Phase modifier network-related test simulation platform based on multi-physics field coupling

Abstract

The invention relates to a phase modifier network-related test simulation platform based on multi-physics field coupling. The platform comprises a power grid fault generation device, an electromagnetic field theoretical calculation device, a thermal field theoretical calculation device, a structure field calculation device and a multi-source input coupling device. According to the multi-physical field coupling simulation platform constructed in the scheme, the limitation of a traditional isolated calculation method is thoroughly overcome through a real-time interaction mechanism of dynamically integrating an electromagnetic field, a thermal field and a structural field. Under a power grid fault working condition, the platform accurately simulates a strong coupling effect among electromagnetic loss, temperature distribution and mechanical deformation (such as local overheating caused by eddy-current loss, air gap offset caused by thermal expansion and closed-loop influence of electromagnetic imbalance aggravated by eccentricity); and the output actual parameter set (electromagnetic force density, temperature rise curve, air gap eccentricity and the like) is highly close to the real operation state of the phase modifier set.

Description

Technical Field

[0001] This application relates to the fields of high-voltage power transmission, synchronous condenser simulation, and physical field coupling technology. Specifically, it relates to a synchronous condenser grid connection test simulation platform based on multi-physics field coupling. Background Technology

[0002] The content in this section provides only background information related to this application and may not constitute prior art.

[0003] Synchronous condensers, as critical reactive power compensation and voltage support devices in modern power systems, directly impact the stability, security, and power quality of the power grid. Therefore, accurate simulation testing is essential before putting these devices into actual operation. The core function of simulation testing is to simulate the operating state of the power grid under various conditions (such as load fluctuations, fault disturbances, and system oscillations), and to evaluate the dynamic response characteristics, reactive power regulation capabilities, overload margin, and stress levels of key components of the synchronous condensers. Through simulation, the behavior of the equipment in a real power grid environment can be predicted, the effectiveness of its design parameters and control strategies can be verified, and potential risks can be identified.

[0004] Currently, the common method for simulation testing of synchronous condenser units is to calculate the adjustment limits of their various physical properties in isolation. This mainly includes: Temperature Limits: Based on a thermodynamic model, calculate the maximum allowable temperature rise of key components such as stator windings, rotor windings, and core under specific cooling conditions.

[0005] Electromagnetic limit: Based on electromagnetic field theory, calculate the maximum output capacity (such as maximum capacitive / inductive reactive power) under constraints such as stator and rotor magnetic circuit saturation, upper limit of excitation current, and end leakage flux.

[0006] Deformation / Mechanical Limits: Considering the mechanical properties of materials, calculate the maximum allowable deformation or stress level of rotor deflection, bearing vibration, and critical structural components (such as end caps and bases) under the combined action of electromagnetic force, thermal stress, and centrifugal force. These limit values ​​are usually calculated individually under idealized or simplified boundary conditions.

[0007] When synchronous condensers operate in actual power grids, their physical fields, such as temperature, electromagnetic, and stress (deformation / vibration), are highly coupled and mutually influential. For example, excessive excitation current (electromagnetic limit) can cause a sharp increase in winding temperature (temperature limit), while high temperature can exacerbate material expansion and mechanical deformation (deformation limit), thus affecting air gap uniformity and, in turn, worsening electromagnetic performance and vibration levels, forming a complex dynamic feedback process. Existing methods that calculate individual limits in isolation fail to fully consider this strong coupling effect of multiple physical fields, leading to overly idealized calculation results. Therefore, the "operating limits" determined based on these individual limits cannot truly reflect the comprehensive performance boundaries of synchronous condensers under complex, dynamic, and multi-constraint coupled actual power grid conditions. This limitation makes it difficult for simulation results to accurately predict the actual withstand capacity and safety margin of the equipment during extreme or transient processes in the power grid, potentially leading to misjudgments of equipment operating risks or underestimations of equipment potential, and failing to provide accurate and reliable simulation basis for the safe and stable operation of the power grid. Summary of the Invention

[0008] The summary section of this application is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0009] Some embodiments of this application propose a simulation platform for network-connected camera switching based on multi-physics coupling to solve the technical problems mentioned in the background section above.

[0010] As a first aspect of this application, some embodiments of this application provide a simulation platform for network-connected testing based on multi-physics coupling, including: A power grid fault generation device is used to provide the power grid supply environment for synchronous condenser groups and simulate fault waveforms; An electromagnetic field theory calculation device, based on the transient finite element method, calculates the theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation of a synchronous condenser group when the input current of a fault waveform is applied. The thermal field theoretical calculation device, based on the transient heat conduction-convection equation, generates the theoretical temperature field and theoretical temperature rise curve of the condenser group according to the theoretical electromagnetic loss and fluid heat transfer coefficient. The structural field calculation device generates theoretical deformation characteristic data of the phase-shifting group and theoretical air gap eccentricity based on theoretical electromagnetic force density and theoretical temperature field. A multi-source input coupling device generates an actual parameter set based on a theoretical parameter set. The theoretical parameter set includes theoretical electromagnetic force density, theoretical electromagnetic loss, theoretical torque pulsation, theoretical temperature field distribution, theoretical temperature rise curve, deformation characteristic data, and air gap eccentricity coupling. The actual parameter set includes actual electromagnetic force density, actual electromagnetic loss, actual torque pulsation, actual temperature field distribution, actual temperature rise curve data, actual deformation characteristic data, and actual air gap eccentricity coupling. The safety boundary calculation device generates the corresponding fault waveform of the camera group based on the actual parameter set, adjusts the fault probability of the camera group, calculates the fault probability of the camera group under different fault waveforms, and generates a simulation experiment report.

[0011] This multiphysics coupled simulation platform overcomes the limitations of traditional isolated calculation methods by dynamically integrating a real-time interaction mechanism between electromagnetic, thermal, and structural fields. Under power grid fault conditions, the platform accurately simulates the strong coupling effects between electromagnetic losses, temperature distribution, and mechanical deformation (e.g., eddy current losses causing local overheating → thermal expansion leading to air gap displacement → eccentricity exacerbating the closed-loop effect of electromagnetic imbalance). This ensures that the output parameter set (electromagnetic force density, temperature rise curve, air gap eccentricity, etc.) closely approximates the actual operating state of the synchronous condenser. Based on the generated fault probability model and safety boundary assessment results, it achieves, for the first time, a quantitative characterization of the multi-constraint coupled limit conditions of synchronous condensers under complex power grid disturbances. The simulation error is reduced by more than 40%, providing a high-confidence prediction of equipment carrying capacity for the safe and stable operation of the power grid, effectively avoiding the risk of equipment damage or system instability due to misjudgments of single limits.

[0012] The electromagnetic field theory calculation device includes: The coordinate generation module is used to determine the origin on the motor shaft of the synchronous condenser group, take the z-axis as the axial direction of the rotation shaft, and make the plane of the x-axis and y-axis perpendicular to the z-axis to establish a three-dimensional rectangular coordinate system. The synchronous condenser response calculation module calculates the stator terminal current, rotor mechanical angle, and excitation current of the synchronous condenser based on the fault waveform. The electromagnetic field theory calculation module calculates the theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation based on the stator end current, rotor mechanical angle, and excitation current.

[0013] The thermal field theory calculation device includes: Temperature information loading module, used to load the heat exchange efficiency at various locations in the heat exchange system; The temperature calculation module generates a theoretical temperature field and a theoretical temperature rise curve based on electromagnetic loss, initial temperature distribution, and heat transfer efficiency at each location. The theoretical temperature field is the temperature distribution of each region in a three-dimensional rectangular coordinate system over time, and the theoretical temperature rise curve is the temperature rise rate of key points in the three-dimensional rectangular coordinate system over time.

[0014] The structural field calculation device includes: The structural information loading module is used to load the initial geometric state of the camera module. The structural information calculation module generates theoretical deformation characteristic data and theoretical air gap eccentricity based on the initial geometric state, theoretical temperature field, and electromagnetic force density.

[0015] In this scheme, the electromagnetic field theoretical calculation device generates stator and rotor current excitation based on the fault waveform, and outputs the spatially distributed theoretical electromagnetic force density / loss / torque pulsation; the thermal field theoretical calculation device uses electromagnetic loss as a heat source and, combined with heat transfer boundary conditions, dynamically derives the theoretical temperature field and temperature rise curve in a three-dimensional coordinate system; the structural field calculation device further uses the electromagnetic force density and temperature field as loads, superimposed with the initial geometric state, to calculate the key theoretical deformation characteristics and air gap eccentricity. This chain-like architecture is the first to completely construct the causal transmission path of "electromagnetic excitation → thermal deformation → mechanical response," reducing the error in theoretical value calculations.

[0016] As a second aspect of this application, when calculating the theoretical parameter sets (electromagnetic force density, temperature field, air gap eccentricity, etc.) of electromagnetic, thermal, and structural fields, inherent calculation biases that are difficult to quantify exist within a single physics field due to factors such as idealized boundary conditions and uncertainties in material parameters. These biases are unrelated to multiphysics coupling effects but are amplified progressively with chained calculations (e.g., electromagnetic loss error → temperature field distortion → deformation result failure), ultimately causing simulation results to deviate from actual operating conditions.

[0017] The simulation platform for network-connected testing of a camera based on multi-physics coupling also includes a correction device; The correction device includes: The electromagnetic correction module has a built-in first correction neural network model, which is used to correct the calculated theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque ripple. The thermal field correction module has a built-in second correction neural network model, which is used to correct the calculated theoretical temperature field and theoretical temperature rise curve. The structural correction module has a built-in third correction neural network model, which is used to correct the calculated theoretical deformation characteristic data and theoretical air gap eccentricity. The first, second, and third corrected neural network models are used to correct the error between the theoretical calculation value and the actual value to obtain the correction parameters.

[0018] This scheme, by adding an independently operating correction device, addresses the inherent biases in theoretical calculations of electromagnetic, thermal, and structural fields caused by model simplification and idealized boundary conditions, while maintaining the integrity of the multiphysics coupling calculation framework. It employs dedicated neural network models (first, second, and third correction neural network models) to perform specific error correction at the single-physics level. This design strictly distinguishes between the dual needs of theoretical calculation bias correction and physical field coupling analysis: the correction device only learns the mapping errors between theoretical and actual values ​​of each single field (such as finite element discretization errors, material nonlinear responses, etc.), without intervening in the interaction mechanisms between physical fields; the corrected parameters serve as input to the multi-source coupling device, significantly improving the physical accuracy of key parameters such as electromagnetic force density, temperature field distribution, and air gap eccentricity.

[0019] Furthermore, the first modified neural network model is a temporal feature extraction network based on bidirectional LSTM; The second modified neural network model is a multi-scale feature fusion network based on 3D convolutional coding; The third modified neural network model is a robust shallow network based on feature expansion.

[0020] In the technical solution provided in this application, the three modified neural network models have different structures, mainly to adapt to different data requirements. Relatively speaking, the changes in electromagnetic force density and time-series information are more significant, and errors are more likely to exist between consecutive time periods. Therefore, an LSTM network is used to capture the signal change relationship between one-dimensional data. For temperature layer data, more attention is paid to the error changes in temperature at three-dimensional locations. Therefore, convolutional and encoding networks are needed to reduce and compress high-dimensional features to find the dependencies between features. For torque information, which is relatively an error information that spreads randomly to both sides from the theoretical value, a shallow neural network is used to find the dependencies between features and avoid model fitting.

[0021] Furthermore, the training data for the first, second, and third modified neural network models are theoretical and actual parameters obtained when the camera group is in a steady state. The theoretical parameters are used as feature data, and the actual parameters are used as labeled data. Among them, the variation ranges of theoretical temperature field, theoretical temperature rise curve, theoretical deformation characteristic data, and theoretical air gap eccentricity in the data used to train the first modified neural network model are maintained within a preset range. In the data used to train the second modified neural network model, the variation ranges of electromagnetic force density, theoretical electromagnetic loss, theoretical torque pulsation, theoretical deformation characteristic data, and theoretical air gap eccentricity are maintained within a preset range. In the data used to train the third modified neural network model, the variation amplitudes of electromagnetic force density, theoretical electromagnetic loss, theoretical torque ripple, electromagnetic force density, theoretical electromagnetic loss, and theoretical torque ripple are maintained within a preset range.

[0022] As a third aspect of this application, when generating actual parameter sets (electromagnetic force density, temperature rise curve, air gap eccentricity, etc.), it is impossible to explicitly quantify the cross-field coupling contribution weights between electromagnetic field, thermal field, and structural field parameters (such as the driving strength of electromagnetic loss on local temperature rise and the influence ratio of thermal expansion on air gap eccentricity). This forces the coupling calculation to rely on linear approximation of empirical formulas or full-end learning of black-box neural networks, resulting in the neglect of key weak coupling effects (such as the hysteresis interaction between transient torque pulsation and temperature field) or the overfitting of strong noise coupling, which seriously restricts the simulation accuracy and physical interpretability.

[0023] The multi-source input coupling device includes: The information input module is used to obtain the theoretical parameter set X; ;in, Represents the empty set. , , These represent the theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation, respectively. , These represent the theoretical temperature field and the theoretical temperature rise curve, respectively. , These represent the corrected theoretical deformation characteristic data and the theoretical air gap eccentricity, respectively. The data augmentation module, based on the Group Lasso model, labels the true contribution of each parameter in the theoretical parameter group X to each parameter in the actual parameter group, and obtains contribution information. The large model coupling calculation module is used to input the theoretical parameter set X and contribution information, and to find the coupling relationship between feature parameters based on the multi-channel cross network to generate the actual parameter set.

[0024] The technical solution provided in this application utilizes a data augmentation module to annotate the true contribution of each parameter in the theoretical parameter set X to each parameter in the actual parameter set. This true contribution information effectively encompasses the coupling relationships between multiple physics fields. The corrected parameter set and the contribution information are input together into a large model for training. The large model can use the contribution information to find the implicit coupling relationships between the corrected parameters, thereby accurately generating the actual parameter set. Because the contribution information is essentially a form of data augmentation, this ensures the prediction accuracy of the large model while avoiding overfitting during intensive training.

[0025] Furthermore, the methods for obtaining contribution information include the following steps: S1: Obtain the theoretical parameter set of the synchronous condenser group under fault conditions, input each parameter in the theoretical parameter set into the trained first correction neural network model, second correction neural network model and third correction neural network model to obtain the theoretical parameter set X; measure the current actual parameter set Y of the synchronous condenser group; All theoretical parameter sets X collected are used as the observation variable matrix XT, and all actual parameter sets collected are used as the response variable matrix YT. Wherein, the dimension of the observation variable matrix XT is n*p, and the dimension of the response variable moment YT is n*p, where n represents the number of samples collected, and p=7; S2: Establish an influence group for each element in the actual parameter set Y. Where b represents the index of an element in the actual parameter group Y, b∈[1,7], and l is the index that affects the grouping. b and l are positive integers. This represents the total number of influence groups for the b-th parameter in the actual parameter group; S3: Get all impact groups Set each affected group Initialization coefficient vector g represents the index that affects the grouping, and k represents the number of iterations; set the initial step size. Step size reduction factor Decrease coefficient and loss function ; S4: Perform the following iterative operation for each influence group until convergence. While performing the iterative operation for each influence group, the coefficient vectors of the other influence groups remain fixed. Based on the current coefficient vector Group design matrix X g Calculate gradient and Heisenberg scalar h g ; ; ; This represents the pre-defined loss function, g represents the index that affects the grouping, and X represents the index that affects the grouping. g =Column indices in the observed variable matrix XT belong to the influence grouping submatrix, Represents the gradient of the loss function. This represents the sub-vector corresponding to the l-th group in the complete coefficient vector of the k-th iteration; ; This represents the loss function with respect to the coefficient vector of the g-th group. Denotes the minimum curvature constant. represents an approximate scalar of a Hessian matrix block, max represents the vector maximum value operation, and diag represents the matrix diagonal extraction operation; Calculate the auxiliary vector u; ; S5: Determine the update direction of the coefficient vector based on the auxiliary vector; ; Indicates the direction of the update. λ represents the penalty threshold, and λ represents the weighting coefficient. S5: Calculate the reference value based on the update direction. Calculate the optimal step size based on the reference value. : ; Indicates the expected decrease. Represents the matrix transpose symbol; S6: Based on the optimal step size Update the coefficient vector; ; S6: Determine whether the iteration should stop based on the convergence condition. The convergence condition is: If the iteration condition is met, the iteration stops; otherwise, the iteration continues. S7: Obtain the coefficient vector of all factors affecting the grouping. The coefficient vector is used as contribution information.

[0026] Based on the Group Lasso design, the influence of the correction parameters on the actual parameters is decomposed into contribution groups with clear physical meanings (such as the combined effect of electromagnetic parameter groups on air gap eccentricity). The sparse coefficient vector of each group is solved through iterative optimization (steps S4-S6) (step S7), which explicitly reveals the quantification weights of causal chains such as "electromagnetic loss → end temperature rise → rotor expansion". This guides the neural network model to find the intrinsic relationship between data during subsequent neural network model training.

[0027] Furthermore, multi-channel cross-connect networks include: The input layer is used to input the theoretical parameter set X and contribution information W; The feature embedding layer projects the input theoretical parameter set X into a high-dimensional space through a fully connected layer to generate a high-dimensional vector: The CrossNet network initializes its weight matrix with contribution information W and sets fine-tuning boundaries. High-dimensional vectors are input into the CrossNet network to generate and display high-order features. The fully connected network is a three-layer fully connected network that extracts the hidden features between the input theoretical parameter set X. The feature fusion layer, based on a gated fusion mechanism, fuses and displays high-order features and latent features, and outputs fused features. The multi-task output layer generates the actual parameter set based on the output fusion features.

[0028] In this scheme, the weight matrix of the cross network is initialized by the contribution information W, that is, the contribution information is transformed into the weight information in the network structure, so that the multi-channel cross network understands and learns some of the implicit relationships between the data in the initial state. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the structure of a simulation platform for network-connected camera switching based on multi-physics coupling.

[0030] Figure 2 This is a schematic diagram of a multi-channel crossover network. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments. The same reference numerals in the accompanying drawings represent the same components. It should be noted that the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the described embodiments of this application without creative effort are within the scope of protection of this application.

[0032] Compared to the embodiments shown in the accompanying drawings, feasible embodiments within the scope of this application may have fewer components, other components not shown in the drawings, different components, differently arranged components, or components with different connections, etc. Furthermore, two or more components in the drawings may be implemented in a single component, or a single component shown in the drawings may be implemented as multiple separate components.

[0033] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” and similar terms used in this specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not necessarily indicate a quantity limitation. Terms such as “upper” and “lower” are used only to indicate relative positional relationships, and these relative positional relationships may change accordingly when the absolute position of the described object changes.

[0034] refer to Figure 1Example 1: A multi-physics coupling-based synchronous condenser grid connection test simulation platform includes a grid fault generation device, an electromagnetic field theory calculation device, a thermal field theory calculation device, a structural field calculation device, a multi-source input coupling device, and a safety boundary calculation device. The grid fault generation device provides the grid power supply environment for the synchronous condenser and simulates the fault waveform. The electromagnetic field theory calculation device calculates the theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation of the synchronous condenser when the input fault waveform current is obtained using the transient finite element method. The thermal field theory calculation device generates the theoretical temperature field and theoretical temperature rise curve of the synchronous condenser based on the transient heat conduction-convection equations, according to the theoretical electromagnetic loss and fluid heat transfer coefficient. The structural field calculation device generates the theoretical deformation characteristic data and theoretical air gap eccentricity of the synchronous condenser based on the theoretical electromagnetic force density and theoretical temperature field. A multi-source input coupling device generates an actual parameter set based on a theoretical parameter set. The theoretical parameter set includes theoretical electromagnetic force density, theoretical electromagnetic loss, theoretical torque pulsation, theoretical temperature field distribution, theoretical temperature rise curve, deformation characteristic data, and air gap eccentricity coupling. The actual parameter set includes actual electromagnetic force density, actual electromagnetic loss, actual torque pulsation, actual temperature field distribution, actual temperature rise curve data, actual deformation characteristic data, and actual air gap eccentricity coupling. A safety boundary calculation device generates the corresponding fault waveform down-adjustment camera group based on the actual parameter set, calculates the fault probability of the down-adjustment camera group under different fault waveforms, and generates a simulation experiment report.

[0035] The theoretical logic of this scheme is as follows: The power grid fault generation device randomly provides synchronous condenser units with fault waveforms that may be encountered during power grid operation. An electromagnetic theory calculation device calculates the electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation obtained from a single theory when the synchronous condenser unit responds to the fault waveform. A thermal field theory calculation device calculates the theoretical temperature field and theoretical temperature rise curve obtained from a single theory when the synchronous condenser unit responds to the fault waveform. A structural field calculation device calculates the theoretical deformation characteristic data and theoretical air gap eccentricity obtained from a single theory when the synchronous condenser unit responds to the fault waveform. The above theoretical parameter sets are all information calculated under a single environment. In complex fault handling operations, there is a significant deviation between theoretical and actual values. This deviation is caused by the interaction between parameters. Therefore, a multi-source input coupling device is used to couple the correspondence between theoretical data to generate an actual parameter set that is as close as possible to the actual values.

[0036] In generating parameter sets, the primary goal is to accurately calculate the theoretical values. After calculating the theoretical values, the actual values ​​can be obtained using sensors. These theoretical and actual values ​​can then be used as training data to train the model. It is foreseeable that the more accurate the theoretical values ​​are, the closer they are to the actual performance of the camera assembly, and the clearer the relationship between the theoretical and actual values ​​during training, resulting in higher model accuracy after training. Based on this, the specific calculation method for the theoretical values ​​is as follows: The electromagnetic field theory calculation device includes: a coordinate generation module, a synchronous condenser group response calculation module, and an electromagnetic field theory calculation module. The coordinate generation module and the synchronous condenser group response calculation module are respectively connected to the electromagnetic field theory calculation module. The coordinate generation module also needs to be connected to the thermal field theory calculation device and the structural field calculation device.

[0037] The coordinate generation module is used to determine the origin on the motor shaft of the synchronous condenser, using the z-axis as the axial direction of the rotation shaft, and placing the x-axis and y-axis planes perpendicular to the z-axis to establish a three-dimensional Cartesian coordinate system. Most parameters in the theoretical parameter set describe various positions within the synchronous condenser's motor, thus requiring an accurate coordinate system. This coordinate system is primarily used to determine these positions.

[0038] The synchronous condenser response calculation module calculates the stator terminal current, rotor mechanical angle, and excitation current of the synchronous condenser based on the fault waveform. When responding to a fault waveform, the built-in control logic adjusts the synchronous condenser's operation. This module simulates how the synchronous condenser will generate stator terminal current, rotor mechanical angle, and excitation current when dealing with fault waveforms or the power supply environment generated by the power grid fault generation device. The stator terminal current, rotor mechanical angle, and excitation current are only related to the power grid's supply conditions and do not participate in multi-physics coupling.

[0039] The electromagnetic field theory calculation module calculates the theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation based on the stator end current, rotor mechanical angle, and excitation current.

[0040] The electromagnetic field theory calculation module is the main calculation module. Relatively speaking, the calculations of theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation are complex. This application uses the following method for calculation: The theoretical electromagnetic force density is f em (x, y, z, t); In the conductive region, f em The formula for calculating (x, y, z, t) is: ; f x f represents the electromagnetic force density component in the x-direction. y f represents the electromagnetic force density component in the y-direction.z J represents the electromagnetic force density component in the z-direction. x J represents the current density component in the x-direction. y J represents the current density component in the y-direction. z B represents the current density component in the z-direction. x B represents the magnetic flux density component in the x-direction. y B represents the magnetic flux density component in the y-direction. z Represents the magnetic flux density component in the z-direction; In the magnetically conductive region, f em The formula for calculating (x, y, z, t) is: ; ; ; H x H y H z These represent the components of the magnetic field strength in the x, y, and z directions, respectively. Indicates the partial derivative sign; ; ; ; ; ; ; ; ; ; Where u represents the magnetic permeability. , , These represent the stator current density components in the x, y, and z directions, respectively. , , These represent the excitation current density components in the x, y, and z directions, respectively. , , These represent the magnetic vector potential components in the x, y, and z directions, respectively, obtained by solving the electromagnetic field control equations using the rotor mechanical angles. Indicates conductivity; The theoretical electromagnetic loss is p em (x, y, z, t); In the conductive region, p em The formula for calculating (x, y, z, t) is: J represents the current density vector; In the magnetically conductive region, p em The formula for calculating (x, y, z, t) is: ; Indicates the hysteresis loss coefficient. This represents the eddy current loss coefficient. represents the additional loss coefficient, and B represents the magnetic flux density; ; Theoretical torque ripple is ; B r B represents the radial magnetic flux density component. t Indicates the tangential magnetic flux density component. Represents the circumferential angle. denoted by , L represents the axial length of the core, and r represents the center radius of the air gap.

[0041] The thermal field theory calculation device includes: The temperature information loading module is used to load the heat exchange efficiency at various locations in the heat exchange system.

[0042] The heat exchange system is the cooling module in the synchronous condenser unit, and its heat exchange performance is set to a constant value in this application. Specifically, liquid cooling results in high heat exchange efficiency, while air cooling results in low efficiency. The heat exchange efficiency is also related to the control logic of the synchronous condenser unit; a higher flow rate of the heat exchange medium leads to higher efficiency, and vice versa. Although the heat exchange efficiency of the synchronous condenser unit affects the temperature field calculation, it is relatively independent of other theoretical parameters. Therefore, the heat exchange efficiency in this scheme is calculated based on the highest efficiency of the synchronous condenser unit.

[0043] The temperature calculation module generates a theoretical temperature field and a theoretical temperature rise curve based on electromagnetic loss, initial temperature distribution, and heat transfer efficiency at each location. The theoretical temperature field is the temperature distribution of each region over time in a three-dimensional Cartesian coordinate system, and the theoretical temperature rise curve is the temperature rise rate of key points in the three-dimensional Cartesian coordinate system over time.

[0044] ; ; Where V represents volume, P em Indicates electromagnetic loss. This represents the heat dissipation coefficient at positions x, y, and z. The time constant is represented by e, and the natural constant is represented by e. The temperature calculation module calculates the temperature field for each location. Including the initial temperature, the key points are several pre-set representative points for the temperature of the interchanged camera groups. The temperature rise rate is actually... . This represents the heat dissipation coefficient at positions x, y, and z. This heat dissipation coefficient is related to the heat dissipation system of the camera condenser. Each position is set to a fixed value based on the work capacity of the heat dissipation system, and the camera condenser needs to conduct experiments to obtain the value.

[0045] The structural field calculation device includes: a structural information loading module, used to load the initial geometric state of the phase-shifting group; and a structural information calculation module, which generates theoretical deformation characteristic data and theoretical air gap eccentricity based on the initial geometric state, theoretical temperature field, and electromagnetic force density.

[0046] The theoretical deformation characteristic data consists of the deformation at key point locations, which are near the camera module connection structure and are prone to deformation. This theoretical deformation characteristic data is obtained by solving the linear elastic-static equations. ; This indicates that the stress tensor is related to the material and temperature. The theoretical electromagnetic force density; ; Indicates the displacement of the inner surface of the stator. This indicates the displacement of the rotor's outer surface.

[0047] The above outlines the modules related to theoretical parameter calculation. By providing the specific methods for calculating theoretical parameters, sufficient data samples can be obtained. These data samples are then used to train the multi-source input coupling device, allowing for the acquisition of the actual parameter set of the camera module under extreme conditions during subsequent practical simulations. Once the actual parameter set is obtained, the safety boundary can be easily determined.

[0048] Example 2: Example 2 provides a correction device based on Example 1.

[0049] Example 1 provides a method for generating a simulation test report on the failure probability of a synchronous condenser. The key is obtaining the actual parameter set of the synchronous condenser. Calculating the actual parameter set requires obtaining theoretical parameters. However, in actual calculations, there are numerous factors related to the theoretical parameter set, and the method for calculating the theoretical parameters provided in Example 1 is actually an assumption with unpredictable errors. During actual operation, the internal information connections of a synchronous condenser are complex, and without multi-physics coupling, it is difficult to accurately calculate the theoretical values ​​using the above-mentioned method. Therefore, this application provides the following correction device, which is mainly used to correct deviations in theoretical value calculations. The correction device differs from the multi-source input coupling device in that it only considers the deviation between the calculated theoretical value and the actual theoretical value, without introducing or learning the coupling relationships between physical fields.

[0050] The correction device includes an electromagnetic correction module, a thermal correction module, and a structural correction module. The electromagnetic correction module incorporates a first correction neural network model to correct the calculated theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation. The thermal correction module incorporates a second correction neural network model to correct the calculated theoretical temperature field and theoretical temperature rise curve. The structural correction module incorporates a third correction neural network model to correct the calculated theoretical deformation characteristic data and theoretical air gap eccentricity. The first, second, and third correction neural network models are used to correct the error between the theoretical calculated values ​​and the actual values ​​to obtain correction parameters.

[0051] The first, second, and third corrected neural network models are three independent neural network models. That is, the three correction modules in the corrected transpose are not related to each other, and each correction module processes the theoretical values ​​related to itself.

[0052] The first modified neural network model is a temporal feature extraction network based on bidirectional LSTM. The second modified neural network model is a multi-scale feature fusion network based on 3D convolutional coding. The third modified neural network model is a robust shallow network based on feature expansion.

[0053] The core functions and specific design ideas of the first, second, and third correction neural networks have already been provided. Their main difference lies in handling different data structures. Relatively speaking, the bidirectional LSTM temporal feature extraction network easily extracts temporal features that are dependent on each other, and it is more accurate and efficient in processing theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation, which have significant temporal variations. Correspondingly, the 3D convolutional coding multi-scale feature fusion network easily handles multi-channel temperature information, can capture the numerical variation relationship (temperature field) between adjacent regions, and is more accurate in processing temperature field information. The robust shallow network based on feature expansion has a relatively simple network structure, can correct linear values, and is more suitable for theoretical deformation feature data and theoretical air gap eccentricity.

[0054] The model structures chosen for the first, second, and third modified neural networks have shown good performance in practice. The key lies in the selection of training data. Based on this, this application provides the following technical solution: The training data for the first, second, and third corrected neural network models consist of theoretical and actual parameters acquired when the camera condenser is in a steady state. The theoretical parameters are used as feature data, and the actual parameters are used as annotation data. Specifically, in the data used to train the first corrected neural network model, the variations in the theoretical temperature field, theoretical temperature rise curve, theoretical deformation feature data, and theoretical air gap eccentricity are maintained within a preset range. In the data used to train the second corrected neural network model, the variations in the electromagnetic force density, theoretical electromagnetic loss, theoretical torque pulsation, theoretical deformation feature data, and theoretical air gap eccentricity are maintained within a preset range. In the data used to train the third corrected neural network model, the variations in the electromagnetic force density, theoretical electromagnetic loss, theoretical torque pulsation, and theoretical torque pulsation are maintained within a preset range.

[0055] The core of this scheme lies in using the relevant parameters under steady-state conditions to train the three neural network models mentioned above. When the synchronous condenser is in a steady state, its performance in various aspects has a certain margin. For example, temperature control can be achieved by adjusting the heat dissipation efficiency of the synchronous condenser, keeping the temperature within a variable range. Correspondingly, under steady-state conditions, the deformation characteristics and air gap eccentricity of the synchronous condenser tend to stabilize and do not fluctuate drastically. Therefore, by collecting the parameters at this time and training the three models, the system can find the dependencies between the theoretical data within the actual synchronous condenser when they are not coupled.

[0056] For example, for the training data of the first neural network model, the heat transfer coefficient of the synchronous condenser is adjusted to ensure that the temperature field does not change (with minimal change). The electromagnetic force density, electromagnetic loss, and torque ripple of the synchronous condenser under different power grid supply environments (non-fault conditions) are detected and used as tag data. Then, the theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque ripple are calculated simultaneously using the scheme of Example 1 and used as input data.

[0057] For the training data of the second neural network model, the heat exchange efficiency of the fixed synchronous condenser group and the power grid supply environment (non-fault state) are used to detect the temperature field and theoretical temperature rise curve at this time, which are then used as label data. Then, the theoretical temperature field and theoretical temperature rise curve are simultaneously calculated using the scheme of Example 1, and these are used as input data.

[0058] For the training data of the second neural network model, the heat transfer coefficient of the synchronous condenser is adjusted to ensure that the temperature field remains unchanged (with minimal change). The power grid supply environment is fixed (non-fault condition), and the deformation characteristic data and air gap eccentricity are detected and used as label data. Then, the theoretical deformation characteristic data and theoretical air gap eccentricity are calculated simultaneously and used as input data.

[0059] Example 3: Example 3 provides a further multi-source input coupling device based on Example 1.

[0060] Multi-source input coupling devices employ big data processing to find the dependencies between theoretical and feature data. However, in the absence of data annotation, the coupling relationship between theoretical and actual parameter sets is complex, especially under fault conditions, where the error between the original theoretical and actual calculated values ​​is amplified. Furthermore, the coupling relationships of multiphysics fields make the connections between data unclear, leading to extreme difficulty in training large models. Based on this, this application provides the following technical solution: The multi-source input coupling device includes an information input module, a data augmentation module, and a large-model coupling calculation module. These modules are connected sequentially.

[0061] The information input module is used to obtain the theoretical parameter set X; ;in, Represents the empty set. , , These represent the theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation, respectively. , These represent the theoretical temperature field and the theoretical temperature rise curve, respectively. , These represent the corrected theoretical deformation characteristic data and the theoretical air gap eccentricity, respectively. In this scheme, if the scheme of embodiment 3 combined with embodiment 2 is applied to embodiment 1, the information input module obtains the set of correction parameters. If embodiment 3 and embodiment 1 are directly combined, the scheme of embodiment 1 can be directly used to obtain the theoretical parameter set.

[0062] The data augmentation module, based on the Group Lasso model, labels the actual contribution of each parameter in the theoretical parameter group X to each parameter in the actual parameter group, and obtains contribution information.

[0063] Contribution information is actually the relationship between data. Before obtaining the corrected relationship, it is necessary to obtain enough training samples X and Y, where X is the labeled dataset and Y is the actual parameter set obtained by synchronous measurement when obtaining the labeled dataset.

[0064] The methods for obtaining contribution information include the following steps: S1: Take all the collected modified parameter sets (theoretical parameter sets) X as the observation variable matrix XT, and take all the collected actual parameter sets as the response variable moments YT; where the dimension of the observation variable matrix XT is n*p, the dimension of the response variable moments YT is n*p, n represents the number of samples collected, and p=7; S2: Establish an influence group for each element in the actual parameter set Y. Where b represents the index of an element in the actual parameter group Y, b∈[1,7], and l is the index that affects the grouping. b and l are positive integers. This represents the total number of influence groups for the b-th parameter in the actual parameter group; S3: Get all impact groups Set each affected group Initialization coefficient vector g represents the index that affects the grouping, and k represents the number of iterations; set the initial step size. Step size reduction factor Decrease coefficient and loss function ; S4: For each impact group Perform the following iterative operation until convergence, affecting each group. During iterative operations, the remaining coefficient vectors that affect grouping fixed; Based on the current coefficient vector Group design matrix X g Calculate gradient and Heisenberg scalar h g ; ; ; This represents the pre-defined loss function, g represents the index that affects the grouping, and X represents the index that affects the grouping. g =Column indices in the observed variable matrix XT belong to the influence grouping submatrix, Represents the gradient of the loss function. This represents the sub-vector corresponding to the l-th group in the complete coefficient vector of the k-th iteration; ; This represents the loss function with respect to the coefficient vector of the g-th group. Denotes the minimum curvature constant. represents an approximate scalar of a Hessian matrix block, max represents the vector maximum value operation, and diag represents the matrix diagonal extraction operation; Calculate the auxiliary vector u; ; S5: Determine the update direction of the coefficient vector based on the auxiliary vector; ; Indicates the direction of the update. λ represents the penalty threshold, and λ represents the weighting coefficient. S5: Calculate the reference value based on the update direction. Calculate the optimal step size based on the reference value. : ; Indicates the expected decrease. Represents the matrix transpose symbol; Specifically, the method involves testing candidate step sizes from largest to smallest, and selecting the first step size that satisfies the condition that "the actual function decreases by the expected threshold" as the optimal step size. ; The conditions for optimization are: ; Indicates step size, Indicates the decrease coefficient. This represents the vector of all coefficients at the k-th iteration. Represents the objective function for optimization; S6: Based on the optimal step size Update the coefficient vector; ; S6: Determine whether the iteration should stop based on the convergence condition. The convergence condition is: If the iteration condition is met, the iteration stops; otherwise, the iteration continues. S7: Obtain the coefficient vector of all factors affecting the grouping. The coefficient vector is used as contribution information.

[0065] The loss function of this scheme is a linear regression function.

[0066] The coefficient vector serves as contribution information, and the coefficient vector corresponds to X. g X g =Column indices in the observed variable matrix XT belong to the influence grouping The submatrix. Therefore, the coefficient vector provided by this scheme is actually the inherent cross-correlation information between the data.

[0067] The large model coupling calculation module is used to input the theoretical parameter set X and contribution information, and to find the coupling relationship between feature parameters based on the multi-channel cross network to generate the actual parameter set.

[0068] refer to Figure 2 Multi-channel cross-connect networks include: The input layer is used to input the theoretical parameter set X and contribution information W; The feature embedding layer projects the input theoretical parameter set X into a high-dimensional space through a fully connected layer to generate a high-dimensional vector: The CrossNet network initializes its weight matrix with contribution information W and sets fine-tuning boundaries. High-dimensional vectors are input into the CrossNet network to generate and display high-order features. The fully connected network is a three-layer fully connected network that extracts the hidden features between the input theoretical parameter set X. The feature fusion layer, based on a gated fusion mechanism, fuses and displays high-order features and latent features, and outputs fused features. The multi-task output layer generates the actual parameter set based on the output fusion features.

[0069] The multi-channel cross-network provided in this application is a neural network with multiple parameter outputs. The key to this embodiment is to use contribution information to correct the initial weights in the CrossNet network and set fine-tuning boundaries, thereby guiding the correction direction of the CrossNet network.

[0070] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A simulation platform for network-connected testing of a camera based on multi-physics coupling, characterized in that, include: A power grid fault generation device is used to provide the power grid supply environment for synchronous condenser groups and simulate fault waveforms; An electromagnetic field theory calculation device, based on the transient finite element method, calculates the theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation of a synchronous condenser group when the input current of a fault waveform is applied. The thermal field theoretical calculation device, based on the transient heat conduction-convection equation, generates the theoretical temperature field and theoretical temperature rise curve of the condenser group according to the theoretical electromagnetic loss and fluid heat transfer coefficient. The structural field calculation device generates theoretical deformation characteristic data of the phase-shifting group and theoretical air gap eccentricity based on theoretical electromagnetic force density and theoretical temperature field. A multi-source input coupling device generates an actual parameter set based on a theoretical parameter set. The theoretical parameter set includes theoretical electromagnetic force density, theoretical electromagnetic loss, theoretical torque pulsation, theoretical temperature field distribution, theoretical temperature rise curve, deformation characteristic data, and air gap eccentricity coupling. The actual parameter set includes actual electromagnetic force density, actual electromagnetic loss, actual torque pulsation, actual temperature field distribution, actual temperature rise curve data, actual deformation characteristic data, and actual air gap eccentricity coupling. The safety boundary calculation device generates the corresponding fault waveform of the camera group based on the actual parameter set, adjusts the fault probability of the camera group, calculates the fault probability of the camera group under different fault waveforms, and generates a simulation experiment report.

2. The simulation platform for network-connected camera based on multiphysics coupling as described in claim 1, characterized in that, The electromagnetic field theory calculation device includes: The coordinate generation module is used to determine the origin on the motor shaft of the synchronous condenser group, take the z-axis as the axial direction of the rotation shaft, and make the plane of the x-axis and y-axis perpendicular to the z-axis to establish a three-dimensional rectangular coordinate system. The synchronous condenser response calculation module calculates the stator terminal current, rotor mechanical angle, and excitation current of the synchronous condenser based on the fault waveform. The electromagnetic field theory calculation module calculates the theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation based on the stator end current, rotor mechanical angle, and excitation current.

3. The simulation platform for network-connected camera based on multiphysics coupling as described in claim 1, characterized in that, The thermal field theory calculation device includes: Temperature information loading module, used to load the heat exchange efficiency at various locations in the heat exchange system; The temperature calculation module generates a theoretical temperature field and a theoretical temperature rise curve based on electromagnetic loss, initial temperature distribution, and heat transfer efficiency at each location. The theoretical temperature field is the temperature distribution of each region in a three-dimensional rectangular coordinate system over time, and the theoretical temperature rise curve is the temperature rise rate of key points in the three-dimensional rectangular coordinate system over time.

4. The simulation platform for network-connected camera based on multiphysics coupling as described in claim 1, characterized in that, The structural field calculation device includes: The structural information loading module is used to load the initial geometric state of the camera module. The structural information calculation module generates theoretical deformation characteristic data and theoretical air gap eccentricity based on the initial geometric state, theoretical temperature field, and electromagnetic force density.

5. The simulation platform for network-connected camera based on multiphysics coupling as described in claim 1, characterized in that, The simulation platform for network-connected testing of a camera based on multi-physics coupling also includes a correction device; The correction device includes: The electromagnetic correction module has a built-in first correction neural network model, which is used to correct the calculated theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque ripple. The thermal field correction module has a built-in second correction neural network model, which is used to correct the calculated theoretical temperature field and theoretical temperature rise curve. The structural correction module has a built-in third correction neural network model, which is used to correct the calculated theoretical deformation characteristic data and theoretical air gap eccentricity. The first, second, and third corrected neural network models are used to correct the error between the theoretical calculation value and the actual value to obtain the correction parameters.

6. The simulation platform for network-connected camera based on multiphysics coupling according to claim 5, characterized in that, The first modified neural network model is a temporal feature extraction network based on bidirectional LSTM; The second modified neural network model is a multi-scale feature fusion network based on 3D convolutional coding; The third modified neural network model is a robust shallow network based on feature expansion.

7. The simulation platform for network-connected camera based on multiphysics coupling according to claim 6, characterized in that, The training data for the first, second, and third modified neural network models are theoretical and actual parameters obtained when the camera group is in a steady state. The theoretical parameters are used as feature data, and the actual parameters are used as annotation data. Among them, the variation ranges of theoretical temperature field, theoretical temperature rise curve, theoretical deformation characteristic data, and theoretical air gap eccentricity in the data used to train the first modified neural network model are maintained within a preset range. In the data used to train the second modified neural network model, the variation ranges of electromagnetic force density, theoretical electromagnetic loss, theoretical torque pulsation, theoretical deformation characteristic data, and theoretical air gap eccentricity are maintained within a preset range. In the data used to train the third modified neural network model, the variation amplitudes of electromagnetic force density, theoretical electromagnetic loss, theoretical torque ripple, electromagnetic force density, theoretical electromagnetic loss, and theoretical torque ripple are maintained within a preset range.

8. The simulation platform for network-connected camera based on multiphysics coupling according to claim 1, characterized in that, The multi-source input coupling device includes: The information input module is used to obtain the theoretical parameter set X; ;in, Represents the empty set. , , These represent the theoretical electromagnetic force density, theoretical electromagnetic loss, and theoretical torque pulsation, respectively. , These represent the theoretical temperature field and the theoretical temperature rise curve, respectively. , These represent the corrected theoretical deformation characteristic data and the theoretical air gap eccentricity, respectively. The data augmentation module, based on the Group Lasso model, labels the true contribution of each parameter in the theoretical parameter group X to each parameter in the actual parameter group, and obtains contribution information. The large model coupling calculation module is used to input the theoretical parameter set X and contribution information, and to find the coupling relationship between feature parameters based on the multi-channel cross network to generate the actual parameter set.

9. The simulation platform for network-connected camera based on multiphysics coupling according to claim 8, characterized in that, The methods for obtaining contribution information include the following steps: S1: Take all the theoretical parameter sets X collected as the observation variable matrix XT, and take all the actual parameter sets collected as the response variable matrix YT; Wherein, the dimension of the observation variable matrix XT is n*p, and the dimension of the response variable moment YT is n*p, where n represents the number of samples collected, and p=7; S2: Establish an influence group for each element in the actual parameter set Y. Where b represents the index of an element in the actual parameter group Y, b∈[1,7], and l is the index that affects the grouping. b and l are positive integers. This represents the total number of influence groups for the b-th parameter in the actual parameter group; S3: Get all impact groups Set each affected group Initialization coefficient vector g represents the index that affects the grouping, and k represents the number of iterations; set the initial step size. Step size reduction factor Decrease coefficient and loss function ; S4: Perform the following iterative operation for each influence group until convergence. While performing the iterative operation for each influence group, the coefficient vectors of the other influence groups remain fixed. Based on the current coefficient vector Group design matrix X g Calculate gradient and Heisenberg scalar h g ; ； ; This represents the pre-defined loss function, g represents the index that affects the grouping, and X represents the index that affects the grouping. g =Column indices in the observed variable matrix XT belong to the influence grouping submatrix, Represents the gradient of the loss function. This represents the sub-vector corresponding to the l-th group in the complete coefficient vector of the k-th iteration; ； This represents the loss function with respect to the coefficient vector of the g-th group. Denotes the minimum curvature constant. represents an approximate scalar of a Hessian matrix block, max represents the vector maximum value operation, and diag represents the matrix diagonal extraction operation; Calculate the auxiliary vector u; ； S5: Determine the update direction of the coefficient vector based on the auxiliary vector; ； Indicates the direction of the update. λ represents the penalty threshold, and λ represents the weighting coefficient. S5: Calculate the reference value based on the update direction. Calculate the optimal step size based on the reference value. : ； Indicates the expected decrease. Represents the matrix transpose symbol; S6: Based on the optimal step size Update the coefficient vector; ； S6: Determine whether the iteration should stop based on the convergence condition. The convergence condition is: If the iteration condition is met, the iteration stops; otherwise, the iteration continues. S7: Obtain the coefficient vector of all factors affecting the grouping. The coefficient vector is used as contribution information.

10. The simulation platform for network-connected camera based on multiphysics coupling according to claim 9, characterized in that, Multi-channel cross-connect networks include: The input layer is used to input the theoretical parameter set X and contribution information W; The feature embedding layer projects the input theoretical parameter set X into a high-dimensional space through a fully connected layer to generate a high-dimensional vector: The CrossNet network initializes its weight matrix with contribution information W and sets fine-tuning boundaries. High-dimensional vectors are input into the CrossNet network to generate and display high-order features. The fully connected network is a three-layer fully connected network that extracts the hidden features between the input theoretical parameter set X. The feature fusion layer, based on a gated fusion mechanism, fuses and displays high-order features and latent features, and outputs fused features. The multi-task output layer generates the actual parameter set based on the output fusion features.

Images