represents the distribution of the generator input noise, G(z) represents the features generated by the generator, D(G(z)) represents the discriminant result of the generated features, λ represents the weight of the gradient penalty term, Indicates that the discriminator is The gradient at .

[0043] Optionally, in a sixth implementation of the first aspect of the present invention, the adopting of the VV&A strategy to verify each layer in the generative AI model, and continuously adjusting the parameters of the generative AI model according to the verification results, includes:

[0044] When there is an error between the output of the generative AI model in the verification result and the expected output, the gradient is calculated based on the error, and then the model parameters are adjusted in the opposite direction of the gradient:

[0045] W l =Wl -αδ l (x l ) T

[0046] b l =b l -αδ l

[0047] Where W l represents the weight matrix of the lth layer, b l represents the bias vector of the lth layer, α represents the learning rate used to control the step size of parameter update, and δ l represents the error term of the lth layer, x l represents the input vector of the lth layer, (x l ) T Represents the transpose of the input vector of the lth layer.

[0048] A second aspect of the present invention provides a cloud simulation automatic modeling device, the cloud simulation automatic modeling device comprising:

[0049] A feature extraction module is used to obtain original signals from various sources, perform preprocessing operations on the original signals, perform feature extraction on the preprocessed original signals, and obtain target signals, wherein the preprocessing operations at least include cleaning, sorting and normalization;

[0050] A model generation module, used to perform modeling based on the target signal using adversarial neural network GANs training to obtain a generative AI model, wherein the generative AI model includes at least a signal layer, an information layer, a unit layer, a module layer, an equipment layer, a system layer, a process layer, a network layer, and a system layer;

[0051] A parameter adjustment module, used to verify each layer of the generative AI model using a VV&A strategy, and continuously adjust the parameters of the generative AI model according to the verification results;

[0052] The simulation deployment module is used to deploy the optimized generative AI model to the cloud platform so that users can access and run simulation tasks through the API interface.

[0053] The third aspect of the present invention provides a cloud simulation automatic modeling device, which includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the cloud simulation automatic modeling device executes each step of the cloud simulation automatic modeling method as described in any one of the above items.

[0054] A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, and when the instructions are executed by a processor, the various steps of the cloud simulation automatic modeling method as described in any of the above items are implemented.

[0055] In the technical solution provided by the present invention, original signals are obtained from various sources, and preprocessing operations are performed on the original signals, and feature extraction is performed on the preprocessed original signals to obtain target signals; based on the target signals, adversarial neural network GANs training is used for modeling to obtain a generative AI model; the VV&A strategy is used to verify each layer in the generative AI model, and the parameters of the generative AI model are continuously adjusted according to the verification results; the generative AI model obtained after tuning is deployed to a cloud platform for users to access and run simulation tasks through an API interface; the present invention adopts a multi-level automatic modeling and self-verification mechanism, and realizes automatic modeling and adaptive optimization at each level from signals to systems. It can efficiently generate simulation models without human intervention, and optimize and adjust through real-time feedback, ensuring the accuracy, flexibility and scalability of the modeling process. At the same time, the VV&A strategy provides a reliable verification mechanism for each level of the model to ensure the authenticity and efficiency of the simulation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the invention.

[0057] Figure 1 A schematic diagram of a first embodiment of a cloud simulation automatic modeling method provided by an embodiment of the present invention;

[0058] Figure 2 A schematic diagram of a second embodiment of the cloud simulation automatic modeling method provided by an embodiment of the present invention;

[0059] Figure 3 A schematic diagram of a third embodiment of the cloud simulation automatic modeling method provided by an embodiment of the present invention;

[0060] Figure 4 A schematic diagram of the structure of a cloud simulation automatic modeling device provided by an embodiment of the present invention;

[0061] Figure 5 A schematic diagram of the structure of a cloud simulation automatic modeling device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0063] The present invention proposes an automatic modeling method based on the combination of generative artificial intelligence (Generative AI) and cloud simulation. Without the need for human intervention, it covers automatic modeling at multiple levels, from signals to systems, from units to networks, from equipment to processes. The modeling method at each level is based on the support of generative AI technology and cloud computing platform, which can achieve high adaptability and intelligence, and has self-verification, verification and confirmation (VV&A) capabilities.

[0064] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 A schematic diagram of a first embodiment of a cloud simulation automatic modeling method provided by an embodiment of the present invention, the method specifically comprises the following steps:

[0065] Step 101, obtaining original signals from various sources, performing preprocessing operations on the original signals, performing feature extraction on the preprocessed original signals, and obtaining target signals, wherein the preprocessing operations at least include cleaning, sorting, and normalization;

[0066] Step 102: Based on the target signal, the adversarial neural network GANs training is used to perform modeling to obtain a generative AI model, wherein the generative AI model at least includes a signal layer, an information layer, a unit layer, a module layer, an equipment layer, a system layer, a process layer, a network layer, and a system layer;

[0067] In this embodiment, signal layer modeling: In the signal layer modeling process, the system first collects the original data related to the target signal (such as frequency, amplitude, phase, etc.), and uses generative artificial intelligence technology to learn and model the characteristics of the signal. Generative AI can generate simulated signals similar to actual signals by learning the distribution laws and behavior patterns of real signals. By training the model, a variety of signal scenarios can be accurately generated and adjusted according to the simulation environment; the VV&A strategy of the signal layer is mainly reflected in two aspects: first, based on the adaptive adjustment of the signal generation process, generative AI optimizes the generated signal according to real-time feedback to ensure the accuracy of the simulated signal and conformity with actual environmental changes. Secondly, the matching degree of the actual signal and the simulated signal is compared, and statistical methods are used for comparison and verification to ensure the authenticity of the generated signal.

[0068] Information layer modeling: Information layer modeling mainly models the process of converting signals into information. The goal of the information layer is to map the original signal to higher-level semantic information through generative AI. For example, the signal received by the sensor can be converted into corresponding tactical data, monitoring data or other types of information through generative models. Generative AI can automatically infer the meaning behind the signal and perform effective encoding and decoding processing by learning from large amounts of data and historical records.

[0069] Unit-level modeling: At the unit level, generative AI models the functions and performance of physical units (such as sensors, actuators, processing units, etc.) to form a virtualized model. Generative AI can model the input-output relationship of physical units and automatically generate response patterns of different physical units in specific situations. Through training on large-scale data sets, the model can automatically identify the characteristics and working rules of different physical units and generate unit models suitable for simulation systems.

[0070] Module-level modeling: Module-level modeling is based on the unit layer, combining multiple physical units into complex modules (such as communication modules, processing modules, control modules, etc.), and modeling the interactions and functions of the modules through generative AI. Generative AI can automatically deduce the functions, interfaces and their relationships of modules without human intervention through deep learning of historical data, thereby generating an accurate modular simulation model.

[0071] Equipment layer modeling: At the equipment layer, generative AI automatically builds equipment layer models (such as radar, communication equipment, sensor systems, etc.) by integrating multiple modules. The equipment model will involve the interaction of multiple modules, including coordination and collaboration between devices. With the support of the cloud simulation platform, generative AI can simulate the working process of the equipment layer and automatically deduce the performance of equipment in different scenarios.

[0072] System-level modeling: Based on the equipment-level model, a comprehensive system simulation model (such as tactical command system, UAV system, etc.) is constructed by combining a variety of equipment, system modules, and technical solutions. In this process, generative AI can automatically identify the complex behaviors and multi-party collaboration processes at the system level by modeling and integrating the various components of the system, thereby generating a system model suitable for simulation and testing.

[0073] Process layer modeling: mainly refers to modeling the entire workflow or operation process based on the system layer. For example, in the fields of military command, production and manufacturing, process layer modeling can automatically derive and optimize the task execution process. Generative AI automatically generates workflows that adapt to different conditions based on historical data, task requirements and real-time feedback, and continuously adjusts process steps according to task objectives.

[0074] Network layer modeling: involves modeling the network relationships of devices, systems, information flows, control flows, and other elements. Generative AI can automatically build network models that adapt to different scenarios by learning network topology, data transmission, and information exchange, and optimize important factors such as data flow, bandwidth allocation, and routing selection, thereby forming an efficient network simulation model.

[0075] System-level modeling: Integrate models at all levels to form a complete system simulation model. Generative AI can not only independently model each level, but also automatically integrate models at all levels to generate a comprehensive simulation model covering multiple levels such as signals, information, equipment, systems, processes, and networks. The system-level model can reflect the complex behavior of the entire system and perform global optimization in different scenarios.

[0076] Step 103: Use the VV&A strategy to verify each layer in the generative AI model, and continuously adjust the parameters of the generative AI model according to the verification results;

[0077] In this embodiment, the VV&A strategy of the information layer includes two aspects: one is to verify the conversion logic between signals and information to ensure that the generated process follows the correct conversion rules and has reliability; the other is to verify the consistency of information and compare it with known rules or data sets to ensure that the generated information matches the actual information; the VV&A strategy of the unit layer mainly ensures that the behavior of each virtual unit is consistent with its real counterparts by comparing with the test results of the real unit. In addition, the model is continuously adjusted and optimized through the online feedback mechanism to ensure its long-term stability and accuracy; the VV&A strategy of the module layer includes the verification of module functions and the verification of interfaces between modules. The modules generated by the generative AI model will be interactively tested with the actual modules. By simulating the data flow and instruction flow between modules, the correctness and coordination of the module layer are verified; the VV&A strategy of the equipment layer includes equipment function verification, performance verification and interoperability verification between equipment. By simulating the operation of equipment in different environments, the VV&A strategy of the equipment layer includes equipment function verification, performance verification and interoperability verification between equipment. The VV&A strategy at the system level focuses on the performance verification and effectiveness evaluation of the entire system. Through large-scale parallel simulation on the cloud simulation platform, the functions and coordination of each subsystem are verified to ensure the work efficiency and reliability of the entire system. The VV&A strategy at the process level includes the logic verification and operation sequence verification of the process. Generative AI can ensure that each stage and each operation of the process meets the predetermined goals and matches the actual task requirements by simulating the execution process of the task process. The VV&A strategy at the network level includes the verification of the network architecture and the verification of the network behavior. Through simulation, the communication effect and information flow path of each node in the network are verified to ensure that the generated network model has high accuracy and reliability in the actual environment. The VV&A strategy at the system level focuses on the overall performance verification and global optimization verification of the system. Through multi-scenario and multi-condition full-system simulation, it ensures that the overall coordination, stability and performance requirements of models at all levels meet the application requirements.

[0078] Step 104: deploy the optimized generative AI model to the cloud platform so that users can access and run simulation tasks through the API interface.

[0079] In the embodiment of the present invention, a high degree of automation of simulation modeling is achieved through generative artificial intelligence technology. The generative AI model can automatically generate simulation models suitable for different scenarios based on input data and adjust them according to real-time feedback. Compared with traditional manual intervention modeling, it can automatically generate new scenarios and scenario parameters, greatly improving modeling efficiency and reducing expert intervention. Automated modeling also improves the iteration speed of the simulation model, and can quickly adapt to changing application scenarios to meet dynamic needs. By combining generative AI with cloud simulation technology, simulation models and parameters can be adjusted according to real-time feedback and environmental changes. Generative AI has strong adaptive capabilities and can dynamically generate new simulation scenarios according to the operating status of the system and user needs, and self-optimize existing scenarios. This adaptability ensures the flexibility of simulation modeling, enabling the model to cope with complex and dynamically changing simulation tasks, avoiding the mismatch problem caused by environmental changes in traditional modeling methods. It can not only perform large-scale parallel simulation through the cloud computing platform and make full use of cloud resources, but also optimize the system through the generative AI model. The simulation calculation process is simplified. The efficient computing power of the cloud platform can support the application of generative AI in different simulation tasks, reducing computing resource consumption and training time. In real-time simulation, generative AI can dynamically adjust the computing allocation of simulation tasks as needed, thereby effectively improving computing efficiency and reducing resource waste. Compared with traditional deep learning methods, generative AI can reduce its dependence on large-scale annotated data sets by learning existing data and generating new simulation scenarios. In the case of insufficient data, generative AI can generate effective simulation models through unsupervised learning or a small number of samples, which improves the generalization ability of the model. In addition, the flexibility of the generative AI model enables it to adapt to different simulation tasks across fields, and has strong universality and migration capabilities. Unlike black box models such as deep learning, the generative AI model in the present invention can provide a more transparent and explainable modeling process by combining rule generation with feedback optimization. In multiple decision-making links, generative AI can provide explanations for model changes, help users understand the modeling process, and enhance the credibility of the system and the transparency of decision-making. This is particularly important in fields that require high reliability and credibility (such as military, aviation, and medical care).

[0080] See also Figure 2 , a schematic diagram of a second embodiment of a cloud simulation automatic modeling method provided by an embodiment of the present invention, the method comprising:

[0081] Step 201, obtaining original signals from various sources, wherein the sources of the original signals at least include sensors, actuators, processing units, radars, and communication equipment;

[0082] Step 202: remove noise from the original signal and fill in missing values, perform format conversion and normalization on the cleaned original signal data to obtain a preliminary processed signal;

[0083] Step 203, decomposing the preliminary processed signal by using variational mode decomposition optimized by subtraction mean optimization algorithm, and screening IMF components by using kurtosis criterion;

[0084] In this embodiment, the preliminary processed signal is decomposed using variational mode decomposition optimized by subtraction mean optimization algorithm, including:

[0085] Get the preliminary processed signal, initialize the VMD parameters, calculate the new position of each search agent in the subtraction mean optimization algorithm,

[0086]

[0087] In the formula, x i Represents an individual, represents a random number between [0,1], N represents the total number of individuals, v represents the scaling factor, X j represents the position of the jth search agent, X i Indicates the location of the current search agent. Indicates the new position after update;

[0088] If the updated position is better, replace the original position, otherwise keep it as it is:

[0089]

[0090] In the formula, and F i Represents the search agent and X i The objective function value of

[0091] During the optimization process, the fitness function is calculated to decompose the signal according to the current VMD parameters to obtain the modal components, and the envelope entropy is used as the fitness function for optimizing the VMD parameters:

[0092]

[0093] Where P j Represents the normalized envelope entropy value of the modal signal, IMF EE (k) represents the envelope entropy of k modal signals;

[0094] Update the VMD parameters based on the calculated fitness function value, and determine whether the preset maximum number of iterations has been reached. If not, continue to use the subtraction mean optimization algorithm to optimize the VMD parameters. If yes, output the optimized VMD parameters.

[0095] The VMD parameters obtained by the subtraction mean optimization algorithm are used to perform variational mode decomposition on the preliminary processed signal to obtain the IMF components.

[0096] In this embodiment, the IMF components are screened using the kurtosis criterion, including:

[0097] Calculate the kurtosis value of each IMF component obtained by variational mode decomposition optimized by the subtraction mean optimization algorithm and the correlation coefficient with the preliminary processed signal:

[0098]

[0099] In the formula, r represents the correlation coefficient, which is used to measure the linear correlation between two variables p and q, and its value range is [-1,1]. i p variable representing the i-th observation, q i The q variable representing the i-th observation, represents the mean value of the p variable, represents the mean value of the q variable, and n represents the total number of observations;

[0100] According to the calculated correlation coefficient, the components with large differences from the preliminary processed signal are screened out, and then according to the calculated kurtosis value, the components with the top three kurtosis values ​​are screened out as representatives;

[0101] The kurtosis and correlation coefficient are combined as the criterion for IMF component screening. According to the calculated kurtosis value and correlation coefficient, the IMF components with characteristic information are screened out.

[0102] Step 204: extract the filtered IMF components to obtain time domain features, frequency domain features, and time-frequency domain features, and fuse the extracted feature vectors to obtain the target signal.

[0103] See also Figure 3 , a schematic diagram of a third embodiment of a cloud simulation automatic modeling method provided by an embodiment of the present invention, the method comprising:

[0104] Step 301: Divide the target signal into a training set, a validation set, and a test set, construct an adversarial neural network GANs including a generator and a discriminator, initialize the parameters of the generator and the discriminator, and use the training set to alternately train the generator and the discriminator;

[0105] Step 302: adopting depthwise separable convolution and introducing residual connection in the generator of the adversarial neural network GANs;

[0106] In this embodiment, assuming that the input of one layer of the generator is e, and the output after transformation is F(e), then the output with residual connection is d=F(e)+e;

[0107] Step 303: construct multiple discriminators of different scales in the adversarial neural network GANs, perform feature matching, and calculate the distance between the generated features and the real features in multiple intermediate feature layers;

[0108] Step 304: Use the training set to perform multiple iterations of training, use the validation set to evaluate the model performance, adjust the hyperparameters to optimize the model, and obtain a generative AI model.

[0109] In this embodiment, a gradient penalty term is added to constrain the behavior of the discriminator in the adversarial neural network GANs. is the linear interpolation between the real feature x and the generated feature G(z), that is Where ∈~U(0,1);

[0110] The gradient penalty term is The loss function of the discriminator is:

[0111]

[0112] The generator loss function of the adversarial neural network GANs is:

[0113]

[0114] Where, L D represents the loss function of the discriminator, L G represents the loss function of the generator, represents the expectation of the discriminator's discrimination result on the real feature, x represents the real feature sample, P r represents the real feature distribution, D(x) represents the discriminant result of the real feature, represents the expectation of the discriminator on the discriminant result of the generated feature, z represents the input noise of the generator, P z represents the distribution of the generator input noise, G(z) represents the features generated by the generator, D(G(z)) represents the discriminant result of the generated features, λ represents the weight of the gradient penalty term, Indicates that the discriminator is The gradient at .

[0115] In this embodiment, when there is an error between the output of the generative AI model in the verification result and the expected output, the gradient is calculated based on the error, and then the model parameters are adjusted in the opposite direction of the gradient:

[0116] W l =W l -αδ l (x l ) T

[0117] b l =bl -αδ l

[0118] Where W l represents the weight matrix of the lth layer, b l represents the bias vector of the lth layer, α represents the learning rate used to control the step size of parameter update, and δ l represents the error term of the lth layer, x l represents the input vector of the lth layer, (x l ) T Represents the transpose of the input vector of the lth layer.

[0119] See also Figure 4 , a schematic diagram of the structure of a cloud simulation automatic modeling device provided by an embodiment of the present invention, the device comprises:

[0120] The feature extraction module 401 is used to obtain original signals from various sources, perform preprocessing operations on the original signals, and perform feature extraction on the preprocessed original signals to obtain target signals, wherein the preprocessing operations at least include cleaning, sorting and normalization;

[0121] A model generation module 402 is used to perform modeling based on a target signal using adversarial neural network GANs training to obtain a generative AI model, wherein the generative AI model includes at least a signal layer, an information layer, a unit layer, a module layer, an equipment layer, a system layer, a process layer, a network layer, and a system layer;

[0122] A parameter adjustment module 403, used to verify each layer in the generative AI model using the VV&A strategy, and continuously adjust the parameters of the generative AI model according to the verification results;

[0123] The simulation deployment module 404 is used to deploy the optimized generative AI model to the cloud platform so that users can access and run simulation tasks through the API interface.

[0124] In the embodiment of the present invention, = generative artificial intelligence technology is combined with a cloud simulation platform to achieve multi-level automated modeling from signal to system. Different from traditional modeling based on artificial rules or templates, the simulation model is automatically generated and optimized through deep learning and generative models, which can effectively improve the modeling efficiency and accuracy, and is particularly suitable for complex and changeable simulation scenarios; the model can be automatically optimized according to real-time feedback without human intervention, and generative artificial intelligence can self-learn and adjust the generated model, so that the simulation system can automatically adapt to different environmental changes, improving the flexibility and accuracy of the model; from the multi-level automatic modeling framework of "signal-information-unit-module-equipment-system-process-network-system", each level Both use generative AI technology for modeling and refine it layer by layer through adaptive optimization and feedback mechanisms. This hierarchical modeling framework can handle simulation tasks of different complexities and dynamic changes and has wide application adaptability. Unlike manual verification and confirmation in traditional simulation methods, a self-verification, verification and confirmation (VV&A) strategy is adopted to self-optimize and verify the model at each level through a real-time feedback mechanism. Through the feedback loop, not only the accuracy of the model can be improved, but also the robustness of the system can be enhanced to ensure the reliability of the generated model. During the simulation process, self-correction is carried out by continuously receiving feedback from the external environment. This mechanism not only enhances the adaptability of the simulation model, but also improves its accuracy and robustness in a dynamic environment.

[0125] above Figure 4 The cloud simulation automatic modeling device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The cloud simulation automatic modeling device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0126] Figure 5 It is a structural schematic diagram of a cloud simulation automatic modeling device provided by an embodiment of the present invention. The cloud simulation automatic modeling device 600 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 610 (for example, one or more processors) and a memory 620, and one or more storage media 630 (for example, one or more massive storage devices) storing application programs 633 or data 632. Among them, the memory 620 and the storage medium 630 can be short-term storage or persistent storage. The program stored in the storage medium 630 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the cloud simulation automatic modeling device 600. Furthermore, the processor 610 can be configured to communicate with the storage medium 630, and execute a series of instruction operations in the storage medium 630 on the cloud simulation automatic modeling device 600 to implement the method provided by the above implementation.

[0127] The cloud simulation automatic modeling device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input and output interfaces 660, and / or one or more operating devices 631, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 5 The cloud simulation automatic modeling device structure shown does not constitute a limitation on the computer device provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0128] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the various steps of the cloud simulation automatic modeling method provided in the above embodiments.

[0129] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described equipment, devices, and units can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0130] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0131] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cloud simulation automatic modeling method, characterized in that: The cloud simulation automatic modeling method comprises: Acquire original signals from various sources, perform preprocessing operations on the original signals, perform feature extraction on the preprocessed original signals, and obtain target signals, wherein the preprocessing operations at least include cleaning, sorting, and normalization; Based on the target signal, adversarial neural network GANs training is used to perform modeling to obtain a generative AI model, wherein the generative AI model at least includes a signal layer, an information layer, a unit layer, a module layer, an equipment layer, a system layer, a process layer, a network layer, and a system layer; Using the VV&A strategy to verify each layer of the generative AI model, and continuously adjusting the parameters of the generative AI model based on the verification results; The optimized generative AI model is deployed to the cloud platform so that users can access it through the API interface and run simulation tasks.

2. A cloud simulation automatic modeling method according to claim 1, characterized in that: The method of obtaining original signals from various sources, performing preprocessing operations on the original signals, and extracting features from the preprocessed original signals to obtain target signals includes: Acquire original signals from various sources, wherein the sources of the original signals at least include sensors, actuators, processing units, radars, and communication equipment; The original signal is subjected to noise removal and missing value filling processing, and the format conversion and normalization processing are performed on the original signal data after the cleaning processing to obtain a preliminary processed signal; Decomposing the preliminary processed signal by using variational mode decomposition optimized by subtractive mean optimization algorithm, and screening IMF components by using kurtosis criterion; The filtered IMF components are extracted to obtain time domain features, frequency domain features, and time-frequency domain features, and the extracted feature vectors are fused to obtain the target signal.

3. A cloud simulation automatic modeling method as claimed in claim 2, characterized in that: Decomposing the preliminary processed signal by using variational mode decomposition optimized by subtraction mean optimization algorithm includes: Get the preliminary processed signal, initialize the VMD parameters, calculate the new position of each search agent in the subtraction mean optimization algorithm, In the formula, x i Represents an individual, represents a random number between [0,1], N represents the total number of individuals, v represents the scaling factor, X j represents the position of the jth search agent, X i Indicates the location of the current search agent. Indicates the new position after update; If the updated position is better, replace the original position, otherwise keep it as it is: In the formula, and F i Represents the search agent and X i The objective function value of During the optimization process, the fitness function is calculated to decompose the signal according to the current VMD parameters to obtain the modal components, and the envelope entropy is used as the fitness function for optimizing the VMD parameters: Where P j Represents the normalized envelope entropy value of the modal signal, IMF EE (k) represents the envelope entropy of k modal signals; Update the VMD parameters based on the calculated fitness function value, and determine whether the preset maximum number of iterations has been reached. If not, continue to use the subtraction mean optimization algorithm to optimize the VMD parameters. If yes, output the optimized VMD parameters. The VMD parameters obtained by the subtraction mean optimization algorithm are used to perform variational mode decomposition on the preliminary processed signal to obtain various IMF components.

4. A cloud simulation automatic modeling method as claimed in claim 2, characterized in that: The method of screening the IMF components by using the kurtosis criterion includes: Calculate the kurtosis value of each IMF component obtained by variational mode decomposition optimized by the subtraction mean optimization algorithm and the correlation coefficient with the preliminary processed signal: In the formula, r represents the correlation coefficient, which is used to measure the linear correlation between two variables p and q, and its value range is [-1,1]. i p variable representing the i-th observation, q i The q variable representing the i-th observation, represents the mean value of the p variable, represents the mean value of the q variable, and n represents the total number of observations; According to the calculated correlation coefficient, the components with large differences from the preliminary processed signal are screened out, and then according to the calculated kurtosis value, the components with the top three kurtosis values ​​are screened out as representatives; The kurtosis and correlation coefficient are combined as the criterion for IMF component screening. According to the calculated kurtosis value and correlation coefficient, the IMF components with characteristic information are screened out.

5. The cloud simulation automatic modeling method according to claim 1, characterized in that: The method of modeling based on the target signal using adversarial neural network GANs training to obtain a generative AI model includes: Dividing the target signal into a training set, a validation set, and a test set, constructing an adversarial neural network GANs including a generator and a discriminator, initializing parameters of the generator and the discriminator, and alternately training the generator and the discriminator using the training set; In the generator of the adversarial neural network GANs, a depth-wise separable convolution is used and a residual connection is introduced. Suppose the input of one layer of the generator is e, and the output after transformation is F(e), then the output with the residual connection is d=F(e)+e; Construct multiple discriminators of different scales in the adversarial neural network GANs, perform feature matching, and calculate the distance between the generated features and the real features in multiple intermediate feature layers; Use the training set for multiple iterative training, use the validation set to evaluate the model performance, adjust the hyperparameters to optimize the model, and obtain a generative AI model.

6. A cloud simulation automatic modeling method as claimed in claim 5, characterized in that: The target signal is divided into a training set, a validation set and a test set, a confrontation neural network GANs including a generator and a discriminator is constructed, parameters of the generator and the discriminator are initialized, and the generator and the discriminator are alternately trained using the training set, including: In the adversarial neural network GANs, a gradient penalty term is added to constrain the behavior of the discriminator. is the linear interpolation between the real feature x and the generated feature G(z), that is Where ∈~U(0,1); The gradient penalty term is The loss function of the discriminator is: The generator loss function of the adversarial neural network GANs is: Where, L D represents the loss function of the discriminator, L G represents the loss function of the generator, represents the expectation of the discriminator's discrimination result on the real feature, x represents the real feature sample, P r represents the real feature distribution, D(x) represents the discriminant result of the real feature, represents the expectation of the discriminator on the discriminant result of the generated feature, z represents the input noise of the generator, P z represents the distribution of the generator input noise, G(z) represents the features generated by the generator, D(G(z)) represents the discriminant result of the generated features, λ represents the weight of the gradient penalty term, Indicates that the discriminator is The gradient at .

7. A cloud simulation automatic modeling method according to claim 1, characterized in that: The VV&A strategy is used to verify each layer of the generative AI model, and the parameters of the generative AI model are continuously adjusted according to the verification results, including: When there is an error between the output of the generative AI model in the verification result and the expected output, the gradient is calculated based on the error, and then the model parameters are adjusted in the opposite direction of the gradient: W l =W l -αδ l (x l ) T b l =b l -ad l Where W l represents the weight matrix of the lth layer, b l represents the bias vector of the lth layer, α represents the learning rate used to control the step size of parameter update, and δ l represents the error term of the lth layer, x l represents the input vector of the lth layer, (x l ) T Represents the transpose of the input vector of the lth layer.

8. A cloud simulation automatic modeling device, characterized in that: The cloud simulation automatic modeling device comprises: A feature extraction module is used to obtain original signals from various sources, perform preprocessing operations on the original signals, perform feature extraction on the preprocessed original signals, and obtain target signals, wherein the preprocessing operations at least include cleaning, sorting and normalization; A model generation module, used to perform modeling based on the target signal using adversarial neural network GANs training to obtain a generative AI model, wherein the generative AI model includes at least a signal layer, an information layer, a unit layer, a module layer, an equipment layer, a system layer, a process layer, a network layer, and a system layer; A parameter adjustment module, used to verify each layer of the generative AI model using a VV&A strategy, and continuously adjust the parameters of the generative AI model according to the verification results; The simulation deployment module is used to deploy the optimized generative AI model to the cloud platform so that users can access and run simulation tasks through the API interface.

9. A cloud simulation automatic modeling device, characterized in that: The cloud simulation automatic modeling device includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the cloud simulation automatic modeling device executes each step of the cloud simulation automatic modeling method as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the cloud simulation automatic modeling method as described in any one of claims 1 to 7 are implemented.

Images