Power equipment data prediction and modeling system and method based on intelligent mimicry

Abstract

The invention provides a power equipment data prediction and modeling system and method based on intelligent mimicry, and belongs to the technical field of power equipment monitoring and intelligent operation and maintenance. The problem that the generalization ability of a model is poor due to the fact that a traditional modeling method often lacks flexibility and is difficult to adapt to the requirements of different types of equipment is solved. Comprising an intelligent mimicry data acquisition module, a data preprocessing and feature extraction module, an adaptive model construction module, a model training and optimization module and a prediction output module. The intelligent mimicry data acquisition module is used for acquiring equipment data in real time through a sensor deployed on power equipment; the data preprocessing module cleans data, removes noise, extracts features and improves data quality, the adaptive model building module forms a modeling branch based on an equipment type, an initial model is trained through a small-batch gradient descent and stochastic optimization algorithm, parameters are adjusted through prediction error feedback, and the accuracy and generalization of the model are enhanced.

Description

Technical Field

[0001] The present invention provides a power equipment data prediction and modeling system and method based on intelligent mimicry, belonging to the technical field of power equipment monitoring and intelligent operation and maintenance. Background Art

[0002] In the field of power equipment monitoring and intelligent operation and maintenance technology, accurate data prediction and modeling of power equipment is the key to ensuring the stable operation of the power system. However, traditional power equipment data prediction and modeling methods often face many challenges.

[0003] The data sources of power equipment are extensive and complex, including electrical parameters, mechanical performance indicators, and operating environment data. These data are highly heterogeneous and time-varying, which brings great difficulties to data processing and feature extraction. Traditional data processing methods often find it difficult to effectively extract key features from these data, thus affecting the prediction accuracy of the model. Different types of power equipment have different operating characteristics and failure modes, so it is necessary to build specific models for different types of equipment. However, traditional modeling methods often lack flexibility and are difficult to adapt to the needs of different types of equipment, resulting in poor generalization of the model. As the scale of the power system continues to expand, the amount of data on power equipment is also increasing dramatically, which puts higher requirements on model training and optimization. Traditional training algorithms are often inefficient when processing large-scale data and are prone to falling into local optimal solutions, thus affecting the prediction performance of the model. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a power equipment data prediction and modeling system and method based on intelligent mimicry.

[0005] The technical solution adopted by the present invention is: a power equipment data prediction and modeling system based on intelligent mimicry, characterized in that it includes:

[0006] Intelligent mimicry data acquisition module: used to collect equipment data in real time through sensors deployed on power equipment, the equipment data includes: electrical parameters, mechanical performance indicators and operating environment data;

[0007] Data preprocessing and feature extraction module: electrically connected to the intelligent mimicry data acquisition module, used to remove noise and outliers using a data cleaning algorithm, and extract key features through principal component analysis and wavelet transform;

[0008] Adaptive model building module: electrically connected to the data preprocessing and feature extraction module, the adaptive model building module is used to introduce a neural network adaptive layer based on the principle of intelligent mimicry, automatically optimize neuron connections and weight distribution according to input features, form specific modeling branches for different device types, and build initial models for each branch;

[0009] Model training and optimization module: electrically connected to the adaptive model building module, used to train the model in batches using the input new collected data through small batch gradient descent combined with a stochastic optimization algorithm, and to retroactively adjust the model parameters when the deviation between the predicted value and the actual monitored value exceeds a threshold value based on the model prediction error feedback mechanism;

[0010] Prediction output module: electrically connected to the model training and optimization module, used to output the prediction results of key indicators of power equipment, and realize the prediction and modeling of power equipment data of different types and sizes.

[0011] Furthermore, the intelligent mimicry data acquisition module includes a data acquisition execution unit and a cache and format regularization unit, and the data acquisition execution unit is responsible for real-time acquisition of electrical, mechanical and environmental data of the power equipment according to a set frequency and method;

[0012] The cache and format regularization unit is used to store massive data and regularize data of different sources and formats into a unified standard format.

[0013] Furthermore, the preprocessing and feature extraction module includes four units: data cleaning, principal component analysis, wavelet transformation and feature integration;

[0014] The data cleaning unit is used to identify and remove noise and outliers;

[0015] The principal component analysis unit is used to reduce the dimension of the cleaned data, retain the core information, and extract key features;

[0016] The wavelet transform unit is used to mine local and detailed features and enhance the model's ability to capture data features;

[0017] The feature integration unit is used to integrate the features extracted by the principal component analysis unit and the wavelet transform unit to construct a complete feature set and provide materials for model construction.

[0018] Furthermore, the adaptive model building module includes four units: intelligent mimicry analysis, neural network construction, branch modeling and parameter initialization;

[0019] The intelligent mimicry analysis unit analyzes the input data features based on the intelligent mimicry principle;

[0020] The neural network building unit builds the neural network architecture by introducing the neural network adaptive layer, dynamically optimizes the neuron connection and updates the weight through the neural network adaptive layer. The formula for updating the weight is:

[0021]

[0022] Where: represents the weight, t represents the number of iterations, η represents the learning rate, L(θ) represents the neural network loss function, and θ represents the weight parameter set;

[0023] The branch modeling unit is used to form specific modeling branches for different equipment types in combination with their characteristics, and to construct the initial model of each branch;

[0024] The parameter initialization unit is used to reasonably set the initial parameters for the initial model of each branch.

[0025] Furthermore, the model training and optimization module includes a training execution unit, an error calculation unit, a feedback judgment unit and a parameter adjustment unit;

[0026] The training execution unit is used to train the initial model using newly collected data batches through mini-batch gradient descent combined with a stochastic optimization algorithm, pushing the model parameters to update in the direction of reducing the loss function;

[0027] The error calculation unit is used to calculate the prediction error based on the model prediction value and the actual monitoring value;

[0028] The feedback judgment unit is used to compare the calculated prediction error with the set threshold value, and decide whether to trigger the operation of backtracking adjustment of model parameters according to the comparison result;

[0029] The parameter adjustment unit is used to adjust the model parameters according to the backtracking strategy when the deviation between the predicted value and the actual monitoring value exceeds the threshold;

[0030] When mini-batch gradient descent is combined with a stochastic optimization algorithm, mini-batch gradient descent divides the data by batch size, and then updates the weights through a stochastic algorithm. The updated weight calculation formula is:

[0031]

[0032] Where: represents the sum of squared historical gradients, ε represents a small constant that prevents the denominator from being zero.

[0033] Furthermore, in the model prediction error feedback mechanism, the deviation between the predicted value and the actual monitored value is calculated, and the mean square error is used as a measurement indicator. When the mean square error exceeds the threshold, the model parameters are adjusted retrospectively;

[0034] The retrospective adjustment method is:

[0035] Assume that the current step size is β. When the mean square error exceeds the threshold, the step size is reduced to γβ, where 0<γ<1. The gradient is recalculated and the weight is updated. By backtracking multiple times to adjust the step size and update the weight, the model prediction error is gradually reduced.

[0036] Furthermore, the equipment types in the branch modeling unit include transformers, circuit breakers and generators. For the transformer modeling branch, a model considering the transformer equivalent circuit is constructed, and the transformation ratio, load circuit and winding resistance are used as model input features. The weights are optimized through the neural network adaptive layer to obtain the initial model of the transformer; the model predicts the future winding temperature by learning the relationship between load current, ambient temperature and winding temperature in historical data;

[0037] For the circuit breaker modeling branch, the relationship between its breaking current capacity and contact material and breaking time is considered, and the neuron connection and weight are adjusted through the neural network adaptive layer to build the initial model of the circuit breaker; the model determines the contact material coefficient through training data, and then predicts the contact wear;

[0038] For the generator modeling branch, the rotor speed, excitation current, and mechanical torque parameters are used as model input features. The neuron connections and weights are adjusted through the neural network adaptive layer to build an initial model that reflects the operating status of the generator. The model determines the output power and mechanical energy of the generator through training data, and then predicts the service life of the generator.

[0039] Further, the prediction output module includes a prediction operation unit, a type adaptation unit and an output display unit;

[0040] The prediction operation unit is used to operate the input power equipment data according to the trained and optimized model to obtain the predicted value of the key indicators of the power equipment;

[0041] The type adaptation unit is used to adapt the prediction results to different types of power equipment;

[0042] The output display unit is used to output the sorted and adapted prediction results, realize the display of prediction results of power equipment data of different types and sizes, and provide intuitive information for users.

[0043] Furthermore, when realizing data prediction and modeling of power equipment of different scales, a distributed modeling and prediction method is adopted for large-scale power systems containing multiple power equipment:

[0044] Assume that there are several devices in the system, and build a local model for one of them;

[0045] Each local model shares some feature information through an information exchange mechanism. The overall system prediction value is obtained by fusing the prediction values ​​of each local model, and the weight is determined by minimizing the overall prediction error.

[0046] For small-scale power systems, a single constructed model is directly used for prediction. By adjusting the model parameters and training data, it can adapt to the data prediction and modeling needs of power equipment of different sizes.

[0047] A method for predicting and modeling power equipment data based on intelligent mimicry, using a power equipment data prediction and modeling system based on intelligent mimicry, includes the following steps:

[0048] S1: Collect electrical parameters, mechanical performance indicators and operating environment data of power equipment in real time through the intelligent mimicry data acquisition module;

[0049] S2: Clean the data and extract features through the data preprocessing and feature extraction module;

[0050] S3: The adaptive model building module automatically builds the initial model of each branch according to the input features;

[0051] S4: The model training and optimization module uses the newly collected data to train the model and retrospectively adjusts the model parameters based on the prediction error feedback mechanism;

[0052] S5: The prediction output module outputs the prediction results of key indicators of power equipment, realizing the prediction and modeling of power equipment data of different types and scales.

[0053] The beneficial effects of the present invention compared with the prior art are as follows: the power equipment data prediction and modeling system proposed in the present invention collects power equipment data in real time through the intelligent mimicry module to ensure timely and continuous data; the data preprocessing module cleans data, denoises, extracts features, and improves data quality; the adaptive model construction module forms a modeling branch based on the equipment type; the initial model is trained by small batch gradient descent and random optimization algorithms; the parameters are adjusted through prediction error feedback to enhance the model accuracy and generalization ability; the prediction output module outputs the prediction results of key indicators to adapt to different equipment and scales; in addition, distributed modeling and prediction are adopted for large-scale systems, and direct prediction is adopted for small-scale equipment to adjust parameters and data to meet the needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The present invention will be further described below in conjunction with the accompanying drawings:

[0055] Figure 1 This is a module framework diagram of the power equipment data prediction and modeling system of the present invention. DETAILED DESCRIPTION

[0056] like Figure 1As shown, the present invention provides a power equipment data prediction and modeling system based on intelligent mimicry, including: an intelligent mimicry data acquisition module, a data preprocessing and feature extraction module, an adaptive model construction module, a model training and optimization module, and a prediction output module;

[0057] The intelligent mimicry data acquisition module is the front-end data entry of the system and plays a key role. The module collects various key data of equipment operation in real time and continuously through sensors deployed in key parts of power equipment, including electrical parameters reflecting the characteristics of power transmission and conversion, mechanical performance indicators reflecting the operating status of mechanical structures, and operating environment data recording external environmental factors. This module is divided into a data acquisition execution unit and a cache and format regularization unit. The data acquisition execution unit is responsible for real-time collection of electrical, mechanical and environmental data according to the set frequency and method to ensure timely and continuous data acquisition. The cache and format regularization unit stores massive data and regularizes data of different sources and formats into a unified standard format, laying the foundation for subsequent processing and analysis.

[0058] The data preprocessing and feature extraction module is closely connected with the intelligent mimicry data acquisition module and is a key link in the data processing process. This module uses advanced data cleaning algorithms to remove noise and outliers in the collected data and restore the real data. At the same time, it uses principal component analysis and wavelet transform technology to mine key features. This module contains four units: data cleaning, principal component analysis, wavelet transform and feature integration. The data cleaning unit is used to identify and remove noise and outliers to make the data reliable. The principal component analysis unit is used to reduce the dimension of the cleaned data, retain the core information, and extract key features. The wavelet transform unit is used to mine local and detailed features to enhance the model's ability to capture data features. The feature integration unit is used to integrate the features extracted by principal component analysis and wavelet transform, build a complete feature set, and provide materials for model construction.

[0059] The adaptive model building module is electrically connected to the data preprocessing and feature extraction module, and is one of the core "smart centers" of the system. Based on the principle of intelligent mimicry, the neural network adaptive layer is introduced to enable the model to have self-evolution capabilities. It automatically and accurately optimizes the neuron connection mode and weight distribution based on the input feature information, forms customized modeling branches for different equipment types such as transformers, circuit breakers, generators, and builds the initial model of each device. The module contains four units: intelligent mimicry analysis, neural network construction, branch modeling, and parameter initialization. The intelligent mimicry analysis unit analyzes data features based on principles and gains insights into regular characteristics. The neural network construction unit builds a neural network architecture by introducing an adaptive layer, and the adaptive layer dynamically optimizes connections and weights. The branch modeling unit is used to study the operating characteristics and data regularities of different devices and create specific modeling branches. The parameter initialization unit reasonably sets initial parameters for the initial and branch models to ensure a good start for training.

[0060] The core of the intelligent mimicry principle is to enable the system to imitate the surrounding environment or target objects. The principle involves multiple levels, including the perception part, information processing stage and execution link.

[0061] In the perception part, environmental information is collected through a variety of sensors. For example, cameras are used to capture the color, shape, and texture of surrounding objects, microphones are used to collect sound characteristics, and various physical sensors are used to detect temperature, humidity, pressure and other data to fully perceive the environment.

[0062] During the information processing phase, the large amount of data collected is transmitted to the central processor. The algorithm quickly analyzes this data and matches it with the massive pattern library stored in the system through pattern recognition technology, thereby accurately identifying the key features of the current environment.

[0063] In the execution phase, based on the processed results, the system will control the actuator to make corresponding changes. To simulate the appearance, special display materials or structures are used to quickly adjust the color and shape to achieve a visual effect that blends in with the environment; to simulate behavior, the dynamic characteristics of the target are imitated by controlling the mechanical structure or movement mode. It is this series of coordinated operations of perception, processing and execution that makes intelligent mimicry possible, playing an important role in military camouflage, bionic robots, intelligent security and other fields.

[0064] The model training and optimization module takes over the adaptive model construction module and is a key stage for improving model performance. This module integrates small batch gradient descent and stochastic optimization algorithms, and uses newly collected data batches to train the model. It is equipped with a model prediction error feedback mechanism, and backtracks to adjust parameters when the deviation exceeds the threshold to maintain the optimal state of the model. The module includes four units: training execution, error calculation, feedback judgment, and parameter adjustment. The training execution unit uses an algorithm to train the model with new data and updates parameters to improve accuracy. The error calculation unit calculates the error based on the difference between the predicted and actual values ​​to provide a basis for adjustment. The feedback judgment unit compares the error with the threshold to decide whether to backtrack and adjust. When the deviation exceeds the threshold, the parameter adjustment unit accurately adjusts the parameters according to the backtracking strategy to optimize the model performance.

[0065] The prediction output module is closely connected with the model training and optimization module, and is the "results presentation window" of the system. Its task is to output the prediction results of key indicators of power equipment and realize the prediction and modeling of equipment data of different types and scales. The module contains three units: prediction operation, type adaptation and output display. The prediction operation unit obtains the prediction value based on the training and optimized model operation data. The type adaptation unit adjusts the prediction results according to the characteristics of different equipment. The output display unit outputs the results in an intuitive way, helping users to understand the equipment operation trends and potential problems, and provide support for equipment maintenance, management and decision-making.

[0066] In the process of intelligent and efficient development in the power sector, the power equipment data prediction and modeling system based on intelligent mimicry is composed of various key modules that work closely together to achieve powerful data processing and analysis capabilities.

[0067] As the data source, the intelligent mimicry data acquisition module collects multi-dimensional data through sensors. Electrical parameters reflect the characteristics of power transmission, mechanical performance indicators reflect the condition of the mechanical structure, and operating environment data records external factors. The two units in the module have clear division of labor. The acquisition unit is responsible for collection, and the regularization unit processes the data format.

[0068] The data preprocessing and feature extraction module uses algorithms and technologies to clean the collected data and mine features. Each unit performs its own duties to ensure data quality and feature extraction results.

[0069] The adaptive model building module builds initial models for different devices based on principles and technologies. Each unit works together to achieve model customization and reasonable parameter setting.

[0070] The model training and optimization module uses algorithms and mechanisms to train models, adjust parameters, and improve performance. Each unit collaborates to ensure that the model adapts to changes in device operation.

[0071] The prediction output module works through each unit to output the prediction results and provide decision support for users.

[0072] When the sensor collects electrical parameters, the root mean square calculation formula is used to accurately obtain the effective value of the voltage for the voltage parameter. The formula for calculating the effective value of the voltage is:

[0073]

[0074] Where: V rms Indicates the effective value of voltage, v 2 (t) represents the instantaneous value of voltage changing with time, T represents the length of a cycle, and dt represents a small increment of time t;

[0075] The current parameter uses a similar RMS calculation method, and the calculation formula is:

[0076]

[0077] Where: I rms Indicates the effective value of current, i 2 (t) represents the instantaneous value of current changing with time;

[0078] When collecting mechanical performance indicators, the time domain vibration signal is converted into a frequency domain signal through Fourier transform, where the conversion formula is:

[0079]

[0080] Where: X(f) represents the frequency domain signal, x(t) represents the frequency domain signal;

[0081] The temperature collection in the operating environment data adopts the temperature measurement principle of thermal resistors, and the relationship between the resistance value and temperature is approximately:

[0082] R t =R 0 (1+αt);

[0083] Where: R t Indicates the resistance value at temperature t, R 0 It represents the resistance value at 0℃, and α represents the temperature coefficient of resistance;

[0084] The use of root mean square to calculate the effective value of voltage and current can accurately reflect their actual working capacity, provide a key basis for power system analysis and evaluation of equipment operating status, and ensure the stable operation of power equipment. The use of Fourier transform to process vibration signals can clearly present different frequency components, help quickly locate potential equipment failure points, and prevent mechanical failures in advance. Thermal resistor temperature measurement uses the approximate relationship between resistance and temperature. It has the characteristics of high precision and good stability. It can accurately monitor the operating environment temperature of the equipment in real time, avoid the impact of abnormal temperature on equipment performance and life, and comprehensively ensure the safe and efficient operation of power equipment.

[0085] When the data cleaning unit removes noise, for the one-dimensional data sequence x 1 ,x 2 ,…,x n , let the window size be m, sort the data in the window, and take the middle value as the filtered value, where the algorithm expression is:

[0086]

[0087] Where: y i represents the value after filtering, Indicates the middle value after sorting;

[0088] The method to remove abnormal data is:

[0089] If the data meets It is judged as an outlier, is the mean, is the standard deviation;

[0090] Window sorting and median filtering can effectively smooth one-dimensional data sequences, suppress random noise interference, and retain the main characteristics of the data. The method of determining outliers based on mean and standard deviation can keenly capture data points that deviate from the normal range, provide relatively pure and accurate data for subsequent analysis, and improve data quality and the reliability of analysis results.

[0091] When the key features extracted by the feature integration unit are integrated to form a complete feature set, the data is first standardized:

[0092]

[0093] Where: represents the data after normalization. represents the mean of the jth variable, s j represents standard deviation;

[0094] Then calculate the covariance matrix:

[0095]

[0096] Where: S represents the covariance matrix;

[0097] Based on the covariance matrix, the eigenvalues ​​and eigenvectors of the covariance matrix are calculated, and the principal components are selected. The steps for the eigenvalues ​​and eigenvectors are:

[0098] Solve the characteristic equation to obtain several eigenvalues. For each eigenvalue, solve the homogeneous linear equation system to obtain the corresponding eigenvector, where the characteristic equation expression is:

[0099] |S-λI|=0;

[0100] Where: λ represents the eigenvalue to be determined, I represents the n-order unit matrix, that is, an n×n matrix with the main diagonal elements being 1 and the remaining elements being 0, and |S-λI| represents the determinant of the matrix S-λI;

[0101] Among them, the expression of the homogeneous linear equation system is:

[0102] |S-λ j I|=0;

[0103] Where: j represents the jth eigenvalue;

[0104] Standardization can eliminate the dimensionality effect between variables and make the data on the same scale for easy comparison. Calculating the covariance matrix can measure the correlation between variables. Finding eigenvalues ​​and eigenvectors and selecting principal components can reduce dimensionality while retaining key information, reduce data redundancy, improve subsequent analysis efficiency and model performance, and explore potential patterns in the data.

[0105] When wavelet transform is used for feature extraction, discrete wavelet transform is used. For discrete signal f(n), its discrete wavelet transform is:

[0106]

[0107] Where: represents the wavelet basis function, a represents the scale parameter, b represents the translation parameter, and ψ(n) represents the basic wavelet function.

[0108] The neural network building unit optimizes the neuron connection and updates the weights through the neural network adaptive layer. The formula for updating the weights is:

[0109]

[0110] Where: represents the sum of squared historical gradients, ε represents a small constant that prevents the denominator from being zero.

[0111] Using discrete wavelet transform for feature extraction can effectively capture the local features of discrete signals at different scales and locations, accurately depict signal details, and provide rich information for subsequent analysis. The adaptive layer of the neural network updates weights according to a specific formula, allowing the neural network to dynamically optimize according to data characteristics, improve the model's ability to fit and generalize complex data, and better adapt to the complex characteristics of power equipment data.

[0112] For the transformer modeling branch, consider the transformer equivalent circuit model, and the relationship between the primary side voltage and the secondary side voltage is:

[0113]

[0114] Where: V 1 Represents the primary side voltage, V 2 Represents the secondary voltage, N 1 Indicates the number of turns of the primary winding, N 2 Indicates the number of turns of the secondary winding;

[0115] According to the relationship between the primary side voltage and the secondary side voltage, the transformation ratio can be obtained as follows:

[0116]

[0117] When constructing the transformer model, the transformation ratio is used as an important characteristic parameter, and the weight is optimized through the neural network adaptive layer in combination with other electrical parameters so that the model can accurately reflect the operating status of the transformer. The electrical parameters include but are not limited to: load current and winding resistance;

[0118] For the circuit breaker modeling branch, the relationship between its breaking current capacity, contact material, and breaking time is considered, and the relationship between the breaking current, contact material coefficient, and breaking time is set as follows:

[0119]

[0120] Where: I br Indicates breaking current, k m Indicates the contact material coefficient, tbr Indicates the breaking time;

[0121] These parameters are used as model input features, and the neuron connections and weights are adjusted through the adaptive layer to build a model suitable for the circuit breaker;

[0122] For the generator modeling branch, the relationship between the electromagnetic induction principle and the mechanical power input of the generator is considered, and the relationship between the generator output power and the rotor speed and the excitation current is assumed to be:

[0123]

[0124] Where: P 发 Indicates the generator output power, C 1 represents a constant related to the generator structure, n 转 Indicates the rotor speed, I f represents the excitation current, Indicates the power factor angle;

[0125] At the same time, the mechanical power input is converted into the mechanical energy of the rotor, and the relationship is expressed as:

[0126] T 机 =C 2 n 转 ;

[0127] Where: T 机 Represents mechanical torque, C 2 represents a constant;

[0128] The rotor speed, excitation current, and mechanical torque parameters are used as model input features, and the neuron connections and weights are adjusted through the neural network adaptive layer to build a model that reflects the operating status of the generator.

[0129] For transformers, the transformation ratio is determined based on its equivalent circuit model and combined with electrical parameters. With the help of neural network adaptive layer optimization, its operating status can be accurately reflected, providing strong support for power system analysis. For circuit breakers, the relationship between breaking current and contact material and breaking time is considered, and these parameters are modeled. Through adaptive layer adjustment, a model that fits its characteristics can be effectively constructed to ensure the safe and stable operation of power equipment. For generators, the electromagnetic and mechanical characteristics of generators can be fully considered, and the accuracy of the model can be improved by accurately reflecting the relationship between output power and various parameters. With the help of neural network adaptive adjustment, the model can flexibly adapt to different working conditions. It can also integrate multiple parameters as input to provide a powerful tool for analyzing the complex operating status of generators.

[0130] When mini-batch gradient descent is combined with a stochastic optimization algorithm, mini-batch gradient descent divides the data by batch size, and its gradient calculation is:

[0131]

[0132] Where: B k represents the kth mini-batch data set, L i (θ) represents the i-th sample loss function, and B represents the batch size of the mini-batch data;

[0133] The weights are updated through a random algorithm, where the weight update formula is:

[0134]

[0135] Where: represents the sum of squared historical gradients, ε represents a small constant that prevents the denominator from being zero. During the model training process, the model gradually converges to a better solution by continuously iterating small batches of data and updating the weights according to the above formula. The random algorithm updates the weights according to a specific formula, introducing randomness to prevent the model from falling into a local optimal solution. The combination of the two, continuous iteration during model training, makes the weight update more reasonable, prompting the model to converge to a better solution quickly and stably, and improving the training effect.

[0136] In the model prediction error feedback mechanism, the deviation between the predicted value and the actual monitored value is calculated, and the mean square error is used as a measurement indicator. When the mean square error exceeds the threshold, the model parameters are adjusted retrospectively;

[0137] The retrospective adjustment method is:

[0138] Assume that the current step size is β. When the mean square error exceeds the threshold, the step size is reduced to γβ, where 0<γ<1. The gradient is recalculated and the weight is updated. The model prediction error is gradually reduced by adjusting the step size and updating the weight multiple times. Using the mean square error to measure the deviation can fully reflect the degree of deviation between the predicted value and the actual value and accurately evaluate the model performance. Backtracking adjustment when the threshold is exceeded, by reducing the step size and recalculating the gradient to update the weight, can avoid the model from missing the optimal solution due to too large a step size, effectively reduce the prediction error, and improve the accuracy and stability of the model prediction.

[0139] When deriving the predicted values ​​of the key indicators of power equipment, for the key indicators of transformers, the heat transfer principle is considered:

[0140] Assume that the transformer load loss generates heat Q loss =I 2 R, I is the load current, R is the winding resistance, and the heat is transferred to the winding to increase its temperature. According to the heat balance equation:

[0141]

[0142] Where: m represents the mass of the winding, c represents the specific heat capacity, h represents the heat dissipation coefficient, A represents the heat dissipation area, T represents the winding temperature, T 0Indicates the ambient temperature;

[0143] The model predicts the future winding temperature by learning the relationship between load current, ambient temperature and winding temperature in historical data and combining it with the above formula;

[0144] For the prediction of key indicators of circuit breakers, the model determines the contact material coefficient through training data, and then predicts the contact wear. For transformers, based on the principle of heat transfer and the heat balance equation, combined with historical data learning, it can accurately predict the winding temperature and prevent failures caused by abnormal temperature in advance. For circuit breakers, the coefficients are determined through training data to predict contact wear, which can detect potential problems in time, ensure the safe and stable operation of equipment, and improve the reliability of the power system. For generators, the model determines the output power and mechanical energy of the generator through training data, and then predicts the service life of the generator.

[0145] When realizing data prediction and modeling of power equipment of different sizes, for large-scale power systems containing multiple power equipment, a distributed modeling and prediction method is adopted:

[0146] Assume that there are several devices in the system, and build a local model for one of them;

[0147] Each local model shares some characteristic information through an information interaction mechanism, and the characteristic information includes but is not limited to: voltage and frequency in the power grid;

[0148] The overall system prediction value is obtained by integrating the prediction values ​​of each local model, using a weighted average method:

[0149]

[0150] Where: Y represents the overall system prediction value, N represents the total number of devices, and w i represents the weight, represents the predicted value of the i-th device;

[0151] The weights are determined by minimizing the overall prediction error:

[0152]

[0153] Where: MSE total represents minimizing the overall prediction error, y j Indicates actual value;

[0154] For small-scale power equipment, a single model is directly used for prediction. By adjusting the model parameters and training data, it can adapt to the data prediction and modeling needs of power equipment of different sizes. Distributed modeling is used for large-scale systems. Each local model shares feature information, and then the prediction value is fused through weighted average, which can effectively utilize system resources and improve prediction accuracy. For small-scale equipment, a single model is directly used, and parameters and data can be flexibly adjusted, taking into account efficiency and adaptability to meet the diverse needs of equipment of different sizes.

[0155] The present invention also proposes a method for predicting and modeling power equipment data based on intelligent mimicry, based on the above system, comprising the following steps:

[0156] S1: Real-time collection of electrical parameters, mechanical performance indicators and operating environment data of power equipment through intelligent mimicry data acquisition module:;

[0157] S2: Clean the data and extract features through the data preprocessing and feature extraction module;

[0158] S3: The adaptive model building module automatically builds the initial model based on the input features;

[0159] S4: The model training and optimization module uses the newly collected data to train the model and retrospectively adjusts the model parameters based on the prediction error feedback mechanism;

[0160] S5: The prediction output module outputs the prediction results of key indicators of power equipment, realizing the prediction and modeling of power equipment data of different types and scales.

[0161] In summary, the advantages of the present invention are: the system can comprehensively and accurately collect various data of power equipment, and through advanced data preprocessing and feature extraction technology, effectively remove noise and outliers, and extract key features. Its adaptive model construction module can form specific modeling branches for different equipment types, build an initial model, and continuously iterate and optimize through model training and optimization modules to improve prediction accuracy. In addition, the system can also realize the prediction and modeling of power equipment data of different types and scales, provide strong support for power equipment monitoring and intelligent operation and maintenance, and help improve the stability and safety of the power system.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power equipment data prediction and modeling system based on intelligent mimicry, characterized by: include: Intelligent mimicry data acquisition module: used to collect equipment data in real time through sensors deployed on power equipment, the equipment data includes: electrical parameters, mechanical performance indicators and operating environment data; Data preprocessing and feature extraction module: electrically connected to the intelligent mimicry data acquisition module, used to remove noise and outliers using a data cleaning algorithm, and extract key features through principal component analysis and wavelet transform; Adaptive model building module: electrically connected to the data preprocessing and feature extraction module, the adaptive model building module is used to introduce a neural network adaptive layer based on the principle of intelligent mimicry, automatically optimize neuron connections and weight distribution according to input features, form specific modeling branches for different device types, and build initial models for each branch; Model training and optimization module: electrically connected to the adaptive model building module, used to train the model in batches using the input new collected data through small batch gradient descent combined with a random optimization algorithm, and to retroactively adjust the model parameters when the deviation between the predicted value and the actual monitored value exceeds a threshold value based on the model prediction error feedback mechanism; Prediction output module: electrically connected to the model training and optimization module, used to output the prediction results of key indicators of power equipment, and realize the prediction and modeling of power equipment data of different types and sizes.

2. According to claim 1, a power equipment data prediction and modeling system based on intelligent mimicry is characterized by: The intelligent mimic data acquisition module includes a data acquisition execution unit and a cache and format regularization unit. The data acquisition execution unit is responsible for real-time acquisition of electrical, mechanical and environmental data of power equipment according to a set frequency and method; The cache and format regularization unit is used to store massive data and regularize data of different sources and formats into a unified standard format.

3. According to claim 1, a power equipment data prediction and modeling system based on intelligent mimicry is characterized by: The preprocessing and feature extraction module includes four units: data cleaning, principal component analysis, wavelet transformation and feature integration; The data cleaning unit is used to identify and remove noise and outliers; The principal component analysis unit is used to reduce the dimension of the cleaned data, retain the core information, and extract key features; The wavelet transform unit is used to mine local and detailed features and enhance the model's ability to capture data features; The feature integration unit is used to integrate the features extracted by the principal component analysis unit and the wavelet transform unit to construct a complete feature set and provide materials for model construction.

4. The power equipment data prediction and modeling system based on intelligent mimicry according to claim 1, characterized in that: The adaptive model building module includes four units: intelligent mimicry analysis, neural network construction, branch modeling and parameter initialization; The intelligent mimicry analysis unit analyzes the input data features based on the intelligent mimicry principle; The neural network building unit builds the neural network architecture by introducing the neural network adaptive layer, dynamically optimizes the neuron connection and updates the weight through the neural network adaptive layer. The formula for updating the weight is: Where: represents the weight, t represents the number of iterations, η represents the learning rate, L(θ) represents the neural network loss function, and θ represents the weight parameter set; The branch modeling unit is used to form specific modeling branches for different equipment types in combination with their characteristics, and to construct the initial model of each branch; The parameter initialization unit is used to reasonably set the initial parameters for the initial model of each branch.

5. The power equipment data prediction and modeling system based on intelligent mimicry according to claim 4 is characterized by: The model training and optimization module includes a training execution unit, an error calculation unit, a feedback judgment unit and a parameter adjustment unit; The training execution unit is used to train the initial model using newly collected data batches through mini-batch gradient descent combined with a stochastic optimization algorithm, pushing the model parameters to update in the direction of reducing the loss function; The error calculation unit is used to calculate the prediction error based on the model prediction value and the actual monitoring value; The feedback judgment unit is used to compare the calculated prediction error with the set threshold value, and decide whether to trigger the operation of backtracking adjustment of model parameters according to the comparison result; The parameter adjustment unit is used to adjust the model parameters according to the backtracking strategy when the deviation between the predicted value and the actual monitoring value exceeds the threshold; When mini-batch gradient descent is combined with a stochastic optimization algorithm, mini-batch gradient descent divides the data by batch size, and then updates the weights through a stochastic algorithm. The updated weight calculation formula is: Where: represents the sum of squared historical gradients, ε represents a small constant that prevents the denominator from being zero.

6. The power equipment data prediction and modeling system based on intelligent mimicry according to claim 5, characterized in that: In the model prediction error feedback mechanism, the deviation between the predicted value and the actual monitored value is calculated, and the mean square error is used as a measurement indicator. When the mean square error exceeds the threshold, the model parameters are adjusted retrospectively; The retrospective adjustment method is: Assume that the current step size is β. When the mean square error exceeds the threshold, the step size is reduced to γβ, where 0<γ<1. The gradient is recalculated and the weight is updated. By backtracking multiple times to adjust the step size and update the weight, the model prediction error is gradually reduced.

7. The power equipment data prediction and modeling system based on intelligent mimicry according to claim 4 is characterized by: The equipment types in the branch modeling unit include transformers, circuit breakers and generators. For the transformer modeling branch, a model that considers the transformer equivalent circuit is constructed. The transformation ratio, load circuit and winding resistance are used as model input features. The weights are optimized through the neural network adaptive layer to obtain the initial model of the transformer. The model predicts the future winding temperature by learning the relationship between load current, ambient temperature and winding temperature in historical data. For the circuit breaker modeling branch, the relationship between its breaking current capacity and contact material and breaking time is considered, and the neuron connection and weight are adjusted through the neural network adaptive layer to build the initial model of the circuit breaker; the model determines the contact material coefficient through training data, and then predicts the contact wear; For the generator modeling branch, the rotor speed, excitation current, and mechanical torque parameters are used as model input features. The neuron connections and weights are adjusted through the neural network adaptive layer to build an initial model that reflects the operating status of the generator. The model determines the output power and mechanical energy of the generator through training data, and then predicts the service life of the generator.

8. The power equipment data prediction and modeling system based on intelligent mimicry according to claim 1, characterized in that: The prediction output module includes a prediction operation unit, a type adaptation unit and an output display unit; The prediction operation unit is used to operate the input power equipment data according to the trained and optimized model to obtain the predicted value of the key indicators of the power equipment; The type adaptation unit is used to adapt the prediction results to different types of power equipment; The output display unit is used to output the sorted and adapted prediction results, realize the display of prediction results of power equipment data of different types and sizes, and provide intuitive information for users.

9. The power equipment data prediction and modeling system based on intelligent mimicry according to claim 8, characterized in that: When realizing data prediction and modeling of power equipment of different sizes, for large-scale power systems containing multiple power equipment, a distributed modeling and prediction method is adopted: Assume that there are several devices in the system, and build a local model for one of them; Each local model shares some feature information through an information exchange mechanism. The overall system prediction value is obtained by fusing the prediction values ​​of each local model, and the weight is determined by minimizing the overall prediction error. For small-scale power systems, a single constructed model is directly used for prediction. By adjusting the model parameters and training data, it can adapt to the data prediction and modeling needs of power equipment of different sizes.

10. A method for predicting and modeling power equipment data based on intelligent mimicry, characterized in that: The power equipment data prediction and modeling system based on intelligent mimicry as described in any one of claims 1 to 9 comprises the following steps: S1: Collect electrical parameters, mechanical performance indicators and operating environment data of power equipment in real time through the intelligent mimicry data acquisition module; S2: Clean the data and extract features through the data preprocessing and feature extraction module; S3: The adaptive model building module automatically builds the initial model of each branch according to the input features; S4: The model training and optimization module uses the newly collected data to train the model and retrospectively adjusts the model parameters based on the prediction error feedback mechanism; S5: The prediction output module outputs the prediction results of key indicators of power equipment, realizing the prediction and modeling of power equipment data of different types and scales.

Images