A method, system, device and medium for determining installation mode combination of electronic transformer acquisition unit

Abstract

The application discloses a kind of methods, systems, equipment and medium for determining electronic mutual inductor acquisition unit installation mode combination, belong to mutual inductor installation technical field, method includes: obtaining multi-source data to construct training set;Construct probability classification and index regression model;Based on multi-objective loss function joint training optimization model;Input measured feature output optimal installation combination.System includes data acquisition and mapping module, sample set construction module, probability classification modeling module, index regression modeling module, joint optimization training module, installation scheme prediction module.The application breaks the barrier of theoretical calculation and engineering practice through multi-model joint architecture and multidimensional constraint mechanism.Electromagnetic interference, construction difficulty and cost control are included in unified optimization framework, which not only solves the technical problem that acquisition unit is easily disturbed in complex electromagnetic environment, but also replaces artificial experience through intelligent decision, significantly improves the standardization level and comprehensive economic benefit of substation construction.

Description

Technical Field

[0001] This invention relates to the field of instrument transformer installation technology, and specifically to a method, system, device, and medium for determining the combination of installation methods for electronic instrument transformer acquisition units. Background Technology

[0002] In the fields of power system relay protection, energy metering, and condition monitoring, electronic instrument transformer acquisition units serve as key nodes connecting primary high-voltage equipment and secondary intelligent terminals. The rationality of their installation scheme directly determines the accuracy of data acquisition and the stability of system operation. Currently, existing installation decisions mainly rely on engineering experience and standardized fixed drawings. Technicians typically determine the installation location, wiring routing, and protection methods of the equipment based on past cases or general manufacturer manuals.

[0003] However, this traditional decision-making model has significant technical limitations in practical applications. First, subjective experience struggles to accurately quantify the complex electromagnetic environment characteristics on-site, and the lack of scientific avoidance mechanisms for areas with strong electromagnetic interference makes it highly susceptible to signal distortion or packet loss due to improper installation of the acquisition unit, thus affecting the reliability of the protection device. Second, standardized solutions lack flexibility and cannot adapt to the special conditions commonly encountered in the renovation of old stations, such as limited space and dense equipment, often resulting in high construction difficulty and renovation costs, and making it difficult to find the optimal balance between technical performance indicators and project resource consumption. Furthermore, existing technologies fail to effectively utilize digital simulation and on-site measurement data for forward-looking evaluation, lacking a comprehensive consideration of the equipment's heat dissipation performance, ease of maintenance, and long-term operational stability throughout its entire lifecycle. Therefore, there is an urgent need for a superior technical solution that can integrate multi-dimensional environmental characteristics and achieve precise matching between installation methods and on-site conditions. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention provides a method, system, device and medium for determining the installation combination of electronic current transformer acquisition units.

[0005] Therefore, the technical problem solved by the present invention is that the existing installation schemes for current transformer acquisition units mainly rely on manual experience and lack quantitative analysis and comprehensive consideration of the on-site electromagnetic environment, construction difficulty and economic cost, resulting in weak anti-interference ability, poor engineering adaptability and high trial and error costs.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for determining the combination of installation methods for an electronic instrument transformer acquisition unit, comprising acquiring multi-source sample data, extracting installation method combination features and environmental interference features from the multi-source sample data, and mapping them into a standard input vector; generating quantitative and qualitative labels based on the multi-source sample data, and combining the standard input vector, the quantitative labels, and the qualitative labels to construct a training sample set; constructing a probabilistic classification model based on a discrete decision architecture to process the standard input vector to output the probability distribution of different installation method combinations; and constructing an index regression model based on a continuous fitting architecture to determine the probability distribution of different installation method combinations based on the installation method. The optimal installation method combination determined by the probability distribution of the formula combination and the environmental interference characteristics are used as inputs; a multi-objective weighted loss function constrained by technical performance indicators, engineering adaptability and comprehensive cost is constructed; the probability classification model and the index regression model are jointly trained using the training sample set; and the model parameters are iteratively corrected through an error feedback mechanism to generate an installation method combination optimization model; the environmental interference characteristics of the installation scenario are obtained, and a test standard input vector is generated by combining the candidate installation method combinations; the test standard input vector is input into the installation method combination optimization model, and the optimal installation method combination and the predicted technical performance indicators corresponding to the optimal installation method combination are output.

[0007] As a preferred embodiment of the method for determining the installation mode combination of an electronic instrument transformer acquisition unit according to the present invention, the step of acquiring multi-source sample data and extracting installation mode combination features and environmental interference features from the multi-source sample data includes: constructing a digital electromagnetic simulation model to simulate the electromagnetic field distribution under various switching operation conditions and outputting theoretical simulation data; building a field physical test environment to collect electrical state fluctuations and communication signal quality indicators during equipment operation and obtain field measured data; analyzing the theoretical simulation data and the field measured data, selecting elements characterizing the physical installation structure and electrical connection attributes as the installation mode combination features, and extracting elements characterizing the background electromagnetic noise level as the environmental interference features.

[0008] The beneficial effects of this preferred technical solution are as follows: it integrates digital simulation with on-site physical testing, constructing a closed-loop data system that combines virtual and real-world data. By simulating various extreme or difficult-to-reproduce switching operation conditions, it overcomes the limitations of single physical testing; combined with on-site measurements, it captures real electrical fluctuations and background noise, ensuring that the data contains complex environmental interference information. This complementary mechanism of multi-source heterogeneous data effectively improves the completeness and realism of feature extraction, providing a high-fidelity data foundation for subsequent model training, solving the problem of large deviations in practical applications caused by relying solely on theoretical models, and also reducing the time and economic costs of relying entirely on on-site data collection.

[0009] As a preferred embodiment of the method for determining the combination of installation methods for electronic instrument transformer acquisition units according to the present invention, the step of generating quantitative and qualitative labels based on the multi-source sample data includes: parsing the electromagnetic response data in the multi-source sample data, extracting performance index data characterizing the degree of electromagnetic interference, and numerically mapping the performance index data to generate the quantitative label; based on preset engineering implementation constraints, multidimensionally evaluating the construction difficulty and resource consumption level of the combination of installation methods to obtain the evaluation result, and converting the evaluation result into a corresponding discrete level identifier as the qualitative label.

[0010] As a preferred embodiment of the method for determining the combination of installation methods of electronic instrument transformer acquisition units according to the present invention, the step of constructing a probabilistic classification model based on a discrete decision architecture includes: establishing a decision logic structure containing multiple basic discrimination units and defining feature judgment rules for each of the basic discrimination units; inputting the standard input vector into each of the basic discrimination units to obtain the local prediction results output by each of the basic discrimination units for different combinations of installation methods; and globally aggregating and calculating all the local prediction results based on a preset result integration strategy to generate a normalized probability distribution of the installation method combination.

[0011] As a preferred embodiment of the method for determining the combination of installation methods for electronic instrument transformer acquisition units according to the present invention, the step of constructing an index regression model based on a continuous fitting architecture includes: constructing feature fusion logic, concatenating the optimal installation method combination with the environmental interference feature dimension to generate a comprehensive regression input vector; establishing a nonlinear mapping mechanism, projecting the comprehensive regression input vector onto a high-dimensional feature space to resolve the nonlinear coupling relationship between features; configuring a numerical fitting strategy, constructing a regression mapping relationship between the comprehensive regression input vector and the predicted technical performance index in the high-dimensional feature space to output continuous predicted values.

[0012] The beneficial effects of this preferred technical solution are as follows: By employing a continuous fitting architecture to process the feature fusion logic, the complex nonlinear coupling problem between installation methods and environmental interference is effectively solved. By establishing a nonlinear mapping mechanism to project the comprehensive regression input vector into a high-dimensional space, the potential correlation patterns in the data are deeply mined, overcoming the bottleneck of traditional linear models' inability to characterize the impact of complex electromagnetic environments. This solution can output continuous and accurate predicted values, not only achieving refined prediction of technical performance indicators but also providing a quantitative basis for selecting the most suitable installation scheme, significantly improving the granularity and accuracy of the prediction results.

[0013] As a preferred embodiment of the method for determining the combination of installation methods for electronic instrument transformer acquisition units according to the present invention, the construction of a multi-objective weighted loss function constrained by technical performance indicators, engineering adaptability, and comprehensive cost includes: constructing a first loss component characterizing the technical performance optimization objective based on the numerical deviation between the predicted technical performance indicators and the quantitative labels; constructing a second loss component characterizing the engineering adaptability and comprehensive cost constraints based on the distribution difference between the probability distribution of the installation method combination and the ideal category indicated by the qualitative labels; and configuring corresponding weighting coefficients for the first loss component and the second loss component according to the degree of importance attached to technical performance, engineering adaptability, and comprehensive cost in different application scenarios, and linearly superimposing them to generate the multi-objective weighted loss function.

[0014] The beneficial effects of this preferred technical solution are as follows: A multi-objective weighted loss function is established, achieving a dynamic balance among technical performance, engineering adaptability, and overall cost. By constructing a first loss component representing performance optimization and a second loss component representing engineering constraints, multi-dimensional evaluation indicators are incorporated into a unified optimization framework. It supports flexible configuration of weighting coefficients according to specific application scenarios, enabling the model to pursue both optimal anti-interference performance and consideration of construction difficulty and economic benefits. This avoids the difficulties in engineering implementation or resource waste caused by single-objective optimization, significantly improving the practical engineering application value of the decision-making solution.

[0015] As a preferred embodiment of the method for determining the installation combination of electronic current transformer acquisition units according to the present invention, the step of jointly training the probabilistic classification model and the index regression model using the training sample set, and iteratively correcting the model parameters through an error feedback mechanism includes: performing a forward inference step, inputting the training sample set into the current probabilistic classification model and the index regression model to generate the current probability distribution prediction result and performance index prediction result; performing an error evaluation step, substituting the probability distribution prediction result and the performance index prediction result into the multi-objective weighted loss function to calculate the comprehensive loss value characterizing the current model deviation; and performing a parameter correction step, generating an error correction amount for each model parameter based on the comprehensive loss value, and synchronously updating the internal parameters of the probabilistic classification model and the index regression model until the comprehensive loss value meets the preset convergence condition.

[0016] To address the aforementioned technical problems, the present invention also provides the following technical solution: a system for determining the combination of installation methods for electronic instrument transformer acquisition units, comprising a data acquisition and mapping module for acquiring multi-source sample data, extracting installation method combination features and environmental interference features from the multi-source sample data, and mapping them into a standard input vector; a sample set construction module for generating quantitative and qualitative labels based on the multi-source sample data, and combining the standard input vector, the quantitative labels, and the qualitative labels to construct a training sample set; a probability classification modeling module for constructing a probability classification model based on a discrete decision architecture, wherein the probability classification model is used to process the standard input vector to output the probability distribution of different installation method combinations; and an index regression modeling module for constructing an index regression model based on a continuous fitting architecture. The index regression model takes the optimal installation method combination determined based on the probability distribution of the installation method combination and the environmental interference characteristics as inputs. The joint optimization training module is used to construct a multi-objective weighted loss function constrained by technical performance indicators, engineering adaptability, and comprehensive cost. It uses the training sample set to jointly train the probability classification model and the index regression model, and iteratively corrects the model parameters through an error feedback mechanism to generate an installation method combination optimization model. The installation scheme prediction module is used to obtain the environmental interference characteristics of the installation scenario, combine the candidate installation method combinations to generate a test standard input vector, input the test standard input vector into the installation method combination optimization model, and output the optimal installation method combination and the predicted technical performance indicators corresponding to the optimal installation method combination.

[0017] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method for determining the combination of installation methods of electronic current transformer acquisition units.

[0018] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for determining the combination of installation methods of electronic current transformer acquisition units.

[0019] The beneficial effects of this invention are as follows: This invention employs a dual-architecture model combining discrete decision-making and continuous fitting to mine optimal installation patterns from multi-source data. By constructing a multi-objective loss function that incorporates technical performance, engineering adaptability, and overall cost, it overcomes the limitations of single-index optimization, ensuring equipment anti-interference performance while also considering construction feasibility and economic benefits. Furthermore, the joint training mechanism based on error feedback significantly improves the model's generalization ability and prediction accuracy, enabling it to quickly output the optimal installation combination that combines superior technical indicators with engineering feasibility for complex and ever-changing substation environments, effectively reducing construction risks. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a method for determining the combination of installation methods for electronic instrument transformer data acquisition units; Figure 2 This is a wiring diagram for short leads; Figure 3 This is a schematic diagram of the TEV waveform simulated under a short lead. Figure 4 This is a wiring diagram for one type of long lead wire; Figure 5 For use such Figure 4 The diagram shows the TEV waveform under the long lead. Figure 6 This is a comparative diagram of two different installation methods; Figure 7 This is another wiring diagram for long leads; Figure 8 For use such Figure 7 The diagram shows a simulated TEV waveform of a long lead wire. Figure 9 This is a wiring diagram for a vertical short lead; Figure 10 A schematic diagram of the TEV waveform for a vertical lead; Figure 11 This is a schematic diagram of a long grounding wire grounding method.

[0022] Figure 12 This is a schematic diagram of a system for determining the combination of installation methods for electronic instrument transformer data acquisition units. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present invention, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.

[0024] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for determining the installation combination of electronic instrument transformer acquisition units is provided, comprising: S1: Obtain multi-source sample data, extract installation method combination features and environmental interference features from the multi-source sample data, and map them into a standard input vector.

[0025] S2: Generate quantitative and qualitative labels based on the multi-source sample data, and combine the standard input vector, the quantitative labels and the qualitative labels to construct a training sample set.

[0026] S3: Construct a probabilistic classification model based on a discrete decision architecture to process the standard input vector and output the probability distribution of different installation method combinations.

[0027] S4: Construct an index regression model based on a continuous fitting architecture, using the optimal combination of installation methods determined based on the probability distribution of the installation method combination and the environmental interference characteristics as input.

[0028] S5: Construct a multi-objective weighted loss function constrained by technical performance indicators, engineering adaptability, and overall cost. Use the training sample set to jointly train the probability classification model and the index regression model. Iteratively correct the model parameters through an error feedback mechanism to generate an optimized model combining installation methods.

[0029] S6: Obtain the environmental interference features of the installation scenario, combine the candidate installation method combinations to generate a test standard input vector, input the test standard input vector into the installation method combination optimization model, and output the optimal installation method combination and the prediction technical performance index corresponding to the optimal installation method combination.

[0030] It should be noted that existing technical solutions mainly rely on manual experience or general atlases to determine the installation method, making it difficult to quantify and assess the specific impact of complex electromagnetic environments on the acquisition unit. In actual engineering, due to the lack of detailed analysis of the electromagnetic field distribution characteristics at different installation locations, the selected installation points are often located in areas of strong interference, causing signal distortion or transmission interruption and reducing the observation accuracy of the power system. Furthermore, existing methods cannot establish a precise mapping relationship between construction difficulty, resource consumption, and technical performance. This leads to installation schemes either excessively pursuing performance while ignoring the objective limitations of on-site construction, or sacrificing key electrical monitoring indicators to reduce construction difficulty, making it difficult to achieve multi-dimensional globally optimal configuration.

[0031] Therefore, to address the aforementioned problems, this invention establishes a scientific decision-making system for installation methods through steps such as acquiring multi-source sample data and extracting features, generating quantitative and qualitative labels, constructing a probabilistic classification model based on a discrete decision-making architecture, constructing an index regression model based on a continuous fitting architecture, constructing a multi-objective weighted loss function, and jointly training the model using a training sample set. This application integrates digital electromagnetic simulation and on-site measured data to comprehensively capture environmental characteristics and equipment response patterns. It utilizes probabilistic classification and index regression models to achieve qualitative applicability judgment and quantitative performance prediction for different installation combinations. Furthermore, it jointly optimizes model parameters based on a multi-objective weighted loss function, ensuring that technical indicators meet requirements while considering engineering adaptability and overall cost. This method can intelligently select the most suitable installation combination for the on-site working conditions from a massive number of potential solutions, effectively solving the problem of difficult selection of acquisition unit installations in complex environments, and significantly improving the digitalization level and engineering economy of power grid construction.

[0032] Example 2, Reference Figures 1 to 12 This is one embodiment of the present invention. Based on the previous embodiment, a method for determining the combination of installation methods of electronic current transformer acquisition units is provided.

[0033] In this embodiment of the application, multi-source sample data is acquired in step S1, and installation method combination features and environmental interference features are extracted from the multi-source sample data and mapped to a standard input vector, including the following steps A1-A3: A1: Construct a digital electromagnetic simulation model to simulate the electromagnetic field distribution under various switching operation conditions and output theoretical simulation data; A2: Set up an on-site physical testing environment, collect electrical state fluctuations and communication signal quality indicators during equipment operation, and obtain on-site measured data; A3: Analyze theoretical simulation data and field measurement data to select elements characterizing the physical installation structure and electrical connection attributes as features for installation method combinations. And extract elements characterizing the background electromagnetic noise level as environmental interference features. Subsequently, a feature normalization mapping function is constructed to concatenate the above features and transform them into a standard input vector. Its expression is: in, and These represent the mean and standard deviation of the corresponding feature dimensions, ensuring that the distribution of the input data meets the stability requirements of model training.

[0034] In constructing the digital electromagnetic simulation model and screening the combination features of installation methods, this embodiment covers a variety of typical physical connection forms as the basic sample space. For example... Figure 2 , Figure 4 , Figure 7 , Figure 9 and Figure 11 As shown, the installation method combinations encompass different lead lengths, lead directions (such as vertical or horizontal), and grounding wire configurations. Figure 2 A short lead connection architecture is demonstrated, which is compact and theoretically less affected by spatial electromagnetic field coupling. Figure 4 and Figure 7 The long lead connection architecture under different path planning is shown to simulate the parasitic inductance effect in complex wiring scenarios. Figure 9 The vertical lead arrangement is shown, and Figure 11 The diagram illustrates long grounding wire connection methods. The model extracts feature vectors characterizing the spatial geometric properties of these different topologies (i.e., the physical installation structures shown in the diagram), serving as the basis for subsequently distinguishing different installation categories.

[0035] In one optional implementation, in step S1, multi-source sample data is acquired. The installation method combination features and environmental interference features are extracted from the multi-source sample data. Furthermore, a high-precision three-dimensional finite element simulation model can be constructed to simulate transient electromagnetic processes under extreme conditions such as lightning overvoltage and short-circuit fault impact, and output electromagnetic field change curves in the entire time domain. These theoretical limit data are used to supplement the boundary samples that are difficult to cover by conventional physical tests, thereby enriching the response characteristics of the installation method combination in harsh electrical environments and ensuring that the model can learn the anti-interference performance boundary of the equipment under critical conditions.

[0036] In another optional implementation, the acquisition of multi-source sample data in step S1, and the extraction of installation method combination features and environmental interference features from the multi-source sample data, can also be achieved by deploying a wide-frequency spectrum monitoring device to capture background noise data across the entire frequency band from power frequency to high-frequency switching oscillation in the field physical test environment. These noise data are then mapped to the physical spatial coordinates of the primary equipment in the substation to generate an electromagnetic interference spectrum with spatial attributes. This allows for the extraction of environmental interference features that characterize the complexity of the electromagnetic environment in different areas, providing accurate data support for subsequent models to distinguish the signal-to-noise ratio differences of different installation locations.

[0037] Specifically, in step S2, quantitative and qualitative labels are generated based on multi-source sample data, including the following steps B1-B2: B1: Analyze electromagnetic response data from multi-source sample data, extract performance index data characterizing the degree of electromagnetic interference, including but not limited to signal-to-noise ratio and bit error rate, and numerically map and process the performance index data to generate quantitative labels. This label is a continuous numerical value used to characterize the quality of technical performance; When outputting theoretical simulation data and extracting performance index data, the model will output the corresponding electromagnetic transient response for the different physical installation structures mentioned above. For example... Figure 3 , Figure 5 , Figure 8 and Figure 10 The figures shown correspond to the TEV simulation waveforms for the different installation structures described above. (Comparison) Figure 3 The short lead waveform shown is similar to Figure 5 As shown in the waveform of the long lead, there are significant differences in signal amplitude, oscillation frequency, and attenuation characteristics under different installation methods. For example... Figure 5 The long-lead waveform shown exhibits more intense oscillations in a specific frequency band, indicating a higher degree of susceptibility to environmental interference. Step B1 specifically involves analyzing these waveform data to extract key elements such as peak amplitude and dominant frequency components, and then numerically mapping these elements to generate quantitative labels. This allows for a quantitative characterization of the technical performance of the current installation method.

[0038] B2: Based on pre-set engineering implementation constraints, conduct a multi-dimensional evaluation of the construction difficulty and resource consumption level of the installation method combination, obtain the evaluation results, and convert the evaluation results into corresponding discrete level labels, including but not limited to optimal, feasible, restricted, and infeasible, as qualitative labels. It is represented using one-hot encoding.

[0039] When generating qualitative labels, it is necessary to compare the overall effects of different installation methods. For example... Figure 6 As shown, by visually comparing the response curves of two different installation methods, the differences in advantages and disadvantages can be clearly identified. For example, the solution represented by the solid line has less fluctuation and converges faster. Based on the differences reflected in this comparative diagram and combined with preset engineering constraints, various installation combinations are classified into optimal, feasible, or infeasible levels, thereby generating qualitative labels for training the classification model. .

[0040] In this embodiment of the application, the probabilistic classification model is constructed based on the discrete decision architecture in step S3, including the following steps C1-C3: C1: Establish a decision logic structure containing multiple basic discrimination units, and define the feature determination rules for each basic discrimination unit; C2: Transfer the standard input vector Input each basic discrimination unit and obtain the local prediction results output by each basic discrimination unit for different combinations of installation methods; C3: Based on a preset result integration strategy, globally aggregate all local prediction results and use the Softmax activation function to generate a normalized probability distribution of installation method combinations. Assume it exists. The possible combinations of installation methods, then the first The probability of a combination The calculation formula is: in, The unnormalized log-probability vector output by the decision logic structure. This characterizes the use of the first input feature under the current input features. Recommended confidence levels for combinations of installation methods.

[0041] In an optional implementation, the probabilistic classification model built on the discrete decision architecture in step S3 can also adopt an ensemble learning strategy, using multiple independent decision tree classifiers as basic discriminant units to build a random forest structure. By performing random subspace sampling of the feature dimension on the standard input vector, different decision trees are driven to focus on the local patterns of physical structure features or environmental interference features respectively. Finally, the output results are aggregated through majority voting or probability averaging, thereby reducing the model's dependence on a single feature source and enhancing its robustness in the face of differences between simulation and measured data distributions.

[0042] In another optional implementation, the probabilistic classification model built on the discrete decision architecture in step S3 can also construct a hierarchical cascaded decision screening network. First, a logic gating unit based on hard rules is deployed as the primary discriminant layer. Based on rigid constraints such as electrical safety distance and mechanical strength limits in power industry standards, the probability propagation of non-compliant installation methods is directly blocked. Then, the candidate combinations that have passed the initial screening are input into the secondary discriminant layer based on Bayesian inference. The conditional probability distribution is calculated by combining prior environmental knowledge. This coarse-to-fine architecture can significantly reduce invalid calculations and improve inference efficiency while ensuring the safety and compliance of the solution.

[0043] Specifically, in step S4, an index regression model is constructed based on a continuous fitting architecture, including the following steps D1-D3: D1: Construct feature fusion logic and select probability distribution. The optimal installation method combination feature with the highest probability value is concatenated with the environmental interference feature dimension to generate a comprehensive regression input vector. ; D2: Establish a nonlinear mapping mechanism, using kernel functions or multilayer perceptrons to project the comprehensive regression input vector to a high-dimensional feature space to analyze the nonlinear coupling relationships between features. Let the nonlinear mapping function be... Then the high-dimensional feature is represented as ; D3: Configure a numerical fitting strategy to construct a regression mapping relationship between the comprehensive regression input vector and the performance index of the prediction technique in a high-dimensional feature space, and output continuous prediction values. The regression model can be expressed as: in, For the regression weight matrix, This is a bias term. This step aims to predict the specific performance values ​​of the selected installation method under a given environment.

[0044] Specifically, in step S5, a multi-objective weighted loss function is constructed, constrained by technical performance indicators, engineering adaptability, and overall cost, including the following steps E1-E3: E1: Based on predictive technology performance indicators With quantitative label The degree of numerical deviation between them is used to construct the first loss component representing the performance optimization objective of the technology. It is usually expressed in the form of mean square error: E2: Probability distribution based on installation method combinations Qualitative Labels The distributional differences between the indicated ideal categories are used to construct a second loss component characterizing engineering adaptability and overall cost constraints. The cross-entropy loss function is used in the following form: E3: Based on the different application scenarios and their varying degrees of emphasis on technical performance, engineering adaptability, and overall cost, corresponding weighting coefficients are configured for the first and second loss components. and And linearly superimpose to generate a multi-objective weighted loss function. : By adjusting and The ratio of these ratios allows the model to achieve a dynamic balance between pursuing the ultimate technical indicators and meeting engineering implementation constraints.

[0045] In this embodiment of the application, regarding step E3, the configuration of corresponding weighting coefficients based on the emphasis placed on technical performance, engineering adaptability, and overall cost in different application scenarios is achieved by employing a static mapping matching strategy based on scenario characteristics, including the following steps E311 to E313: E311: Parse the task description text and environmental constraint parameters of the current installation scenario to identify the standard category label to which the scenario belongs, such as a new high-voltage substation, renovation of an old power distribution room, or a temporary emergency repair project.

[0046] E312: Retrieve a pre-set expert experience database and locate a weight configuration template that matches the standard category label. This template contains weight values ​​for the first and second loss components that have been verified by historical engineering data.

[0047] E313: Directly retrieve the values ​​from the weight configuration template and assign them to the weighting coefficients respectively. and This is to ensure that the optimization direction of model training conforms to the general engineering specifications for this type of scenario.

[0048] In an optional implementation, in step E3, corresponding weighting coefficients are configured based on the emphasis placed on technical performance, engineering adaptability, and overall cost according to different application scenarios. The configuration method can also be a dynamic gradient adjustment strategy based on the training process, including the following steps E321 to E323: E321: In the initial stage of model training, initialize the weighting coefficients of the second loss component. A larger value forces the model to prioritize learning basic installation rules that satisfy engineering adaptability and overall cost constraints, quickly eliminating infeasible solutions.

[0049] E322: Real-time monitoring of the convergence trend of training iterations and overall loss value, constructing a decay function that monotonically decreases with the number of iterations, and calculating the dynamic weight factor for the current stage.

[0050] E323: Gradually reduce the weighting coefficient of the second loss component based on the aforementioned dynamic weighting factor. And simultaneously increase the weighting coefficient of the first loss component. This guides the model to shift its focus to fine-tuning of technical performance metrics during later training stages.

[0051] In another optional implementation, in step E3, corresponding weighting coefficients are configured according to the importance attached to technical performance, engineering adaptability, and overall cost in different application scenarios. The configuration method can also be a multivariate inference strategy based on fuzzy logic control, including the following steps E331 to E333: E331: Define cost sensitivity and performance requirement as fuzzy input variables, and design corresponding membership functions to transform specific scenario parameters into low, medium, and high fuzzy linguistic variables.

[0052] E332: Construct a fuzzy inference rule base, formulate logical rules that increase the weight of the second loss component if the cost sensitivity is high and the performance requirement is medium, and perform implication operations through the fuzzy inference engine.

[0053] E333: The centroid method is used to perform defuzzification processing on the fuzzy set of inference output, and the accurate continuous values ​​are calculated as the weighting coefficients of the first loss component and the second loss component, so as to achieve flexible adaptation to non-standardized scenarios.

[0054] Specifically, in step S6, the probabilistic classification model and the index regression model are jointly trained using the training sample set, and the model parameters are iteratively corrected through an error feedback mechanism, including the following steps F1-F3: F1: Perform the forward inference step, input the training sample set into the current probability classification model and index regression model, and generate the current probability distribution prediction result and performance index prediction result; F2: Perform the error assessment step, substituting the probability distribution prediction results and performance index prediction results into the multi-objective weighted loss function to calculate the comprehensive loss value representing the current model bias. ; F3: Perform parameter correction steps, calculating parameters for each model based on the comprehensive loss value. gradient The gradient descent algorithm is used to simultaneously update the internal parameters of the probabilistic classification model and the index regression model. The parameter update formula is as follows: in, For learning rate, This is the number of iterations. The process is repeated until the combined loss value is reached. The model converges to a preset threshold, thereby generating the final optimized model for the combination of installation methods.

[0055] In this embodiment of the application, regarding step F3, which involves generating error correction amounts for each model parameter based on the comprehensive loss value and synchronously updating the internal parameters, the update method employs a stochastic gradient descent algorithm with momentum, including the following steps F311 to F313: F311: Calculate the overall loss value The gradient vector at the current moment is obtained by taking the partial derivative of each weight parameter relative to the probabilistic classification model and the index regression model.

[0056] F312: Introducing a momentum term variable, the gradient vector at the current time step is weighted and averaged with the momentum update at the previous time step to suppress oscillations during training and accelerate convergence.

[0057] F313: Multiply the preset learning rate by the calculated momentum update amount to obtain the final correction magnitude of each parameter, and perform a subtraction operation to complete the parameter update iteration.

[0058] In an optional implementation, in step F3, for generating error correction amounts for each model parameter based on the comprehensive loss value and synchronously updating the internal parameters, the updating method can also be to use an adaptive moment estimation optimization algorithm, including the following steps F321 to F323: F321: Calculate the Chebyshev first moment estimate (mean of the gradient) and the second moment estimate (uncentered variance of the gradient) of the gradient of the integrated loss value to capture the distribution characteristics of the gradient.

[0059] F322: Performs bias correction operation on the first-order moment estimate and the second-order moment estimate to eliminate the influence of zero bias caused by gradient sparsity in the early stage of training and obtain the corrected moment estimate.

[0060] F323: An adaptive adjustment factor is constructed based on the corrected first and second moments, which dynamically allocates an independent update step size to each model parameter, thereby balancing the learning rate difference between the probabilistic classification model and the index regression model during the parameter update process.

[0061] In another optional implementation, in step F3, for generating error correction amounts for each model parameter based on the comprehensive loss value and synchronously updating the internal parameters, the updating method can also be a dual-channel asynchronous update strategy using the alternating direction multiplier method, including the following steps F331 to F333: F331: Decouple the model parameter set into a subset of classifier parameters and a subset of regressor parameters. In odd-numbered iterations, keep the subset of regressor parameters unchanged and only use the classification-related components in the comprehensive loss value to calculate the gradient and update the subset of classifier parameters.

[0062] F332: In even-numbered iterations, the subset of classifier parameters is kept unchanged, and the gradient is calculated and the subset of regressor parameters is updated only using the regression correlation component in the comprehensive loss value, so as to achieve alternating optimization of the two sub-models.

[0063] F333: Perform a global joint fine-tuning once every preset synchronization period. Use the complete comprehensive loss value to apply a small perturbation correction to the subset of classifier parameters and the subset of regressor parameters at the same time, to ensure the coupling consistency between the two in the feature sharing layer.

[0064] Example 3, Reference Figure 12The above is an illustrative scheme for a method of determining the installation combination of electronic instrument transformer data acquisition units. It should be noted that the technical solution of this system for determining the installation combination of electronic instrument transformer data acquisition units and the technical solution of the method for determining the installation combination of electronic instrument transformer data acquisition units described above belong to the same concept. Details not described in detail in the technical solution of the system for determining the installation combination of electronic instrument transformer data acquisition units in this embodiment can be found in the description of the technical solution of the method for determining the installation combination of electronic instrument transformer data acquisition units described above.

[0065] This embodiment also provides a system for determining the combination of installation methods for electronic instrument transformer data acquisition units, including: The data acquisition and mapping module is used to acquire multi-source sample data, extract installation method combination features and environmental interference features from the multi-source sample data, and map them into a standard input vector. The sample set construction module is used to generate quantitative and qualitative labels based on the multi-source sample data, and to combine the standard input vector, the quantitative labels and the qualitative labels to construct a training sample set; The probability classification modeling module is used to construct a probability classification model based on a discrete decision architecture. The probability classification model is used to process the standard input vector to output the probability distribution of different installation method combinations. The index regression modeling module is used to construct an index regression model based on a continuous fitting architecture, taking the optimal combination of installation methods determined based on the probability distribution of the installation method combination and the environmental interference characteristics as inputs. The joint optimization training module is used to construct a multi-objective weighted loss function constrained by technical performance indicators, engineering adaptability and comprehensive cost. It uses the training sample set to jointly train the probability classification model and the index regression model, and iteratively corrects the model parameters through an error feedback mechanism to generate an installation method combined optimization model. The installation scheme prediction module is used to obtain the environmental interference characteristics of the installation scenario, generate a test standard input vector by combining the candidate installation method combinations, input the test standard input vector into the installation method combination optimization model, and output the optimal installation method combination and the prediction technical performance index corresponding to the optimal installation method combination.

[0066] This embodiment also provides an electronic device applicable to a method for determining a combination of installation methods for electronic instrument transformer acquisition units, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for determining a combination of installation methods for electronic instrument transformer acquisition units as proposed in the above embodiment.

[0067] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a method for determining the combination of installation methods for electronic instrument transformer acquisition units as proposed in the above embodiments.

[0068] The storage medium proposed in this embodiment and the method for determining the combination of installation methods of electronic current transformer acquisition units proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0069] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0070] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for determining the installation combination of electronic instrument transformer data acquisition units, characterized in that, include: Acquire multi-source sample data, extract installation method combination features and environmental interference features from the multi-source sample data, and map them into a standard input vector; Quantitative and qualitative labels are generated based on the multi-source sample data, and a training sample set is constructed by combining the standard input vector, the quantitative labels, and the qualitative labels. A probabilistic classification model is constructed based on a discrete decision architecture to process the standard input vector and output the probability distribution of different installation method combinations; An index regression model is constructed based on a continuous fitting architecture, with the optimal combination of installation methods determined based on the probability distribution of the installation method combination and the environmental interference characteristics as inputs. A multi-objective weighted loss function constrained by technical performance indicators, engineering adaptability, and overall cost is constructed. The probability classification model and the index regression model are jointly trained using the training sample set. The model parameters are iteratively corrected through an error feedback mechanism to generate an optimized model for installation method combination. The environmental interference characteristics of the installation scenario are obtained, and a standard input vector to be tested is generated by combining the candidate installation method combinations. The standard input vector to be tested is then input into the installation method combination optimization model, and the optimal installation method combination and the prediction technical performance index corresponding to the optimal installation method combination are output.

2. The method for determining the installation combination of electronic instrument transformer acquisition units as described in claim 1, characterized in that: The process of acquiring multi-source sample data and extracting installation method combination features and environmental interference features from the multi-source sample data includes: Construct a digital electromagnetic simulation model to simulate the electromagnetic field distribution under various switching operation conditions and output theoretical simulation data; A physical testing environment was set up on-site to collect electrical state fluctuations and communication signal quality indicators during equipment operation, thereby obtaining on-site measured data. The theoretical simulation data and the field measurement data are analyzed to select elements that characterize the physical installation structure and electrical connection attributes as the installation method combination features, and elements that characterize the background electromagnetic noise level are extracted as the environmental interference features.

3. The method for determining the installation combination of electronic instrument transformer acquisition units as described in claim 1, characterized in that: The generation of quantitative and qualitative labels based on the multi-source sample data includes: The electromagnetic response data in the multi-source sample data is analyzed, the performance index data characterizing the degree of electromagnetic interference is extracted, and the performance index data is numerically mapped and processed to generate the quantitative label. Based on preset engineering implementation constraints, the construction difficulty and resource consumption level of the installation method combination are evaluated in multiple dimensions to obtain evaluation results. The evaluation results are then converted into corresponding discrete level identifiers as qualitative labels.

4. The method for determining the installation combination of electronic instrument transformer acquisition units as described in claim 1, characterized in that: The probabilistic classification model constructed based on the discrete decision architecture includes: Establish a decision logic structure containing multiple basic discrimination units, and define the feature determination rules for each of the basic discrimination units; The standard input vector is input into each of the basic discrimination units to obtain the local prediction results output by each of the basic discrimination units for different combinations of installation methods; Based on a preset result integration strategy, all the local prediction results are globally aggregated and calculated to generate a normalized probability distribution of the installation method combination.

5. The method for determining the installation combination of electronic instrument transformer acquisition units as described in claim 1, characterized in that: The construction of the index regression model based on the continuous fitting architecture includes: Construct a feature fusion logic, which combines the optimal installation method with the environmental interference feature dimension to generate a comprehensive regression input vector; A nonlinear mapping mechanism is established to project the comprehensive regression input vector onto a high-dimensional feature space in order to resolve the nonlinear coupling relationship between features; A numerical fitting strategy is configured to construct a regression mapping relationship between the comprehensive regression input vector and the performance index of the prediction technology in the high-dimensional feature space, so as to output continuous prediction values.

6. A method for determining the installation combination of an electronic instrument transformer data acquisition unit as described in claim 1 or 5, characterized in that: The construction of the multi-objective weighted loss function constrained by technical performance indicators, engineering adaptability, and overall cost includes: Based on the degree of numerical deviation between the predicted technical performance index and the quantitative label, a first loss component characterizing the technical performance optimization objective is constructed. Based on the difference between the probability distribution of the installation method combination and the ideal category indicated by the qualitative label, a second loss component characterizing engineering adaptability and overall cost constraints is constructed. Based on the different application scenarios and their emphasis on technical performance, engineering adaptability, and overall cost, corresponding weighting coefficients are configured for the first loss component and the second loss component, and then linearly superimposed to generate the multi-objective weighted loss function.

7. The method for determining the installation combination of electronic instrument transformer acquisition units as described in claim 1, characterized in that: The step of jointly training the probability classification model and the index regression model using the training sample set, and iteratively correcting the model parameters through an error feedback mechanism, includes: Perform a forward inference step, inputting the training sample set into the current probability classification model and the index regression model to generate the current probability distribution prediction result and performance index prediction result; Perform an error assessment step, substituting the probability distribution prediction results and the performance index prediction results into the multi-objective weighted loss function to calculate the comprehensive loss value characterizing the current model bias; The parameter correction step is performed, which generates error correction amounts for each model parameter based on the comprehensive loss value, and synchronously updates the internal parameters of the probability classification model and the index regression model until the comprehensive loss value meets the preset convergence condition.

8. A system for determining the combination of installation methods for electronic instrument transformer data acquisition units, using the method for determining the combination of installation methods for electronic instrument transformer data acquisition units as described in any one of claims 1 to 7, characterized in that, include: The data acquisition and mapping module is used to acquire multi-source sample data, extract installation method combination features and environmental interference features from the multi-source sample data, and map them into a standard input vector. The sample set construction module is used to generate quantitative and qualitative labels based on the multi-source sample data, and to combine the standard input vector, the quantitative labels and the qualitative labels to construct a training sample set; The probability classification modeling module is used to construct a probability classification model based on a discrete decision architecture. The probability classification model is used to process the standard input vector to output the probability distribution of different installation method combinations. The index regression modeling module is used to construct an index regression model based on a continuous fitting architecture, taking the optimal combination of installation methods determined based on the probability distribution of the installation method combination and the environmental interference characteristics as inputs. The joint optimization training module is used to construct a multi-objective weighted loss function constrained by technical performance indicators, engineering adaptability and comprehensive cost. It uses the training sample set to jointly train the probability classification model and the index regression model, and iteratively corrects the model parameters through an error feedback mechanism to generate an installation method combined optimization model. The installation scheme prediction module is used to obtain the environmental interference characteristics of the installation scenario, generate a test standard input vector by combining the candidate installation method combinations, input the test standard input vector into the installation method combination optimization model, and output the optimal installation method combination and the prediction technical performance index corresponding to the optimal installation method combination.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of a method for determining the combination of installation methods of an electronic instrument transformer acquisition unit as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a method for determining the combination of installation methods of an electronic instrument transformer acquisition unit as described in any one of claims 1 to 7.

Images