Simulation data monitoring and analysis system and method based on artificial intelligence

Abstract

The invention discloses a simulation data monitoring and analysis system and method based on artificial intelligence, and belongs to the field of big data and artificial intelligence, and the system comprises an intelligent sensing module which is used for collecting multi-source data fusion; and the dynamic modeling module is used for constructing an evolutionary simulation model and realizing self-adaptive evolution of the simulation model through a multi-level technical architecture. According to the invention, the equipment abnormity identification accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the field of big data and artificial intelligence technology, and in particular to an artificial intelligence-based simulation data monitoring and analysis system and method. Background Art

[0002] Traditional industrial equipment monitoring systems often rely on threshold alarms based on single sensor data, resulting in high false alarm rates and difficulty detecting hidden faults. Existing simulation data monitoring solutions typically employ centralized processing architectures, facing challenges such as poor real-time performance and high bandwidth requirements. Furthermore, they lack dual verification mechanisms based on physical laws and statistical data, making root cause analysis difficult. Especially under complex operating conditions, traditional methods struggle to promptly detect progressive anomalies such as equipment performance degradation. Summary of the Invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes an artificial intelligence-based simulation data monitoring and analysis system and method, which can achieve predictive maintenance of equipment anomalies.

[0004] On the one hand, an embodiment of the present invention provides an artificial intelligence-based simulation data monitoring and analysis system, including: an intelligent perception module for collecting multi-source data fusion, including: a heterogeneous protocol adaptation unit for adopting an input-output two-layer conversion architecture, wherein the input layer identifies the message type through a pre-trained protocol classifier and deconstructs it into a JSON Schema intermediate model, and the output layer implements protocol conversion through a dynamic field mapping table; a spatiotemporal alignment unit for implementing microsecond-level data alignment using the PTP time synchronization protocol; a data quality assessment unit for detecting abnormal sampling values ​​in real time based on an isolation forest algorithm; a dynamic modeling module for constructing an evolvable simulation model, and implementing adaptive evolution of the simulation model through a multi-level technical architecture; including: a parameterized modeling unit for encoding device geometric features and physical properties into programmable parameter vectors, wherein the geometric feature encoding is based on an MLP network, the geometric constraints adopt the Lagrange multiplier method, and support automatic parameter optimization through error gradient back propagation, the formula is as follows: ; Where y is the physical quantity measurement value of the simulation target, θ is the trainable parameter vector, is a neural network prediction model, g(X) is a physical constraint function, and λ∈(0,1) is a constraint strength coefficient. The physical engine coupling unit is used to integrate the outputs of different solvers through a multi-scale modeling framework, adopt a dynamic time warping algorithm to achieve cross-solver time step synchronization, and eliminate cross-domain simulation deviations based on a weighted fusion algorithm based on residual analysis. The weights of the weighted fusion algorithm are: ;in, is the fusion weight of the i-th solver, is the residual vector, Represents the Euclidean norm, and the weight range satisfies The model validation unit is used for global sensitivity analysis of variance decomposition based on a sliding window. The window length is automatically adjusted according to the coefficient of variation of the parameters, dynamically identifying key parameters and optimizing the allocation of computing resources. The formula for automatic adjustment of the window length is as follows: ; Where L(t) is the window length at time t, is the coefficient of variation, which is the ratio of the standard deviation to the mean and is used to quantify the degree of dispersion of the parameter; are the maximum and minimum window lengths, respectively; k is the slope factor of the sigmoid function, which controls the sensitivity of window length changes; s0 is the critical threshold of the coefficient of variation; k and s0 are preset parameters obtained based on cross-validation of historical simulation data sets.

[0005] According to some embodiments of the present invention, the data quality assessment unit is used to receive a multidimensional time series data stream from a sensor or control system as input, perform standardization preprocessing on the data to eliminate dimensional differences, and construct a set of randomly divided isolated trees; identify outliers through the path length of the tree structure of the isolated tree, output as an anomaly score with confidence, and mark the abnormal time point and feature dimension.

[0006] According to some embodiments of the present invention, the parametric modeling unit is used to input three-dimensional geometric topology data and physical property data of the device, decompose the geometric structure into hierarchical parameter groups through a feature extraction network, and quantify the physical properties into differentiable variables; generate embedding vectors for the geometric features through a graph neural network, convert the physical properties into a standardized parameter space through a normalization layer, and output a programmable parameter vector of unified dimension, each dimension establishing a differentiable mapping relationship with the device performance indicator; when real-time sensor data is received, the error gradient between the predicted output and the actual measurement is calculated, and the adjustment value is backpropagated along each dimension of the parameter vector to achieve automatic parameter optimization.

[0007] According to some embodiments of the present invention, the physics engine coupling unit is configured to input the spatiotemporal discrete data output by each physics field solver, establish a scale correlation matrix, and map the simulation data of different resolutions to a unified reference frame. A weighted fusion algorithm is then employed to dynamically adjust the fusion weights based on the confidence metrics of each solver, where the confidence metrics are verified through residual analysis and energy conservation. The output is composite physics field data with scale-free data and a deviation rate controlled within a preset tolerance.

[0008] According to some embodiments of the present invention, the model verification unit is used to input a parameter space sampling data set, construct a Sobol index calculation framework, and decompose the total variance into individual effects of each parameter and cross-effect terms; use the Monte Carlo integration method to calculate the first-order and high-order sensitivity indicators, and quantify the importance of parameters through variance contribution rate; output a parameter sensitivity ranking list and confidence interval, where parameters whose contribution rate exceeds a threshold are marked as key parameters; based on the parameter sensitivity ranking list, confidence interval and key parameters, dynamically adjust the allocation of computing resources: increase the sampling density and number of iterations for key parameters, and reduce the calculation accuracy requirements for non-key parameters.

[0009] According to some embodiments of the present invention, the edge computing module includes: a streaming storage unit for adopting a time-series-sensitive cache management strategy, wherein the cache management strategy includes: event-triggered data coverage rules, a dynamically adjusted window size mechanism, and multi-priority data shard storage.

[0010] According to some embodiments of the present invention, the edge computing module includes: a streaming storage unit for adopting a time-series-sensitive cache management strategy, wherein the cache management strategy includes: event-triggered data coverage rules, a dynamically adjusted window size mechanism, and multi-priority data shard storage.

[0011] According to some embodiments of the present invention, the edge computing module includes: a communication optimization unit for realizing compression and selective uploading of transmitted data, including: adaptive adjustment of encoding parameters based on content characteristics, differential transmission of key frames and non-key frames, and network status-aware bit rate control.

[0012] According to some embodiments of the present invention, the system further comprises: an intelligent warning module for dynamically generating a graded warning signal based on simulation data, including: a composite event detection unit for performing multi-dimensional abnormal correlation analysis:

[0013] ;

[0014] in, is the deviation degree of the physical constraint of type i, which is calculated by the Hellinger distance between the real-time simulation value and the theoretical value; is the statistical anomaly score, which is obtained by normalizing the standard deviation of the residual distribution output by the dynamic modeling module; α, β∈[0,1] are weight coefficients, satisfying α+β=1, and τ is the dynamically adjusted expert experience threshold.

[0015] An artificial intelligence-based simulation data monitoring and analysis system according to an embodiment of the present invention includes at least the following beneficial effects: the embodiment of the present invention achieves three major breakthroughs through multi-source heterogeneous data fusion of the intelligent perception module, physical-data hybrid driven analysis of the dynamic modeling module, distributed real-time processing of the edge computing module, and multi-dimensional abnormal correlation detection of the intelligent early warning module: 1) improving the accuracy of equipment abnormality identification; 2) reducing cloud bandwidth occupancy through lightweight processing on the edge side; 3) supporting closed-loop diagnosis from data anomalies to physical object failures.

[0016] Another embodiment of the present invention provides an artificial intelligence-based simulation data monitoring and analysis method, comprising the following steps: S100, collecting multi-source data fusion, including: S110, adopting an input-output two-layer conversion architecture, the input layer identifies the message type through a pre-trained protocol classifier and deconstructs it into a JSON Schema intermediate model, and the output layer implements protocol conversion through a dynamic field mapping table; S120, adopting the PTP time synchronization protocol to achieve microsecond-level data alignment; S130, detecting abnormal sampling values ​​in real time based on the isolation forest algorithm; S200, constructing an evolvable simulation model, and realizing adaptive evolution of the simulation model through a multi-level technical architecture; including: S210, encoding the device geometric features and physical properties into a programmable parameter vector, wherein the geometric feature encoding is based on an MLP network, the geometric constraints adopt the Lagrange multiplier method, and support automatic parameter optimization through error gradient back propagation, the formula is as follows: ; Where y is the physical quantity measurement value of the simulation target, θ is the trainable parameter vector, is a neural network prediction model, g(X) is a physical constraint function, and λ∈(0,1) is a constraint strength coefficient. S220 integrates the outputs of different solvers through a multi-scale modeling framework, uses a dynamic time warping algorithm to achieve cross-solver time step synchronization, and eliminates cross-domain simulation deviations based on a weighted fusion algorithm based on residual analysis. The weights of the weighted fusion algorithm are: ;in, is the fusion weight of the i-th solver, is the residual vector, Represents the Euclidean norm, and the weight range satisfies S230: Global sensitivity analysis of variance decomposition based on sliding windows. The window length is automatically adjusted according to the coefficient of variation of the parameters, dynamically identifying key parameters and optimizing the allocation of computing resources. The formula for automatic adjustment of the window length is as follows: ; Where L(t) is the window length at time t, is the coefficient of variation, which is the ratio of the standard deviation to the mean and is used to quantify the degree of dispersion of the parameter; are the maximum and minimum window lengths, respectively; k is the slope factor of the sigmoid function, which controls the sensitivity of window length changes; s0 is the critical threshold of the coefficient of variation; k and s0 are preset parameters obtained based on cross-validation of historical simulation data sets.

[0017] An artificial intelligence-based simulation data monitoring and analysis method according to an embodiment of the present invention includes at least the following beneficial effects: the method according to an embodiment of the present invention is used to implement any of the above-mentioned systems, so the method according to an embodiment of the present invention has all the beneficial effects of the above-mentioned systems.

[0018] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 A schematic block diagram of modules of a system according to an embodiment of the present invention;

[0021] Figure 2 Schematic diagram of a method according to an embodiment of the present invention.

[0022] Reference numerals:

[0023] Intelligent perception module 100, heterogeneous protocol adaptation unit 110, spatiotemporal alignment unit 120, data quality assessment unit 130, dynamic modeling module 200, parameterized modeling unit 210, physical engine coupling unit 220, model verification unit 230. DETAILED DESCRIPTION

[0024] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0025] In the description of the present invention, "several" means one or more, "more" means two or more, "greater than," "less than," and "exceed" are understood to exclude the number itself, while "above," "below," and "within" are understood to include the number itself. The use of "first" and "second" in the description is solely for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, implicitly specifying the number of the indicated technical features, or implicitly specifying the order of the indicated technical features.

[0026] Glossary:

[0027] Multiscale modeling: Simultaneously integrate physical models with different spatial / temporal resolutions (such as macroscopic continuum mechanics and microscopic molecular dynamics), and align the output time series of different solvers using the Dynamic Time Warping (DTW) algorithm.

[0028] Solver: A numerical computing engine (e.g., a finite element solver or a computational fluid dynamics solver).

[0029] Solver integration: Eliminate the limitations of a single solver in cross-physical domain (such as thermal-mechanical-fluid coupling) simulations and achieve consistency in multi-physics coupling through weighted fusion.

[0030] Global sensitivity analysis: Analyze the impact of the joint changes of all input parameters in the entire parameter space on the output (as opposed to local sensitivity which only examines single parameter perturbations).

[0031] Reference Figure 1 , an embodiment of the present invention proposes an artificial intelligence-based simulation data monitoring and analysis system, comprising:

[0032] The intelligent sensing module 100 is used for multi-source data fusion and collection, and includes:

[0033] The heterogeneous protocol adapter unit 110 is used to support automatic conversion between Modbus / TCP, OPC UA, and MQTT protocols. This embodiment utilizes a two-layer input-output conversion architecture to achieve automatic multi-protocol conversion. The input layer receives raw protocol data packets, and the protocol parsing engine automatically identifies the protocol type based on packet characteristics (Modbus / TCP based on function codes, OPC UA based on NS0 header identification, and MQTT based on topic levels). The data structures of different protocols are then deconstructed into an intermediate common data model. The output layer dynamically binds source protocol tags to target protocol nodes through a field mapping table, completing data type conversion (e.g., Modbus register values ​​to OPC UA Variant types) and semantic adaptation (e.g., mapping MQTT topic trees to OPC UA address spaces). The data packets are then repackaged according to the target protocol specifications, with a quality verification module verifying data integrity and timing consistency. This process utilizes a stateless design, with conversion rules continuously optimized through online learning, ultimately achieving bidirectional transparent transmission between protocols.

[0034] A time-space alignment unit 120 is configured to implement microsecond-level data alignment using the PTP time synchronization protocol;

[0035] The data quality assessment unit 130 is used to detect abnormal sampling values ​​in real time based on the isolation forest algorithm. The data quality assessment unit 130 of this embodiment receives a multi-dimensional time series data stream from a sensor or control system as input through an algorithm, first performs standardization preprocessing on the data to eliminate dimensional differences, and then constructs a set of randomly divided isolation trees. Abnormal values ​​can be quickly identified because they usually have a shorter path length in the tree structure. The detection result is output as an abnormality score with confidence, and the abnormal time point and feature dimension are marked. In industrial scenarios, data anomalies often hide key information, and traditional threshold methods are difficult to deal with complex correlated anomalies. In equipment health management, this embodiment can capture sensor drift or sudden fault signals in real time to avoid misdiagnosis; for process optimization, this embodiment can identify abnormal operating data that deviates from the normal parameter range to ensure the data quality of the training model.

[0036] The dynamic modeling module 200 is used to build an evolvable simulation model and realize the adaptive evolution of the simulation model through a multi-level technical architecture. In this embodiment, the digital twin is used to map the equipment degradation process in real time in equipment health management, and a closed-loop prediction model of parameters and performance is built in process optimization. The dynamic modeling module drives the continuous evolution of the model through real-time data flow. Its input is equipment operating parameters, environmental variables and historical performance data, and its output is a prediction result with confidence assessment. The model evolution trigger mechanism is based on three criteria: 1) triggering parameter recalibration when the online error index exceeds the threshold; 2) starting structural optimization when data distribution drift is detected; 3) performing model attenuation compensation updates regularly. The data processing adopts a sliding window mechanism. After feature extraction, the data in the window is used for both real-time prediction and model iterative training.

[0037] The dynamic modeling module 200 of this embodiment includes:

[0038] The parametric modeling unit 210 is used to encode the geometric features and physical properties of the device into a programmable parameter vector. The geometric feature encoding is based on the MLP network, and the geometric constraints are mathematically expressed using the Lagrange multiplier method. It supports automatic parameter optimization through error gradient back propagation. The formula is as follows: ; Where y is the physical quantity measurement value of the simulation target, θ is the trainable parameter vector, is a neural network prediction model, g(X) is a physical constraint function, and λ∈(0,1) is the constraint strength coefficient. The parametric modeling unit 210 of this embodiment implements a digital representation of device features using a structured coding framework. Its operating principle is as follows: It inputs 3D device geometric topology data (such as a boundary representation or voxel grid) and physical property data (such as material parameters and operating constraints). A feature extraction network first decomposes the geometric structure into hierarchical parameter groups (such as shape parameters and connectivity parameters), while simultaneously quantizing the physical properties into differentiable variables. A dual encoding strategy is employed during the data processing phase: geometric features are embedded using a graph neural network, and physical properties are converted into a standardized parameter space using a normalization layer. The output is a programmable parameter vector of uniform dimension, each dimension of which is differentially mapped to a device performance indicator. Upon receiving real-time sensor data, the system calculates the error gradient between the predicted output and the actual measurement, and then backpropagates adjustments along each dimension of the parameter vector to achieve automatic parameter optimization. This process maintains the mathematical integrity of the geometric constraints and physical laws, ensuring that the optimized parameter combination consistently meets engineering feasibility.

[0039] The physics engine coupling unit 220 is used to integrate the outputs of different solvers through a multi-scale modeling framework, use a dynamic time warping algorithm to synchronize the time steps across solvers, and eliminate cross-domain simulation deviations using a weighted fusion algorithm based on residual analysis. The weights of the weighted fusion algorithm are: ;in, is the fusion weight of the i-th solver, is the residual vector, Represents the Euclidean norm, and the weight range satisfies . The physical engine coupling unit 220 of this embodiment realizes cross-domain simulation collaboration through a multi-scale modeling framework, and its working principle is: input the spatiotemporal discrete data output by each physical field solver (such as structural stress field, fluid velocity field, etc.), first establish a scale correlation matrix, and map the simulation data of different resolutions to a unified reference system. An adaptive weighted fusion algorithm is adopted in the data processing stage to dynamically adjust the fusion weight based on the confidence index of each solver, where the confidence is double-verified by residual analysis and energy conservation rate. The output is composite physical field data that eliminates scale conflicts, and its deviation rate is controlled within the preset tolerance range.

[0040] The model validation unit 230 is used for global sensitivity analysis of variance decomposition based on a sliding window. The window length is automatically adjusted according to the coefficient of variation of the parameters, dynamically identifying key parameters and optimizing the allocation of computing resources. The formula for automatic adjustment of the window length is as follows: ; Where L(t) is the window length at time t, is the coefficient of variation, which is the ratio of the standard deviation to the mean and is used to quantify the degree of dispersion of the parameter; where s0 represents the maximum and minimum window lengths, respectively; k represents the slope factor of the sigmoid function, controlling the sensitivity of window length changes; and s0 represents the critical threshold of the coefficient of variation. k and s0 are preset parameters derived from cross-validation of historical simulation datasets. The model validation unit 230 of this embodiment inputs a parameter space sampling dataset (containing system response values ​​for various parameter combinations). It first constructs a Sobol index calculation framework, decomposing the total variance into individual parameter contributions and interaction terms. During the data processing phase, the Monte Carlo integration method is used to calculate first-order and higher-order sensitivity indices, quantifying parameter importance using variance contribution ratios. A ranked parameter sensitivity list and confidence intervals are output, with parameters whose contribution ratios exceed the threshold being marked as critical parameters. Based on these analysis results, the system dynamically adjusts computing resource allocation: sampling density and iteration counts are increased for critical parameters, while computational accuracy requirements are reduced for non-critical parameters. The entire process employs an adaptive sampling strategy to minimize computational overhead while ensuring analytical accuracy.

[0041] In some embodiments, the system of an embodiment of the present invention also includes: an edge computing module for realizing distributed processing of near data sources, including: a model acceleration unit for deploying lightweight inference models optimized by hardware awareness, and the optimization includes at least one of the following technologies: numerical precision quantization, computational graph operator fusion, and target hardware instruction set adaptation.

[0042] In some embodiments, the edge computing module of an embodiment of the present invention includes: a streaming storage unit for adopting a time series-sensitive cache management strategy, the cache management strategy including: event-triggered data coverage rules, a dynamically adjusted window size mechanism, and multi-priority data shard storage.

[0043] In some embodiments, the edge computing module of an embodiment of the present invention includes: a communication optimization unit for realizing compression and selective uploading of transmitted data, including: adaptive adjustment of encoding parameters based on content characteristics, differential transmission of key frames and non-key frames, and network status-aware bit rate control.

[0044] In some embodiments, the system of the present invention includes an intelligent warning module for dynamically generating graded warning signals based on simulation data, including:

[0045] Composite event detection unit, used to perform multi-dimensional anomaly correlation analysis:

[0046] ;

[0047] in, is the deviation degree of the physical constraint of type i, which is calculated by the Hellinger distance between the real-time simulation value and the theoretical value; is the statistical anomaly score, which is obtained by normalizing the standard deviation of the residual distribution output by the dynamic modeling module; α, β∈[0,1] are weight coefficients, satisfying α+β=1, and τ is the dynamically adjusted expert experience threshold.

[0048] The "abnormalities" involved in the intelligent early warning module in this embodiment include but are not limited to the following two categories: (1) Inconsistency between the simulation model and the physical laws, which is manifested in the simulation data (such as temperature field distribution, stress-strain curve) violating the preset physical constraints (such as energy conservation, material constitutive equation). The root cause of the abnormality is that the actual state of the simulation object (such as equipment, process system) deviates from the normal operating mode, which may cause equipment failure or process out of control. For example: the stress simulation value of the turbine blade exceeds the material yield limit → warning that the blade may undergo plastic deformation. (2) Significant deviation of data-driven statistical characteristics, which is manifested in the time series distribution of the monitoring data (such as residual fluctuation, multi-parameter correlation) significantly deviating from the statistical characteristics of historical normal operating conditions. The root cause of the abnormality may be sensor drift, unmodeled external interference (such as environmental mutation) or hidden failure mode (such as wear accumulation). For example: the variance of the hydraulic system pressure fluctuation suddenly increases by 3 standard deviations → warning of seal leakage risk.

[0049] Reference Figure 2 , an embodiment of the present invention proposes a simulation data monitoring and analysis method based on artificial intelligence, comprising the following steps:

[0050] S100, collects multi-source data fusion, including:

[0051] S110 adopts an input-output two-layer conversion architecture. The input layer uses a pre-trained protocol classifier to identify the message type and deconstruct it into a JSON Schema intermediate model. The output layer implements protocol conversion through a dynamic field mapping table.

[0052] S120, using PTP time synchronization protocol to achieve microsecond-level data alignment;

[0053] S130, detecting abnormal sampling values ​​in real time based on the isolation forest algorithm;

[0054] S200: Build an evolvable simulation model and implement adaptive evolution of the simulation model through a multi-level technical architecture; including:

[0055] S210, encode the geometric features and physical properties of the device into a programmable parameter vector, where the geometric feature encoding is based on the MLP network, and the geometric constraints use the Lagrange multiplier method, supporting automatic parameter optimization through error gradient back propagation. The formula is as follows:

[0056] ;

[0057] Among them, y is the physical quantity measurement value of the simulation target, θ is the trainable parameter vector, is the neural network prediction model, g(X) is the physical constraint function, and λ∈(0,1) is the constraint strength coefficient;

[0058] S220: The outputs of different solvers are integrated through a multi-scale modeling framework. The dynamic time warping algorithm is used to synchronize the time steps across solvers. A weighted fusion algorithm based on residual analysis is used to eliminate cross-domain simulation deviations. The weights of the weighted fusion algorithm are:

[0059] ;

[0060] in, is the fusion weight of the i-th solver, is the residual vector, Represents the Euclidean norm, and the weight range satisfies ;

[0061] S230: Global sensitivity analysis of variance decomposition based on sliding windows. The window length is automatically adjusted according to the coefficient of variation of the parameters, dynamically identifying key parameters and optimizing the allocation of computing resources. The formula for automatic adjustment of the window length is as follows:

[0062] ;

[0063] Among them, L(t) is the window length at time t, is the coefficient of variation, which is the ratio of the standard deviation to the mean and is used to quantify the degree of dispersion of the parameter; are the maximum and minimum window lengths, respectively; k is the slope factor of the sigmoid function, which controls the sensitivity of window length changes; s0 is the critical threshold of the coefficient of variation; k and s0 are preset parameters obtained based on cross-validation of historical simulation data sets.

[0064] Although specific embodiments are described herein, those skilled in the art will recognize that many other modifications or alternative embodiments are also within the scope of the present disclosure. For example, any of the functions and / or processing capabilities described in conjunction with a particular device or component may be performed by any other device or component. In addition, although various exemplary implementations and architectures have been described in accordance with the embodiments of the present disclosure, those skilled in the art will recognize that many other modifications to the exemplary implementations and architectures described herein are also within the scope of the present disclosure.

[0065] Those skilled in the art will appreciate that all or some of the steps and systems disclosed above can be implemented as software, firmware, hardware, or any suitable combination thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is well known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVDs) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0066] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the above implementation mode. Technical personnel familiar with the field can also make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A simulation data monitoring and analysis system based on artificial intelligence, characterized in that: include: Intelligent perception module, used to collect and fuse multi-source data, including: Heterogeneous protocol adapter unit, which uses an input-output two-layer conversion architecture. The input layer uses a pre-trained protocol classifier to identify the message type and deconstruct it into a JSON Schema intermediate model. The output layer implements protocol conversion through a dynamic field mapping table. A time-space alignment unit, used to achieve microsecond-level data alignment using the PTP time synchronization protocol; Data quality assessment unit, used to detect abnormal sampling values ​​in real time based on the isolation forest algorithm; The dynamic modeling module is used to build an evolvable simulation model and realize the adaptive evolution of the simulation model through a multi-level technical architecture; it includes: The parametric modeling unit is used to encode the geometric features and physical properties of the device into a programmable parameter vector. The geometric feature encoding is based on the MLP network, and the geometric constraints are based on the Lagrange multiplier method. It supports automatic parameter optimization through error gradient backpropagation. The formula is as follows: ； Among them, y is the physical quantity measurement value of the simulation target, θ is the trainable parameter vector, is the neural network prediction model, g(X) is the physical constraint function, and λ∈(0,1) is the constraint strength coefficient; The physics engine coupling unit is used to integrate the outputs of different solvers through a multi-scale modeling framework, use a dynamic time warping algorithm to synchronize the time steps across solvers, and eliminate cross-domain simulation deviations using a weighted fusion algorithm based on residual analysis. The weights of the weighted fusion algorithm are: ； in, is the fusion weight of the i-th solver, is the residual vector, Represents the Euclidean norm, and the weight range satisfies ; The model validation unit is used for global sensitivity analysis of variance decomposition based on a sliding window. The window length is automatically adjusted according to the coefficient of variation of the parameters, dynamically identifying key parameters and optimizing the allocation of computing resources. The formula for automatic adjustment of the window length is as follows: ； Among them, L(t) is the window length at time t, is the coefficient of variation, which is the ratio of the standard deviation to the mean and is used to quantify the degree of dispersion of the parameter; are the maximum and minimum window lengths, respectively; k is the slope factor of the sigmoid function, which controls the sensitivity of window length changes; s0 is the critical threshold of the coefficient of variation; k and s0 are preset parameters obtained based on cross-validation of historical simulation data sets.

2. The artificial intelligence-based simulation data monitoring and analysis system according to claim 1, characterized in that: The data quality assessment unit is used to receive a multidimensional time series data stream from a sensor or control system as input, perform standardization preprocessing on the data to eliminate dimensional differences, and construct a set of randomly divided isolation trees; identify outliers through the path length of the tree structure of the isolation tree, output an anomaly score with confidence, and mark the abnormal time point and feature dimension.

3. The artificial intelligence-based simulation data monitoring and analysis system according to claim 1, characterized in that: The parametric modeling unit is used to input the three-dimensional geometric topology data and physical property data of the device, decompose the geometric structure into hierarchical parameter groups through a feature extraction network, and quantify the physical properties into differentiable variables; generate embedding vectors for the geometric features through a graph neural network, convert the physical properties into a standardized parameter space through a normalization layer, and output a programmable parameter vector of unified dimension, each dimension of which establishes a differentiable mapping relationship with the device performance indicator; when receiving real-time sensor data, the error gradient between the predicted output and the actual measurement is calculated, and the adjustment value is backpropagated along each dimension of the parameter vector to achieve automatic parameter optimization.

4. The artificial intelligence-based simulation data monitoring and analysis system according to claim 1, characterized in that: The physics engine coupling unit inputs the spatiotemporal discrete data output by each physics solver, establishes a scale correlation matrix, and maps the simulation data of different resolutions to a unified reference frame. A weighted fusion algorithm dynamically adjusts the fusion weights based on the confidence level of each solver, where the confidence level is verified through residual analysis and energy conservation. The output is composite physics data with scale-free data and a deviation rate within a preset tolerance.

5. The artificial intelligence-based simulation data monitoring and analysis system according to claim 1, characterized in that: The model verification unit is used to input a parameter space sampling data set, construct a Sobol index calculation framework, and decompose the total variance into individual effect terms of each parameter and cross-action terms; The Monte Carlo integration method is used to calculate the first-order and high-order sensitivity indices, and the importance of parameters is quantified by the variance contribution rate; Output parameter sensitivity ranking list and confidence interval, where parameters whose contribution rate exceeds the threshold are marked as key parameters; based on the parameter sensitivity ranking list, confidence interval and key parameters, dynamically adjust computing resource allocation: increase sampling density and number of iterations for key parameters, and reduce calculation accuracy requirements for non-key parameters.

6. The artificial intelligence-based simulation data monitoring and analysis system according to claim 1, characterized in that: The system further comprises: Edge computing module, used to implement distributed processing near data sources, including: A model acceleration unit for deploying lightweight inference models that have been optimized with hardware awareness, including at least one of the following techniques: numerical precision quantization, computational graph operator fusion, and target hardware instruction set adaptation.

7. The artificial intelligence-based simulation data monitoring and analysis system according to claim 6, characterized in that: The edge computing module includes: The streaming storage unit is used to adopt a time-series-sensitive cache management strategy, which includes: event-triggered data overwriting rules, a dynamically adjusted window size mechanism, and multi-priority data sharding storage.

8. The artificial intelligence-based simulation data monitoring and analysis system according to claim 6, characterized in that: The edge computing module includes: The communication optimization unit is used to achieve compression and selective upload of transmitted data, including: adaptive adjustment of encoding parameters based on content characteristics, differential transmission of key frames and non-key frames, and network status-aware bit rate control.

9. The artificial intelligence-based simulation data monitoring and analysis system according to claim 1, characterized in that: The system further comprises: The intelligent early warning module is used to dynamically generate graded early warning signals based on simulation data, including: Composite event detection unit, used to perform multi-dimensional anomaly correlation analysis: ； in, is the deviation degree of the physical constraint of type i, which is calculated by the Hellinger distance between the real-time simulation value and the theoretical value; is the statistical anomaly score, which is obtained by normalizing the standard deviation of the residual distribution output by the dynamic modeling module; α, β∈[0,1] are weight coefficients, satisfying α+β=1, and τ is the dynamically adjusted expert experience threshold.

10. A simulation data monitoring and analysis method based on artificial intelligence, used in the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: S100, collects multi-source data fusion, including: S110 adopts an input-output two-layer conversion architecture. The input layer uses a pre-trained protocol classifier to identify the message type and deconstruct it into a JSON Schema intermediate model. The output layer implements protocol conversion through a dynamic field mapping table. S120, using PTP time synchronization protocol to achieve microsecond-level data alignment; S130, detecting abnormal sampling values ​​in real time based on the isolation forest algorithm; S200: Build an evolvable simulation model and implement adaptive evolution of the simulation model through a multi-level technical architecture; including: S210, encode the geometric features and physical properties of the device into a programmable parameter vector, where the geometric feature encoding is based on the MLP network, and the geometric constraints use the Lagrange multiplier method, supporting automatic parameter optimization through error gradient back propagation. The formula is as follows: ； Among them, y is the physical quantity measurement value of the simulation target, θ is the trainable parameter vector, is the neural network prediction model, g(X) is the physical constraint function, and λ∈(0,1) is the constraint strength coefficient; S220. Integrate the outputs of different solvers through a multi-scale modeling framework, use a dynamic time warping algorithm to synchronize the time steps across solvers, and eliminate cross-domain simulation deviations using a weighted fusion algorithm based on residual analysis; the weights of the weighted fusion algorithm are: ； in, is the fusion weight of the i-th solver, is the residual vector, Represents the Euclidean norm, and the weight range satisfies ; S230: Global sensitivity analysis of variance decomposition based on sliding windows. The window length is automatically adjusted according to the coefficient of variation of the parameters, dynamically identifying key parameters and optimizing the allocation of computing resources. The formula for automatic adjustment of the window length is as follows: ； Among them, L(t) is the window length at time t, is the coefficient of variation, which is the ratio of the standard deviation to the mean and is used to quantify the degree of dispersion of the parameter; are the maximum and minimum window lengths, respectively; k is the slope factor of the sigmoid function, which controls the sensitivity of window length changes; s0 is the critical threshold of the coefficient of variation; k and s0 are preset parameters obtained based on cross-validation of historical simulation data sets.

Images