Multi-physical-quantity data alignment method for power system

Abstract

The invention discloses a power system multi-physical-quantity data alignment method, and relates to the technical field of power system data alignment, and the method comprises the following steps: collecting multi-physical-quantity data in real time through a plurality of sensors and devices in a power system, and calibrating the original data according to the types and installation positions of the sensors, and all data sources are ensured to provide accurate time sequence information. By calibrating and synchronizing sensor data, time sequence consistency of multi-physical-quantity data is ensured, and errors caused by equipment difference and transmission delay are eliminated. Abnormal values are automatically detected and corrected by adopting a deep learning algorithm, the data quality is improved, and manual intervention is reduced. Meanwhile, the generative adversarial network is used for synthesizing and augmenting data, the data diversity and coverage are enhanced, the generalization ability and prediction precision of the large electric power model are improved, especially in a complex scene, the overfitting problem is avoided, and the accuracy and high efficiency of the model in practical application are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power system data alignment, and in particular to a method for aligning multi-physical quantity data of a power system. Background Art

[0002] The alignment of multi-physical quantity data in power systems refers to the synchronization and coordination of multiple physical quantity data (such as voltage, current, frequency, power, etc.) from different sources in the power system in terms of time and space. These data often come from different sensors, devices or monitoring systems. Due to the different sampling frequencies, timestamps or data transmission delays of sensors, they have time deviations. The purpose of data alignment is to adjust these data to a unified time reference through algorithmic means to ensure that they can reflect the status of the power system at the same time, facilitate comprehensive analysis, monitoring and decision-making, especially in scenarios such as fault diagnosis and predictive maintenance, to improve the accuracy and reliability of data analysis.

[0003] Since the physical quantity data in the power system comes from different sensors and devices, these data may have sampling bias, transmission delay or timestamp errors. In the process of data alignment, if there is no sufficiently accurate time synchronization mechanism, the data of each physical quantity may be inconsistent, which will introduce errors when building a large power model. Especially in large-scale power systems, if errors are not discovered or corrected in time, it may lead to errors in the judgment of the system operation status, which in turn may cause load scheduling errors, equipment failure prediction failures or misjudgment of catastrophic events. For example, incorrect frequency, current or voltage data alignment may cause the system to react slowly to impending failures, or even miss critical safety warnings, thereby causing large-scale power outages, equipment damage or power supply interruptions, and ultimately have a significant impact on the social economy and people's livelihood.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0005] The purpose of the present invention is to provide a method for aligning multi-physical quantity data in an electric power system, which ensures the consistency of multi-physical quantity data in timing by calibrating and synchronizing the sensor data of the electric power system, and eliminates errors caused by equipment differences, transmission delays or clock drifts. A deep learning algorithm is used to automatically detect and correct outliers, improve data quality and reduce manual intervention. Data synthesis and augmentation are performed through generative adversarial networks (GANs), the diversity and coverage of data are increased, and the generalization ability of large electric power models is enhanced, especially for scarce extreme case data, to improve the prediction accuracy of the model in complex scenarios, avoid overfitting problems, and ensure the efficiency and accuracy of the model in practical applications, so as to solve the problems in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a method for aligning multi-physical quantity data of a power system, comprising the following steps:

[0007] Multiple sensors and devices in the power system collect data on multiple physical quantities, such as voltage, current, and frequency, in real time, and calibrate the raw data according to the sensor type and installation location to ensure that all data sources can provide accurate timing information;

[0008] The time synchronization method based on the Kalman filter algorithm is used to accurately align the data collected by different sensors. The timestamp is dynamically estimated through the Kalman filter to correct the time error caused by transmission delay or system clock inconsistency in real time to ensure the consistency of data source time.

[0009] Preprocess the collected multi-physical quantity data, including noise filtering, outlier detection and elimination, etc. Use the outlier detection algorithm based on deep learning to automatically identify and correct abnormal data in the sampling process to ensure the high quality of input data;

[0010] A deep data alignment algorithm based on convolutional neural network (CNN) is used to perform nonlinear feature extraction and alignment processing on data with different time intervals and sampling frequencies. The deep learning model automatically learns the temporal and spatial associations between different physical quantities and optimizes the data alignment results, so that the aligned multi-physical quantity data has high consistency and accuracy in time series.

[0011] The high-quality data set is synthesized and augmented through the Generative Adversarial Network (GAN) algorithm to generate new training samples to enhance the sample diversity of the large power model and ensure that the power system model has stronger generalization ability and prediction accuracy in practical applications. The synthetic data can be further used for model pre-training and performance evaluation to improve the overall stability and security of the system.

[0012] Preferably, multiple sensors and devices in the power system collect multi-physical quantity data such as voltage, current, frequency, etc. in real time, and calibrate the original data according to the sensor type and installation location to ensure that all data sources can provide accurate timing information. The specific steps are as follows:

[0013] First, multiple sensors and devices in the power system begin to collect multi-physical quantity data such as voltage, current, and frequency in real time. These sensors include but are not limited to current transformers, voltage transformers, frequency sensors, temperature sensors, etc., which are distributed in different locations of the power system, such as substations, distribution networks, power plants, etc. In order to ensure the accuracy and consistency of the data, it is first necessary to ensure that the clocks of each sensor are synchronized. Different sensors may have different timestamps or different sampling time intervals, and the clocks of all devices must be calibrated to a unified standard through a synchronization mechanism. For example, the Network Time Protocol (NTP) or the Precision Synchronization Protocol (PTP) can be used to calibrate the clocks of the sensors to ensure that the timestamps of all devices can be aligned, thereby ensuring that when the subsequent data is aligned, no deviation is introduced due to time errors.

[0014] During the data collection process, the type and installation location of each sensor are crucial to the accuracy and validity of the data. Different sensors have different measurement accuracy, measurement range, and response characteristics, so they need to be calibrated according to the specifications of the sensor. For example, current transformers (CT) and voltage transformers (VT) may have differences in conversion ratio, accuracy level, installation location, etc., which may affect the accuracy of the measurement results. Therefore, when collecting data, the raw data needs to be corrected according to the type and installation location of the equipment. The calibration process involves comparing the sensor with a known standard value, obtaining the calibration coefficient of the sensor output, and adjusting the data based on these coefficients. For example, if the installation location of the sensor introduces a phase error, it can be corrected through a mathematical model to ensure that the phase relationship between voltage and current can correctly reflect the actual state of the power system.

[0015] After the data calibration is completed, the next step is to verify the quality of the collected multi-physical quantity data. The core goal of this process is to ensure the accuracy and reliability of the data and to detect potential anomalies or errors in a timely manner. The outlier detection algorithm is usually based on statistical methods or machine learning techniques, combined with the physical laws of the power system and historical data for detection. For example, by calculating the deviation of each set of data from the historical normal value, it is determined whether the data is abnormal. If it is found that the data of a certain sensor does not match the expected physical quantity range, it may be caused by sensor failure, transmission delay or external interference. For abnormal data, the usual measures are to repair the data through interpolation or prediction models, or to directly eliminate invalid data. In addition, periodic health checks and calibrations can be performed on sensors to ensure that the data quality can continue to be maintained within a controllable range during the long-term operation of data acquisition.

[0016] Data labeling and time series association are key steps to ensure that multi-physical quantity data can be accurately aligned. The labeling process involves labeling the collected multi-physical quantity data and associating each set of data with the corresponding power system status (such as load, equipment status, fault information, etc.). This process usually relies on real-time data from the power system monitoring center and on-site feedback from operation and maintenance personnel. Labeling not only involves aligning the timestamp of the data, but also requires semantic labeling of the data according to the working status of the equipment (such as equipment start and stop, load fluctuation, etc.). By establishing a standardized data structure, it is ensured that each set of data can be accurately matched with the real-time operating status of the power system. For example, if a device failure or a drastic load fluctuation is found during the data collection process, the system can automatically label the data and associate it with the time of the event, providing valuable input data for subsequent fault diagnosis and prediction. At the same time, it is also necessary to ensure the accuracy and consistency of data labeling to avoid the accumulation of errors during data alignment due to labeling errors, thereby affecting the training and prediction accuracy of the entire power system model.

[0017] Preferably, a time synchronization method based on a Kalman filter algorithm is used to accurately align data collected by different sensors. The specific steps are as follows:

[0018] First, the data from different sensors in the power system are pre-processed, including data cleaning and standardization. Different sensors may have different errors or deviations in timestamps due to differences in sampling frequency, transmission delay, etc. In order to effectively align these data, it is necessary to first make a preliminary estimate of the timestamp of each data source. This process mainly includes reading the raw data of the sensor, obtaining the timestamp and marking its source. Since the sensor may have clock offset or transmission delay, the system will record the sampling time of each sensor and use it as a preliminary timestamp reference. At this time, the Kalman filter algorithm is not directly involved, and the preliminary timestamp estimation mainly provides the initial value for the subsequent filtering process. This step provides basic data for the dynamic correction of the subsequent Kalman filter algorithm.

[0019] After data preprocessing is completed, the next step is to establish a time synchronization model based on Kalman filtering. The Kalman filter algorithm estimates the optimal time synchronization value by combining the system state and measurement data. First, define the state variables of the system, including the offset of the sensor clock and the transmission delay. Assume that the time offset of the sensor is a dynamically changing variable, and the transmission delay can be modeled by network delay or other factors. The dynamic model of the system is set according to the time relationship between sensors and the actual data flow pattern in the power system. The Kalman filter uses a state space model, regards the timestamp error of each sensor as part of the system state, and describes its change process through state transfer equations and observation equations. In this way, the Kalman filter can estimate the time synchronization deviation between each sensor in real time.

[0020] After the Kalman filter model is established, the Kalman filter algorithm is executed in real time to dynamically estimate the timestamp. Whenever new sensor data arrives, the Kalman filter will be corrected according to the current predicted state and actual measurement results. The Kalman filter is divided into a prediction step and an update step. In the prediction step, the time synchronization deviation at the current moment is predicted based on the state estimate at the previous moment; in the update step, the time error at the current moment is calculated by comparing it with the timestamp actually measured by the sensor, and the estimated value is corrected according to the Kalman gain. In this way, the Kalman filter can effectively handle problems such as system clock offset, transmission delay, noise interference, etc., and adjust the timestamp of each sensor in real time to ensure the synchronization and consistency of data from different sensors.

[0021] After completing the real-time timestamp correction of the Kalman filter, the synchronization results need to be verified and the errors corrected. At this point, the system will once again check the corrected timestamp of each sensor to ensure that all data has been accurately aligned. In practical applications, due to the complexity of the power system, some sensors may still have large synchronization errors. At this time, it is necessary to iteratively optimize the correction process. For example, the accuracy of the synchronization results can be verified by comparing with the data of other high-precision sensors, or redundant data sources can be introduced into the system for cross-validation. For situations with large errors, the parameters in the Kalman filter (such as the process noise covariance matrix and the observation noise covariance matrix) can be adjusted to improve the correction accuracy. After multiple corrections and verifications, it is finally ensured that the data of all sensors in the power system can be accurately aligned under the same time reference, providing reliable data support for subsequent large power model training, fault prediction and load scheduling.

[0022] Preferably, an outlier detection algorithm based on deep learning is used to automatically identify and correct abnormal data during the sampling process. The specific steps are as follows:

[0023] Before using the outlier detection algorithm based on deep learning, it is necessary to preprocess and extract features from the multi-physical quantity data collected in the power system. Data preprocessing includes operations such as cleaning the original data, removing noise, and filling missing values ​​to ensure the integrity and consistency of the data. Since the data in the power system usually has time series characteristics and complex physical laws, it is necessary to extract features suitable for deep learning models from the original data. For example, by normalizing multi-physical quantity data such as voltage, current, and frequency, the values ​​of different dimensions can be converted into a relatively uniform scale. In addition, by performing sliding window processing on the time series data, the time domain features and frequency domain features of the data, such as mean, variance, maximum value, minimum value, frequency spectrum, etc., are extracted. These features will help the deep learning model better identify anomalies in the data.

[0024] After data preprocessing and feature extraction, it is necessary to design and train a deep learning model suitable for outlier detection. Common deep learning models include autoencoders, long short-term memory networks (LSTM), convolutional neural networks (CNN), etc. Specifically, autoencoders are an unsupervised learning method that can reconstruct input data by training the network and determine whether the data is abnormal based on the reconstruction error. The goal of the training process is to allow the model to learn the normal pattern of the data by minimizing the reconstruction error, so that abnormal data that deviates from the normal pattern can be identified in practical applications. During the training process, a large amount of historical data is used to train the model so that the model can adapt to the common change patterns in the power system and distinguish between normal and abnormal data on this basis. During the training process, the network structure and hyperparameters need to be continuously adjusted to improve the detection ability and generalization performance of the model.

[0025] After the deep learning model training is completed, it enters the practical application stage. In this stage, the model will calculate the reconstruction error or prediction error of each data according to the input real-time data, and compare it with the pre-set threshold. If the reconstruction error exceeds the threshold, it means that the data may be an outlier. In order to improve the accuracy of outlier detection, the output results of multiple deep learning models can be combined for integrated discrimination. For example, an autoencoder can be used in combination with an LSTM network to detect anomalies in time series data, thereby making full use of time dependency and nonlinear characteristics of data. The model can also be updated and adjusted online by continuously learning new data patterns to ensure strong adaptability to complex and changeable operating conditions in the power system. Through the deep learning algorithm, the model can automatically identify abnormal data in the sampling process, such as sudden sensor failures, instantaneous high-frequency noise, etc.

[0026] In the process of outlier detection, in addition to identifying abnormal data, it is also necessary to correct the abnormal data to ensure the high quality of the final data. The correction method can be implemented by interpolation, smoothing or correction based on model prediction values. Common correction methods include using interpolation algorithms (such as linear interpolation, spline interpolation, etc.) to fill the abnormal value interval, or using prediction models (such as regression models or deep learning models) to correct abnormal data. In the process of abnormal data correction, it is also necessary to combine the physical characteristics of the power system for correction, such as considering the upper and lower limits of physical quantities such as voltage and current and the physical relationship between them, to ensure that the corrected data conforms to the physical laws of the actual power system. In addition, the establishment of a feedback mechanism is also very critical. By monitoring the effect of abnormal data correction, the effectiveness and accuracy of the correction method can be evaluated in real time, and the outlier detection and correction process can be continuously optimized based on system feedback. This feedback mechanism can enable the system to maintain high standards for data quality in long-term operation, ensure that the data is continuously accurate and can provide reliable input for subsequent analysis and modeling.

[0027] Preferably, the high-quality data set is synthesized and augmented by a generative adversarial network (GAN) algorithm to generate new training samples. The specific steps are as follows:

[0028] Before using generative adversarial networks (GANs) for data synthesis and augmentation, it is necessary to prepare and preprocess existing high-quality data sets. The purpose of data preprocessing is to ensure the quality and consistency of the input data and convert it into a format suitable for GAN training. For multi-physical quantity data in power systems, such as voltage, current, frequency, etc., the data must first be cleaned to remove noise, fill missing values, and unify dimensions. Next, the data can be standardized or normalized to eliminate the dimensional differences between different physical quantities and ensure that the input data will not affect the learning effect of the model due to different scales during training. In order to adapt to GAN training, time series data also needs to be converted into an input form suitable for deep learning models. Usually, the sliding window technology is used to convert continuous time series data into fixed-size input samples so that the model can capture the temporal characteristics and change patterns in the data.

[0029] Once the data preprocessing is complete, we can start building the Generative Adversarial Network (GAN) model. GAN consists of two main parts: the generator and the discriminator. The role of the generator is to generate samples that are as realistic as possible based on random noise (usually a high-dimensional vector), while the role of the discriminator is to determine whether the input sample comes from a real data set. Through adversarial training, the generator and the discriminator constantly compete with each other, and eventually the generator can generate samples that are very similar to the real data. Specifically, in the synthesis of power system data, the generator will generate new samples based on the statistical characteristics of the existing data (such as the amplitude of voltage and current, frequency fluctuations, etc.); while the discriminator will train the generator to improve its output by analyzing the differences between the generated data and the real data. The training process of GAN is very critical, and it is usually necessary to continuously iterate and optimize between the generator and the discriminator to ensure that the quality of the generated data meets the expectations.

[0030] Through the trained GAN model, the generator can generate new training samples based on the existing power system data. Since the data generated during the GAN training process may be noisy or biased, the generated data needs to be verified to ensure that it meets the physical laws and characteristics of the power system. For example, in the process of generating voltage and current, the generated data should follow a certain amplitude range and phase relationship, and conform to the common electrical parameters in the power system. The new data generated by GAN can not only increase the number of samples, but also increase the diversity of samples, helping the power system model to better adapt to different operating environments and fault conditions. For example, the robustness and generalization ability of the model can be improved by generating power system data under different load conditions, or generating power system data under different meteorological conditions. This data augmentation method can significantly expand the coverage of the training data set, thereby improving the training effect of large power models.

[0031] After generating new training samples, the generated data needs to be strictly verified and quality controlled. The verification process includes two aspects: first, verifying the physical consistency of the generated data to ensure that the generated data is reasonable in the power system. For example, it is possible to check whether the generated voltage and current data conform to the electrical laws of the power system such as the phase relationship and power factor; second, by comparing with the real data, ensure that the generated data is close to the real data in terms of statistical distribution, fluctuation characteristics, etc. If the generated data can meet the requirements in both aspects, it can be used for subsequent model training and testing. In terms of quality control, some indicators such as data diversity, boundary consistency, and feature retention can be used to evaluate the quality of the generated data. In addition, expert verification or automated quality inspection processes can be introduced to further ensure the availability and validity of the generated data. Finally, the generated data samples will be combined with the original data to expand and enhance the training set of the power system model and improve the accuracy and robustness of the model.

[0032] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0033] The present invention can ensure the accurate consistency of timing information between different data sources by calibrating and synchronizing multiple sensors in the power system. Sensor calibration eliminates the errors caused by differences in equipment installation location and type, and ensures the accuracy of the collected voltage, current, frequency and other multi-physical quantity data. Through the time synchronization mechanism, the time deviation caused by factors such as transmission delay and system clock drift is further eliminated, effectively avoiding the problem of inconsistent data timestamps, and ensuring the accurate integration of data collected by different sensors in the large power model.

[0034] The present invention adopts an outlier detection algorithm based on deep learning, which can automatically identify and correct abnormal data in the data collection process. By analyzing the time series characteristics of multi-physical quantity data through a deep learning model, it can accurately identify the noise and deviation in the data, ensuring that only high-quality data that conforms to physical laws enters the training and prediction process of the power model. The automatic detection and correction of outliers not only reduces the need for manual intervention, but also improves processing speed and accuracy.

[0035] The present invention synthesizes and augments the data set through generative adversarial networks, which can effectively improve the diversity and coverage of training data, thereby enhancing the generalization ability of large power models. In the power system, many abnormal or extreme situations (such as voltage fluctuations caused by drastic changes in wind speed) are relatively scarce in real data, which may lead to the model's weak prediction ability for these situations. The GAN algorithm can generate power data under different environments and different load conditions, fill these data gaps, and ensure that the model can learn more complex power system operation characteristics. It not only helps to improve the model's adaptability to different scenarios, but also effectively avoids the overfitting problem caused by insufficient or unbalanced data, so that the large power model has stronger prediction accuracy and wider applicability in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0037] Figure 1 The present invention is a method flow chart of a method for aligning multi-physical quantity data of a power system. DETAILED DESCRIPTION

[0038] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0039] The present invention provides Figure 1 A method for aligning multi-physical quantity data of a power system is shown, comprising the following steps:

[0040] Multiple sensors and devices in the power system collect data on multiple physical quantities, such as voltage, current, and frequency, in real time, and calibrate the raw data according to the sensor type and installation location to ensure that all data sources can provide accurate timing information;

[0041] The time synchronization method based on the Kalman filter algorithm is used to accurately align the data collected by different sensors. The timestamp is dynamically estimated through the Kalman filter to correct the time error caused by transmission delay or system clock inconsistency in real time to ensure the consistency of data source time.

[0042] Preprocess the collected multi-physical quantity data, including noise filtering, outlier detection and elimination, etc. Use the outlier detection algorithm based on deep learning to automatically identify and correct abnormal data in the sampling process to ensure the high quality of input data;

[0043] A deep data alignment algorithm based on convolutional neural network (CNN) is used to perform nonlinear feature extraction and alignment processing on data with different time intervals and sampling frequencies. The deep learning model automatically learns the temporal and spatial associations between different physical quantities and optimizes the data alignment results, so that the aligned multi-physical quantity data has high consistency and accuracy in time series.

[0044] The high-quality data set is synthesized and augmented through the Generative Adversarial Network (GAN) algorithm to generate new training samples to enhance the sample diversity of the large power model and ensure that the power system model has stronger generalization ability and prediction accuracy in practical applications. The synthetic data can be further used for model pre-training and performance evaluation to improve the overall stability and security of the system.

[0045] Implementation 1: Real-time multi-physical quantity data acquisition and processing based on sensor calibration and data synchronization

[0046] In the power system, multiple sensors and devices are used to collect multi-physical quantity data in real time, such as voltage, current, frequency, etc. Since these sensors may use different technologies, installation locations and measurement ranges, the timing information between data sources may be inconsistent. Therefore, in order to ensure that the data of all sensors can be efficiently and accurately integrated into the large model of the power system, each sensor needs to be calibrated first.

[0047] The core goal of sensor calibration is to ensure that all sensors can provide accurate and consistent timing information. First, the position of each sensor in space, especially in the three-dimensional coordinate system, is confirmed through accurate measurement of the installation position. Since the installation position and spatial layout of different sensors may affect the measurement results, spatial calibration is required to eliminate the errors caused by installation deviations. For example, the relative position error between the voltage sensor and the current sensor may cause the signals they measure to deviate in time, thus affecting the timing alignment of the data.

[0048] Secondly, the sensor type is identified and calibrated. Different types of sensors (such as voltage sensors, current sensors, and frequency sensors) may have different response times and sampling frequencies. In order to ensure the accuracy and consistency of the data, the time deviation and measurement accuracy of each sensor must be determined through experimental calibration and theoretical analysis. Common calibration methods include calibration using a standard signal source, a signal generator, or through known physical laws. During the calibration process, the sensor will be tested and calibrated one-to-one to ensure that the timing of its output data is consistent with the standard reference signal.

[0049] In order to further ensure that the timing information provided by all sensors can be correctly synchronized during the real-time data collection process, a unified clock system needs to be established. Different devices and sensors in the power system often run on their own local clocks, which may have different drifts or deviations. During the data collection process, a centralized clock synchronization mechanism can ensure that the data of each sensor can be aligned on the time axis. This is usually achieved through the Network Time Protocol (NTP), IEEE 1588 Precision Time Protocol (PTP), etc., to ensure that the data of all sensors is recorded under a unified timestamp.

[0050] In actual operation, by calibrating and synchronizing the sensors, it is possible to ensure that the data collected from different sensors have the same timing information. This is the basis for power system data analysis. Only through accurate data collection and synchronization can the subsequent data processing and analysis ensure the reliability and accuracy of the results. After this stage of processing, the collected data will enter the subsequent process of time synchronization and data alignment to ensure that multi-physical quantity data can be consistently and accurately represented in different time periods and different sampling frequencies.

[0051] Implementation method 2: time synchronization and error correction method based on Kalman filtering;

[0052] In power systems, due to factors such as sensor distribution and communication delays, the data collected by multiple sensors often have problems such as inconsistent timestamps, transmission delays, or clock drift. In order to eliminate these errors and ensure that the data from different sensors can be accurately aligned, a time synchronization method based on the Kalman filter algorithm can be used.

[0053] Kalman filtering is a dynamic filtering algorithm based on state estimation. It can dynamically correct the time deviation in sensor data by estimating the system state. In power systems, the input of the Kalman filter is usually the timestamps collected by the sensors and the time difference between them. First, the Kalman filter makes a preliminary estimate of time synchronization based on the initial timestamps of the sensors and the known system model. Due to factors such as transmission delays and system clock drift, these preliminary estimated timestamps may deviate from the actual values, so they need to be corrected through Kalman filtering.

[0054] Specifically, the Kalman filter calculates the error between the predicted timestamp and the actual measured value, and dynamically adjusts the sensor's timestamp based on the error. When sensor data is collected, the Kalman filter uses its current estimated timestamp to compare with the actual measured timestamp and updates the filter state to correct any inconsistencies. Every time new data arrives, the Kalman filter re-estimates the timestamp based on the updated state, making the time correction process dynamic at every moment.

[0055] The advantage of Kalman filtering is that it can accurately estimate the system state in the presence of noise. Sensor data in power systems usually contain noise, such as signal attenuation, clock drift, etc. Kalman filtering can accurately estimate time synchronization errors in these noisy backgrounds. The filter can not only correct the data timestamp of each sensor, but also further improve the overall time synchronization accuracy of the system by jointly optimizing multi-sensor data.

[0056] In addition, the real-time nature of the Kalman filter enables it to adapt to the dynamically changing environment in the power system. In the power system, the location and state of the sensor may change over time. The Kalman filter can continuously adjust the system's clock synchronization state and correct the time error between sensors in real time. This flexible dynamic correction capability is the key to ensuring that multiple sensor data in the power system can be accurately aligned over a long period of time.

[0057] By adopting the Kalman filter algorithm for time synchronization, multi-sensor data in the power system can be effectively aligned within a precise time range, thereby providing high-quality input data for subsequent data analysis, modeling, and prediction.

[0058] Implementation method 3: Data enhancement and optimization processing based on deep learning and GAN;

[0059] In the power system, especially in the modeling and prediction process of large-scale systems, the quality and diversity of data have an important impact on the accuracy and generalization ability of the model. However, due to the rarity of power system failures or the lack of certain special cases, the training data may have problems with insufficient samples or scenario bias. Therefore, the use of deep learning-based outlier detection algorithms and generative adversarial networks (GANs) to synthesize and augment data can effectively improve the quality and sample diversity of the dataset.

[0060] First, in the data preprocessing stage, the outlier detection algorithm based on deep learning can automatically detect and correct abnormal data in the sampling process. In order to ensure the high quality of the input data of the power system model, the detection and elimination of outliers are crucial. The use of deep learning models, such as the anomaly detection model based on convolutional neural network (CNN), can automatically identify abnormal data that does not conform to physical laws by analyzing the time series patterns of multi-physical quantity data. For example, if the current waveform has sudden irregular fluctuations, it may be caused by sampling noise or sensor failure. The deep learning model can capture these abnormal patterns and correct them, thereby improving the quality of the data.

[0061] Secondly, in terms of data synthesis and augmentation, the Generative Adversarial Network (GAN) can generate synthetic data with physical consistency to fill in the gaps in the training data. GAN consists of a generator and a discriminator. The generator generates new samples based on the existing data pattern, while the discriminator determines whether the generated data is realistic enough. Through adversarial training, the generator can continuously optimize the samples it generates so that they are very similar to the real data in terms of statistical characteristics. In the process of power system data synthesis, GAN can generate power data under different loads and different climatic conditions, enriching the diversity of training samples. For example, under weather conditions with large changes in wind speed, different voltage fluctuation patterns may occur in the power system. GAN can generate data in such scenarios to enhance the adaptability of the model.

[0062] The new samples generated in this way can not only improve the diversity of the data set, but also help the power system model better adapt to changing environments and extreme situations. Ultimately, the generated augmented data will be used together with the original data to train the deep learning model, so that the model can learn more comprehensive and extensive power system characteristics, thereby improving its prediction accuracy and robustness.

[0063] During the data verification phase, in order to ensure that the generated synthetic data is consistent with the real data, in addition to using physical constraints for inspection, expert verification or automated evaluation can also be performed to ensure that the synthetic data can effectively improve the training quality of the model.

[0064] The present invention can ensure the accurate consistency of timing information between different data sources by calibrating and synchronizing multiple sensors in the power system. Specifically, sensor calibration eliminates the errors caused by differences in equipment installation location and type, and ensures the accuracy of the collected voltage, current, frequency and other multi-physical quantity data. Through the time synchronization mechanism, especially the Kalman filter algorithm, the time deviation caused by factors such as transmission delay and system clock drift is further eliminated. This process effectively avoids the problem of inconsistent data timestamps and ensures the accurate integration of data collected by different sensors in the large power model. For real-time monitoring and decision analysis in power systems, this accurate consistency provides high-quality input data for subsequent fault diagnosis, load forecasting and other tasks, significantly improving the stability and safety of system operation.

[0065] The present invention adopts an outlier detection algorithm based on deep learning, which can automatically identify and correct abnormal data in the data collection process. By analyzing the time series characteristics of multi-physical quantity data through a deep learning model (such as a convolutional neural network), the noise and deviation in the data can be accurately identified, ensuring that only high-quality data that conforms to physical laws enters the training and prediction process of the large power model. The automatic detection and correction of outliers not only reduces the need for manual intervention, but also improves processing speed and accuracy. For example, in the power system, some sudden current fluctuations or voltage instability may be caused by sensor failures or data collection errors. The deep learning algorithm can capture and correct these outliers in time, thereby avoiding misjudgments caused by abnormal data. This technology improves the reliability of data and provides a more solid foundation for the safe operation of the power system.

[0066] The present invention synthesizes and augments the data set through generative adversarial networks (GANs), which can effectively improve the diversity and coverage of training data, thereby enhancing the generalization ability of large power models. In power systems, many abnormal or extreme situations (such as voltage fluctuations caused by drastic changes in wind speed) are relatively scarce in real data, which may lead to weak prediction capabilities of the model for these situations. The GAN algorithm can generate power data under different environments and different load conditions, fill these data gaps, and ensure that the model can learn more complex power system operation characteristics. This data enhancement technology not only helps improve the model's adaptability to different scenarios, but also effectively avoids the overfitting problem caused by insufficient or unbalanced data, so that the large power model has stronger prediction accuracy and wider applicability in practical applications.

[0067] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for aligning multi-physical quantity data in a power system, characterized in that: The following steps are involved: Collect multi-physical quantity data in real time through multiple sensors and devices in the power system, and calibrate the raw data according to the sensor type and installation location to ensure that all data sources provide accurate timing information; A time synchronization method based on the Kalman filter algorithm is used to accurately align data collected by different sensors. The timestamp is dynamically estimated through the Kalman filter to correct the time error caused by transmission delay or inconsistent system clocks in real time. Preprocess the collected multi-physical quantity data and use the outlier detection algorithm based on deep learning to automatically identify and correct abnormal data in the sampling process; A deep data alignment algorithm based on convolutional neural networks is used to perform nonlinear feature extraction and alignment processing on data with different time intervals and sampling frequencies. The deep learning model automatically learns the temporal and spatial associations between different physical quantities and optimizes the data alignment results. By generating adversarial network algorithms, high-quality data sets are synthesized and augmented to generate new training samples, thereby enhancing the sample diversity of large power models and ensuring that power system models have stronger generalization capabilities and prediction accuracy in practical applications.

2. A method for aligning multi-physical quantity data in a power system according to claim 1, characterized in that: Multiple sensors and devices in the power system are used to collect multi-physical quantity data in real time, and the raw data is calibrated according to the sensor type and installation location. The specific steps are as follows: First, multiple sensors and devices in the power system start to collect multi-physical quantity data in real time, and calibrate the clocks of all devices to a unified standard through a synchronization mechanism to ensure that the timestamps of all devices can be aligned, thereby ensuring the alignment of subsequent data; When collecting data, the raw data is corrected according to the type and installation location of the equipment. The calibration process includes comparing the sensor with a known standard value, obtaining the calibration coefficients of the sensor output, and adjusting the data according to these coefficients; After data calibration is completed, the quality of the collected multi-physical quantity data is verified to ensure the accuracy and reliability of the data, and to promptly detect potential anomalies or errors. For abnormal data, the data is repaired through interpolation or prediction models, or invalid data is directly eliminated; The collected multi-physical quantity data are labeled and each set of data is associated with the corresponding power system status.

3. A method for aligning multi-physical quantity data in a power system according to claim 1, characterized in that: The time synchronization method based on the Kalman filter algorithm is used to accurately align the data collected by different sensors. The specific steps are as follows: First, preliminary preprocessing is performed on the data from different sensors in the power system, including data cleaning and standardization, and preliminary estimation of the timestamp of each data source, including reading the raw data from the sensor, obtaining the timestamp and marking its source; After data preprocessing is completed, a time synchronization model based on Kalman filtering is established. The Kalman filtering algorithm estimates the optimal time synchronization value by combining the system state and measurement data; By establishing the Kalman filter model, the Kalman filter algorithm is executed in real time to dynamically estimate the timestamp. Whenever new sensor data arrives, the Kalman filter will be corrected according to the current prediction state and actual measurement results; After the real-time timestamp correction of the Kalman filter is completed, the verification and error correction of the synchronization results are performed. At this time, the corrected timestamp of each sensor will be checked again to ensure that all data has been accurately aligned.

4. A method for aligning multi-physical quantity data of a power system according to claim 3, characterized in that: The Kalman filter is divided into a prediction step and an update step. In the prediction step, the time synchronization deviation at the current moment is predicted based on the state estimate at the previous moment; in the update step, the time error at the current moment is calculated by comparing it with the timestamp actually measured by the sensor, and the estimated value is corrected according to the Kalman gain.

5. A method for aligning multi-physical quantity data of a power system according to claim 1, characterized in that: Use the outlier detection algorithm based on deep learning to automatically identify and correct abnormal data during the sampling process. The specific steps are as follows: Before using the deep learning-based outlier detection algorithm, it is necessary to first preprocess and extract features from the multi-physical quantity data collected in the power system; After data preprocessing and feature extraction, design and train a deep learning model for outlier detection; After the deep learning model training is completed, it enters the actual application stage. The model calculates the reconstruction error or prediction error of each data according to the input real-time data, and compares it with the pre-set threshold to determine the outlier; In the process of outlier detection, in addition to correcting the abnormal data, the high quality of the final data is ensured.

6. A method for aligning multi-physical quantity data of a power system according to claim 5, characterized in that: During the training process, a large amount of historical data is used to train the model so that the model can adapt to the common change patterns in the power system and distinguish between normal data and abnormal data. During the training process, the network structure and hyperparameters are continuously adjusted to improve the model's detection capabilities and generalization performance.

7. A method for aligning multi-physical quantity data of a power system according to claim 1, characterized in that: The high-quality data set is synthesized and augmented through the generative adversarial network algorithm to generate new training samples. The specific steps are as follows: Prepare and preprocess existing high-quality datasets before using generative adversarial networks for data synthesis and augmentation; Once the data preprocessing is completed, start building the generative adversarial network model; Through the trained GAN model, the generator generates new training samples based on the existing power system data; After generating new training samples, the generated data is strictly verified and quality controlled.

Images