Intelligent sample generation platform for power system

Abstract

The invention discloses an intelligent sample generation platform for a power system, and the platform comprises a data collection module which is used for collecting the real-time operation data, including voltage, current, power and frequency, of the power system; the data preprocessing module is used for cleaning, denoising and normalizing the real-time operation data; the feature extraction module is used for extracting key features from the preprocessed real-time operation data by adopting a machine learning algorithm, wherein the key features comprise a time domain feature, a time-frequency domain feature and a frequency domain feature; the sample generation module is used for generating a power system sample by adopting a generative adversarial network according to the key features; the sample evaluation module is used for evaluating the power system sample; and the sample output module is used for outputting the samples conforming to the sample evaluation indexes according to the evaluation result. According to the invention, by generating samples, practitioners in the power industry can better understand system operation characteristics and potential problems.

Description

Technical Field

[0001] The present invention relates to the technical field of power system sample generation, and in particular to a sample intelligent generation platform for power systems. Background Art

[0002] With the rise of generative artificial intelligence big model technology, new solutions have been provided for the field of power system operation analysis. However, deep learning algorithms headed by big models require massive samples, and the current sample generation framework makes it difficult to intelligently realize the generation and analysis management of massive samples. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture.

[0004] A sample intelligent generation platform for power systems includes:

[0005] Data acquisition module: used to collect real-time operation data of the power system, including voltage, current, power and frequency;

[0006] Data preprocessing module: used for cleaning, denoising and normalizing the real-time operation data;

[0007] Feature extraction module: used to extract key features from pre-processed real-time running data using machine learning algorithms, including time domain features, time-frequency domain features, and frequency domain features

[0008] Sample generation module: used to generate power system samples according to the key features by using a generative adversarial network;

[0009] Sample evaluation module: used for evaluating the power system sample;

[0010] Sample output module: used to output samples that meet the sample evaluation indicators based on the evaluation results.

[0011] Preferably, the sample generation module includes:

[0012] Generative adversarial network initialization unit: used to initialize the generative adversarial network;

[0013] A sample generating unit: used for generating a power system sample by generating an adversarial network according to the key features;

[0014] Sample optimization unit: used to optimize power system samples using reinforcement learning algorithms and reward functions;

[0015] Sample compression and reconstruction unit: used to compress and reconstruct power system samples through an autoencoder;

[0016] Sample post-processing unit: used to post-process the power system samples, including sample verification, correction and formatting.

[0017] Preferably, the calculation formula of the reward function is as follows:

[0018] R=α·R rcalism +β·R diυersity +γ·R utility ,

[0019] Among them, R realism is the authenticity reward value, R diversity is the diversity bonus value, R utility is the practical reward value; α, β, γ are the weights for adjusting the importance of each reward indicator.

[0020] Preferably, the sample generating unit includes:

[0021] Key feature acquisition subunit: used to acquire key features in the feature extraction module;

[0022] A power system sample generation subunit: used for receiving the key features using a generator, and generating a power system sample using a neural network;

[0023] Sample judgment subunit: used to judge the generated samples and real samples through the discriminator and generate feedback data;

[0024] An adjustment subunit: used to adjust the parameters and reward function of the generative adversarial network according to the feedback data.

[0025] Preferably, the sample evaluation module comprises:

[0026] Sample evaluation index definition unit: used to define sample evaluation indicators including: sample diversity, authenticity and consistency;

[0027] Sample evaluation unit: used to evaluate the power system sample using a statistical analysis method according to the sample evaluation index;

[0028] Evaluation report generating unit: used to generate a sample evaluation report according to the evaluation results; the sample evaluation report includes authenticity evaluation results, diversity evaluation results and consistency evaluation results;

[0029] Feedback unit: used to feed back the sample evaluation report to the sample generation module.

[0030] Preferably, the feedback unit comprises:

[0031] Evaluation report analysis subunit: used to analyze the evaluation report;

[0032] Improvement suggestion generation unit: used to generate improvement suggestions for samples that do not meet the sample evaluation indicators based on the analysis results;

[0033] Feedback to generation module unit: used to feed back improvement suggestions and analysis results to the sample generation module, and adjust the generation adversarial network parameters and reward function through the generation module.

[0034] Preferably, the sample output module includes:

[0035] Sample output unit 161: outputs samples that meet sample evaluation indicators according to the sample evaluation report;

[0036] Sample output formatting unit: used to format samples that meet the sample evaluation indicators, including data type, unit and precision;

[0037] Sample storage unit: used to store and manage samples that meet sample evaluation indicators through a preset sample database;

[0038] Sample retrieval unit: used to provide a sample retrieval interface, allowing users to retrieve samples that meet sample evaluation indicators based on time and power system operation data.

[0039] The present invention provides an intelligent generation platform for power system samples. By adopting advanced technologies such as generative adversarial networks to generate diversified virtual samples, it can not only increase the size of the training set, but also simulate the operating states under various extreme conditions that are difficult to obtain in reality, thereby enhancing the generalization ability and robustness of the GAN model. By evaluating the generated samples and screening out high-quality data as the final output, it can effectively improve the performance of any prediction, diagnosis or optimization GAN model built based on these data. The platform provides a complete set of workflows from data collection to sample output, greatly simplifying the process of researchers developing new applications and shortening the cycle from concept verification to productization. Compared with traditional methods that require a large number of real-world tests to obtain sufficient information, the use of this platform can reduce the need for physical experiments, thereby saving costs and reducing the safety risks that may be caused by frequent testing. By analyzing the samples generated by the platform, it is beneficial for power industry practitioners to better understand the system operation characteristics and potential problems, provide a scientific basis for maintenance planning, fault prevention, etc., and assist in making more informed business decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 An architecture diagram of a sample intelligent generation platform for power systems provided by the present invention;

[0042] Figure 2 A unit structure diagram of the sample generation module provided by the present invention;

[0043] Figure 3 A subunit structure diagram of a sample generation unit in a sample generation module provided by the present invention;

[0044] Figure 4 A unit structure diagram of a sample evaluation module provided by the present invention;

[0045] Figure 5 A subunit structure diagram of the feedback unit 154 in the sample evaluation module provided by the present invention;

[0046] Figure 6 This is a unit structure diagram of the sample output module provided by the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0049] refer to Figure 1 , an intelligent generation platform for power system samples, including the following:

[0050] Data acquisition module 11: used to collect real-time operation data of the power system, including voltage, current, power and frequency;

[0051] Data preprocessing module 12: cleaning, denoising and normalizing the real-time operation data;

[0052] Feature extraction module 13: Use machine learning algorithms to extract key features from pre-processed real-time running data, including time domain features, time-frequency domain features, and frequency domain features

[0053] Sample generation module 14: Generate power system samples using a generative adversarial network based on key features;

[0054] Sample evaluation module 15: used to evaluate the power system sample;

[0055] Sample output module 16: used to output samples that meet the sample evaluation indicators according to the evaluation results.

[0056] The intelligent generation platform for power system samples provided by the present invention reduces errors or outliers through data cleaning, denoising and normalization processing; adopts advanced technologies such as generative adversarial networks to generate diversified virtual samples, which can not only increase the scale of training sets, but also simulate the operating states under various extreme conditions that are difficult to obtain in practice, thereby enhancing the generalization ability and robustness of GAN models; by evaluating the generated samples and screening out high-quality data as the final output, the performance of any prediction, diagnosis or optimization GAN model built based on these data can be effectively improved; the platform provides a complete set of workflows from data collection to sample output, greatly simplifying the process of researchers developing new applications and shortening the cycle from concept verification to productization; compared with traditional methods that require a large number of real-world tests to obtain sufficient information, the use of this platform can reduce the need for physical experiments, thereby saving costs and reducing the safety risks that may be caused by frequent testing; by analyzing the samples generated by the platform, it is beneficial for power industry practitioners to better understand the system operation characteristics and potential problems, provide a scientific basis for maintenance planning, fault prevention and other aspects, and assist in making more informed business decisions.

[0057] Specifically, in the data acquisition module 11, various sensors and power measurement devices, such as current transformers, voltage transformers and power transmitters, are used to collect data such as voltage, current and power; and these sensors are installed on transmission lines, transformers and load nodes to monitor the status of each point in real time.

[0058] In the data preprocessing module 12, for voltage or current data with missing values, linear interpolation or a method based on historical average values ​​can be used to complete the missing values. For example, when the voltage data at a certain moment is missing, the voltage values ​​at the two moments before and after can be used for linear interpolation to estimate the missing values.

[0059] Use a low-pass filter to eliminate noise from real-time operation data. When the current signal may contain high-frequency noise, a Butterworth low-pass filter can be used to denoise the signal.

[0060] When extracting features from large-scale power system data, the voltage and current data are usually normalized to the range of [0, 1], and all voltage values ​​are normalized using Min-Max to avoid difficulties in model training due to large differences in the value range.

[0061] In the feature extraction module 13, the time domain features include mean, variance, peak-to-peak value, skewness, and kurtosis; the time domain signal is converted into a frequency domain signal through fast Fourier transform, and the amplitude and phase information of the frequency component are extracted; the dual information of time and frequency is obtained through wavelet transform or short-time Fourier transform, and the spectrum distribution of the signal at different time points is captured.

[0062] refer to Figure 2 In a preferred embodiment, in the sample generation module 14, it includes

[0063] Generative adversarial network initialization unit 141: initialize the generative adversarial network;

[0064] LeakyReLU is used as the activation function in the generator and discriminator to avoid the "death" phenomenon of neurons and improve the stability of GAN model training;

[0065] The Adam optimizer is used to optimize the GAN model to ensure the efficiency and convergence speed of the training process;

[0066] Sample generating unit 142: generating power system samples through a generative adversarial network according to key features;

[0067] Sample optimization unit 143: uses a reinforcement learning algorithm and a reward function to optimize the power system samples; if the value of the reward function calculation result is less than a preset threshold, the parameters of the generator are adjusted using a reinforcement learning mechanism. Among them, the reinforcement learning algorithm, such as a deep Q network or a policy gradient method, optimizes the generated samples.

[0068] The reward function is as follows:

[0069] R=α·R realism +β·R diυersity +γ·R uIility

[0070] Among them, R realism is the authenticity reward value, R diversity is the diversity bonus value, R utility is the practical reward value; α, β, Y are the weights for adjusting the importance of each reward indicator.

[0071] Among them, calculate R realism The formula is as follows:

[0072] Rrealism =D(G(z)),

[0073] G(z) is the generated sample, D(G(z)) is the authenticity reward of the discriminator for the generated sample, and D(G(z))∈[0,1] represents the probability that the generated sample is judged as a real sample. The closer this probability is to 1, the more real the sample is.

[0074] Calculate R diversity The formula is as follows:

[0075]

[0076] N is the number of samples, d(x i , x j ) is the Euclidean distance or other similarity measure; the diversity is measured by calculating the average of the pairwise distances between these samples. The larger the average distance between samples, the higher the diversity of the samples, thus providing a higher reward to the generator.

[0077] Calculate R utility The formula is as follows:

[0078]

[0079] E gen is the error of using generated samples for the task, E aseline is the error of the baseline model. When the prediction error of the generated sample is lower than the baseline, the practical reward is higher. The value of this formula is between [0, 1], and the larger the value, the better the performance of the generated sample in the specific task.

[0080] Sample compression and reconstruction unit 144: compresses and reconstructs the generated samples using an autoencoder; the compression process reduces the dimensionality of the high-dimensional samples to reduce storage overhead while retaining key features of the samples;

[0081] During the sample reconstruction process, the mean square error is used as the reconstruction loss to ensure that the reconstructed sample maintains a high similarity with the original sample.

[0082] Sample post-processing unit 145: performs post-processing on the generated samples, including sample verification, correction and formatting.

[0083] Specifically, the rationality of the sample is checked through a data verification algorithm, including the limitations of physical constraints and electrical parameters;

[0084] Sample correction is to use rule algorithms to correct samples that do not meet the preset requirements;

[0085] Formatting is the process of converting samples into the standard format required by power system analysis tools.

[0086] refer to Figure 3 In a preferred embodiment, in the sample generation unit 142, it includes

[0087] Key feature acquisition subunit 1421: used to acquire key features in the feature extraction module;

[0088] The power system sample generation subunit 1422 is used to receive the key features using the generator and generate the power system sample using the neural network;

[0089] Sample judgment subunit 1423: used to judge the generated samples and the real samples through the discriminator, and generate feedback data;

[0090] An adjustment subunit 1424 is used to adjust the parameters of the generative adversarial network according to the feedback data of the discriminator;

[0091] The power system sample output subunit 1425 uses the generator to output the optimized power system sample.

[0092] refer to Figure 4 In a preferred embodiment, the sample evaluation module 15 comprises

[0093] Sample evaluation index definition unit 151: used to define sample evaluation indexes, including sample diversity, authenticity and consistency;

[0094] Sample evaluation unit 152: evaluates the power system sample using a statistical analysis method according to the sample evaluation index;

[0095] Evaluation report generating unit 153: generates a sample evaluation report according to the evaluation result of the sample evaluation unit;

[0096] Feedback unit 154: used to feed back the sample evaluation report to the sample generation module.

[0097] refer to Figure 5 In a preferred embodiment, the feedback unit 154 includes:

[0098] Evaluation report analysis subunit 1541: used to analyze the evaluation report; wherein the evaluation report includes authenticity evaluation results, diversity evaluation results and consistency evaluation results;

[0099] Improvement suggestion generation subunit 1542: Generate improvement suggestions based on the analysis results; for example, to address the problem of insufficient authenticity, the parameters of the generator in the generative adversarial network can be adjusted; to address the problem of insufficient diversity, the randomness in the sample generation process can be increased.

[0100] Feedback to generation module subunit 1543: Feedback improvement suggestions and analysis results to the sample generation module, and adjust the generation adversarial network parameters through the generation module.

[0101] Specifically, network parameters mainly refer to the specific parameters of the generator and discriminator in the generative adversarial network, such as learning rate, type of activation function, number of network layers, convolution kernel size, etc. Adjustment of these parameters can effectively improve the performance of the generator and make the generated samples more in line with the actual needs of the power system.

[0102] refer to Figure 6 In a preferred embodiment, the sample output module 16 includes

[0103] Sample output unit 161: outputs samples that meet sample evaluation indicators according to the evaluation report;

[0104] Sample output formatting unit 162: formats samples that meet the sample evaluation indicators, including data type, unit and precision; reduces data processing errors caused by format differences.

[0105] Sample storage unit 163: stores and manages samples that meet sample evaluation indicators through a preset sample database;

[0106] Sample retrieval unit 164: provides a sample retrieval interface, allowing users to retrieve samples that meet sample evaluation indicators based on time and power system operation data, so that users can quickly locate required information according to specific needs, thereby improving data query efficiency.

Claims

1. A sample intelligent generation platform for power systems, characterized in that: These include: Data acquisition module: used to collect real-time operation data of the power system, including voltage, current, power and frequency; Data preprocessing module: used for cleaning, denoising and normalizing the real-time operation data; Feature extraction module: used to extract key features from pre-processed real-time operation data using machine learning algorithms, including time domain features, time-frequency domain features, and frequency domain features; Sample generation module: used to generate power system samples according to the key features by using a generative adversarial network; Sample evaluation module: used for evaluating the power system sample; Sample output module: used to output samples that meet the sample evaluation indicators based on the evaluation results.

2. The sample intelligent generation platform for power systems according to claim 1, characterized in that: The sample generation module includes: Generative adversarial network initialization unit: used to initialize the generative adversarial network; A sample generating unit: used for generating a power system sample by generating an adversarial network according to the key features; Sample optimization unit: used to optimize power system samples using reinforcement learning algorithms and reward functions; Sample compression and reconstruction unit: used to compress and reconstruct power system samples through an autoencoder; Sample post-processing unit: used to post-process the power system samples, including sample verification, correction and formatting.

3. The sample intelligent generation platform for power systems according to claim 2, characterized in that: The calculation formula of the reward function is as follows: R=α·R realism +β·R diversity +γ·R utility , Among them, R realism is the authenticity reward value, R diversity is the diversity reward value, R utility is the practical reward value; α, β, γ are the weights for adjusting the importance of each reward indicator.

4. The sample intelligent generation platform for power systems according to claim 2, characterized in that: The sample generation unit includes: Key feature acquisition subunit: used to acquire key features in the feature extraction module; A power system sample generation subunit: used for receiving the key features using a generator, and generating a power system sample using a neural network; Sample judgment subunit: used to judge the generated samples and real samples through the discriminator and generate feedback data; An adjustment subunit: used to adjust the parameters and reward function of the generative adversarial network according to the feedback data.

5. The sample intelligent generation platform for power systems according to claim 4, characterized in that: The sample evaluation module includes: Sample evaluation index definition unit: used to define sample evaluation indicators including: sample diversity, authenticity and consistency; Sample evaluation unit: used to evaluate the power system sample using a statistical analysis method according to the sample evaluation index; An evaluation report generating unit is used to generate a sample evaluation report according to the evaluation results, wherein the sample evaluation report includes an authenticity evaluation result, a diversity evaluation result and a consistency evaluation result; Feedback unit: used to feed back the sample evaluation report to the sample generation module.

6. The sample intelligent generation platform for power systems according to claim 5, characterized in that: The feedback unit comprises: Evaluation report analysis subunit: used for analyzing the sample evaluation report; Improvement suggestion generation subunit: used to generate improvement suggestions for samples that do not meet the sample evaluation indicators based on the analysis results; Feedback to the generation module subunit: used to feed back improvement suggestions and analysis results to the sample generation module, and adjust the generation adversarial network parameters and reward function through the generation module.

7. The sample intelligent generation platform for power systems according to claim 6, characterized in that: The sample output module comprises: Sample output unit 161: outputs samples that meet sample evaluation indicators according to the sample evaluation report; Sample output formatting unit: used to format samples that meet sample evaluation indicators, including: data type, unit and precision; Sample storage unit: used to store and manage samples that meet sample evaluation indicators through a preset sample database; Sample retrieval unit: used to provide a sample retrieval interface, allowing users to retrieve samples that meet sample evaluation indicators based on time and power system operation data.

Images