Power system source network coordination performance anomaly detection method
Through the semi-supervised learning method and the semi-supervised K-mean clustering model, combined with marked and unlabeled data, the problem of abnormal detection of coordinated performance of power system source network is solved, and the stability and reliability of the power system are improved.
Patent Information
- Application Number
- CN202411980202.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-31
- Publication Date
- 2025-05-02
AI Technical Summary
The difficulty in detecting abnormal coordination performance of power system source network, especially in the absence of a large amount of abnormal data marked, it is difficult for the prior art to effectively identify and detect abnormal phenomena, affecting the stability and reliability of the power system.
The semi-supervised learning method is adopted, combining marked normal data and unlabeled data, and the semi-supervised K-mean clustering model is trained, and the abnormal coordination performance of the power system source network is identified and detected through feature extraction and clustering analysis.
Effectively capture the characteristics of abnormal data, improve the detection accuracy of abnormal coordination performance of power system source network, and is especially suitable for the absence of labeled samples, and enhance the stable operation guarantee of the power system.
Smart Images
Figure CN119917972A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of power systems, and in particular to a method for detecting abnormality in source-network coordination performance of a power system. Background Art
[0002] In recent years, with the increasing proportion of new energy, the power system is facing increasingly severe challenges. The rapid consumption of new energy puts greater pressure on the power system's regulation resources. Its intermittent, random and volatile characteristics increase the difficulty of controlling the system operation, making the system balance and safety issues more prominent. In this case, the occurrence of abnormal power system source-grid coordination performance may have a serious impact on the stability and reliability of the power system, and even cause system instability or failure. Therefore, studying the abnormal detection method of power system source-grid coordination performance is of great significance to improving the operation stability and reliability of the power system. The study of the characteristics and laws of abnormal data of power system source-grid coordination performance involves the collection, processing and analysis of abnormal data. By extracting the characteristics and analyzing the laws of abnormal data, we can better understand the causes and mechanisms of abnormal occurrence. Establishing an effective abnormal data analysis method is helpful to detect abnormal phenomena early and provide a basis for abnormal processing.
[0003] The rapid development of deep learning has provided new ideas for anomaly detection methods. Commonly used neural networks are usually implemented based on convolution or attention mechanisms, and have powerful feature extraction and application capabilities. These networks support end-to-end training and can effectively solve the problems of complex processes and poor migration in traditional data processing. In the field of deep learning, supervised learning usually has the best performance, but it is difficult to obtain a large number of labeled anomaly samples in actual scenarios. In contrast, semi-supervised learning can better adapt to actual application scenarios by combining labeled and unlabeled data to train models in the absence of labeled samples. Summary of the invention
[0004] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a method for detecting abnormal performance of source-grid coordination in a power system. By combining labeled normal data and unlabeled data, a model is trained to capture the characteristics of abnormal data. According to the characteristics of the abnormal data and the model learning results, the abnormal performance of source-grid coordination in the power system is identified and detected, thereby providing guarantee for the stable operation of the power system.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for detecting abnormality of power system source-grid coordination performance, which comprises the following steps: S1. Source network coordinated frequency response curve collection; S2, source network coordination signal preprocessing; S3, divide the data set into labeled and unlabeled data sets; S4, divide the training set and test set; S5. Build a semi-supervised K-means clustering model and perform training; S6. Optimize model.
[0006] Furthermore, S1 is specifically as follows: when the hydropower unit adjusts the output to adjust the grid frequency, a corresponding frequency response curve is generated, and these frequency response curves are collected and transmitted to the control workstation.
[0007] Furthermore, S2 is specifically as follows: after the control workstation receives these frequency response curve signals, it performs feature extraction on them to obtain steady-state values, rise time, peak time, overshoot and adjustment time values; these values together with the parameters for controlling the hydropower unit at this time constitute a vector, which becomes a data sample in the subsequent training data set.
[0008] Furthermore, S3 is specifically as follows: during the data collection process, a vector, i.e., a data sample, is obtained after each frequency modulation of the hydropower unit; a group of data samples are collected through multiple frequency modulations, and the hydropower unit data previously recorded in the hydropower station are integrated to extract vector data of the same dimension to form a data set for model training; in the data set, labels are added to the data that already contains the unit status; if it can be determined that the unit status is normal at this time, a "normal" label is added to the data sample; through preliminary label addition work, the accuracy of the semi-supervised learning model of subsequent training can be increased; after completing the label addition work, the data set can be divided into labeled and unlabeled parts.
[0009] Furthermore, in the process of dividing the training set and the test set in S4, the training data set consists of some labeled data and all unlabeled data, while the test data set only contains a small amount of labeled data.
[0010] Furthermore, the S5 is specifically: 5.1. Initialization: Use labeled data points to initialize the cluster centers; the cluster center of each labeled point is its true label; 5.2. Assignment: Assign the unlabeled data points to the nearest cluster center; calculate the distance from each data point to all cluster centers, the distance formula uses Euclidean distance or cosine distance, and then select the cluster center with the minimum distance; 5.3. Update: Use the distribution results of all labeled and unlabeled data to update the cluster centers; the cluster centers of labeled data remain unchanged, and the cluster centers of unlabeled data are determined by the mean of the data points assigned to them; 5.4. Iteration: Repeat the assignment and update until the cluster center no longer changes or the predetermined number of iterations is reached. Furthermore, S6 specifically includes: after completing the construction and training of the semi-supervised K-means model, verifying the accuracy of the trained model through a test data set, and optimizing the model by adjusting parameters.
[0011] Furthermore, the accuracy of the trained model is verified by the test data set, and the specific process of optimizing the model by adjusting the parameters is as follows: Input the test set samples into the trained model to obtain the model output results; then, compare the output results with the labels of the test samples to evaluate the accuracy of the model; if a large number of sample outputs do not match the labels, it means that the performance of the model needs to be improved; In order to improve the accuracy of the model, adjust the parameters of the model, including the number of clusters and the distance measurement method; by adjusting the parameters and retraining the model, continuously optimize the model until a satisfactory accuracy is achieved; through this process, the performance of the model can be verified and the model can be continuously optimized.
[0012] The beneficial effects of the present invention are as follows: 1. The present invention combines labeled normal data and unlabeled data to train a model to capture the characteristics of abnormal data. According to the characteristics of abnormal data and model learning results, it identifies and detects abnormalities in the coordination performance of the power system source network, thereby providing guarantee for the stable operation of the power system.
[0013] 2. The patent of this invention successfully extracts features from the data generated by the power system source-grid coordination, prepares sufficient data samples, divides labeled data from unlabeled data, determines the training set and data set, and effectively trains the anomaly detection model.
[0014] 3. The unsupervised learning method of the present invention can effectively discover abnormal data, and is particularly suitable for situations where it is difficult to obtain a large amount of labeled abnormal data in actual scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 A flow chart of a method for detecting abnormal performance of power system source-grid coordination; Figure 2 It is the frequency response curve; Figure 3 This is a visualization result of semi-supervised K-means clustering dimensionality reduction. DETAILED DESCRIPTION
[0016] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0017] Example 1: In the new power system, hydropower, as a highly efficient energy source with both continuous regeneration and flexible adjustment capabilities, faces the volatility and randomness of wind and solar power generation, and undertakes the important task of maintaining the stability of the grid frequency. In the process of power system source-grid coordination, the output change of the hydropower unit will produce a response characteristic curve. Although there is a frequency dead zone, due to the volatility and randomness of wind and solar energy, the hydropower unit needs to be adjusted frequently, and each adjustment will form a frequency response curve. These curves contain information such as steady-state value, rise time, peak time, overshoot and adjustment time, reflecting the actual state of the hydropower unit. Figure 2 What is shown is a frequency response curve, through which we can extract the rise time of 2.0877s, the adjustment time of 17.3158s, the peak time of 3.5614s, the overshoot of 40.1285%, and the steady-state value of 0.111. The parameter information extracted from a frequency response curve is a data sample. The response characteristic curves of the same hydropower unit under different control adjustment parameters will be different, and the characteristics of the obtained data samples will also be different.
[0018] The present invention proposes a set of schemes for collecting and analyzing the response characteristic curves generated by the power system source network coordination. Through this scheme, the semi-supervised clustering model suitable for the power system source network coordination proposed by the present invention can be used to detect whether the hydropower unit has abnormal conditions. The implementation of this scheme will help to timely discover and solve the problems in the power system source network coordination, thereby improving the stability and reliability of the power system. Figure 1 It is a flow chart of the present invention. The specific process is as follows: 1. Data collection and processing: When the hydroelectric unit adjusts its output to adjust the grid frequency, a corresponding frequency response curve is generated. The present invention records these response characteristic curves and transmits them to the control workstation. Once the control workstation receives these signals, it extracts features to obtain values such as steady-state value, rise time, peak time, overshoot, and adjustment time. These values, together with the parameters controlling the hydroelectric unit at this time, form a vector, which becomes a data sample in the subsequent training data set. Figure 2 The parameters contained in and the control parameters of the hydropower unit together constitute a data sample. Data collection requires obtaining a large number of data samples, and sufficient data samples can ensure the accuracy of the training model.
[0019] Information collection of power system source-grid coordinated response characteristic curve: Hydropower units control hydropower output by adjusting the guide vane opening to stabilize the grid frequency. However, in this process, bearing vibration, guide vane vibration, and hydraulic self-excited vibration may cause noise interference to the response signal. Transmitting this data from the hydropower unit to the workstation is also a key point solved by the patent of this invention. In order to solve this problem, the present invention uses autocorrelation analysis to filter the signal and uses a 10G shielded network cable to ensure fast and stable transmission of signal data.
[0020] 2. Build and train the model: In the above data collection process, a vector, i.e., a data sample, is obtained each time the hydropower unit is frequency modulated. Through multiple frequency modulations, this patent can collect a set of data samples, and integrate the hydropower unit data previously recorded by the hydropower station, extract vector data of the same dimension, and together form a data set that can be used for model training. In this data set, this patent adds labels to data that already contains the unit status. For example, a hydropower unit obtains a frequency response curve under a set of control parameters, and the corresponding steady-state value, rise time, peak time, overshoot, adjustment time and other information can be obtained through data extraction. The unit control parameters and the corresponding response characteristic curve information together constitute a vector. If it can be determined that the unit status is normal at this time, a "normal" label can be added to the data sample.
[0021] The accuracy of the semi-supervised learning model in subsequent training can be increased through the preliminary labeling work. After completing the labeling work, the patent can divide the data set into labeled and unlabeled parts, and then further divide it into training sets and test sets. The training data set consists of some labeled data and all unlabeled data, while the test data set only contains a small amount of labeled data. By organizing the data set, it provides a basis for subsequent model training, and also increases the speed and accuracy of model training.
[0022] The training model constructed by the present invention is a semi-supervised K-means clustering algorithm model. This model is improved under the unsupervised K-means clustering model. Unsupervised K-means clustering divides data points into K different clusters, where K is set in advance. It does not require label information to guide the clustering process, but determines the optimal cluster division by the distance between data points. The following is an introduction to the unsupervised K-means clustering model: Input data sample set: ; The number of clusters is k, and the output is cluster C= ; (1) Randomly initialize k points as cluster centroids; (2) Assign each point in the sample set to a cluster; calculate the distance between each point and the centroid. The distance formula can be Euclidean distance or cosine distance. The Euclidean distance is: ; The cosine distance is: ; Choose a distance formula to calculate according to the actual situation, and assign each point to the cluster corresponding to the nearest centroid after calculation; (3) Update the centroid of the cluster: The centroid of each cluster is updated to the average value of all points in the cluster; (4) Repeat steps (2) to (3) until a certain termination condition is reached. Common termination conditions include: 1) reaching a specified number of iterations; 2) the cluster center no longer changes significantly, i.e. convergence; 3) minimum sum of squared errors (SSE); (5) Clustering effect evaluation index SSE (Sum of Square Error). The smaller the SSE value, the closer the data points are to their centroid, and the better the clustering effect.
[0023] ; ; Semi-Supervised K-Means is an extension of the K-Means clustering algorithm, which combines labeled data and unlabeled data for clustering. In the traditional K-Means algorithm, all data points are unlabeled, while in the semi-supervised K-Means, some data points can be labeled while others are unlabeled. The basic idea of semi-supervised K-Means is labeled data: (1) Initialize cluster centers using labeled data points.
[0024] (2) Unlabeled data: Assign unlabeled data points to the nearest cluster center.
[0025] (3) Update cluster centers: Use the distribution results of all labeled and unlabeled data to update the cluster centers.
[0026] (4) Iterative optimization: Repeat the above steps until the cluster center is stable and no longer changes or the predetermined number of iterations is reached.
[0027] Semi-supervised K-means has better generalization ability than unsupervised K-means, and is also suitable for complex scenarios such as hydropower units, and can produce more accurate and robust clustering results. The following are the steps for semi-supervised K-means: (1) Initialization: Use labeled data points to initialize the cluster centers. The cluster center of each labeled point is its true label.
[0028] (2) Assignment: Assign the unlabeled data points to the nearest cluster center. This can be done by calculating the distance from each data point to all cluster centers using the same distance formula as the unsupervised K-means method described above, and then selecting the cluster center with the smallest distance.
[0029] (3) Update: Use the assignment results of all labeled and unlabeled data to update the cluster centers. The cluster centers of labeled data remain unchanged, and the cluster centers of unlabeled data are determined by the mean of the data points assigned to them.
[0030] (4) Iteration: Repeat the assignment and update until the cluster center no longer changes or the predetermined number of iterations is reached.
[0031] The above process completes the construction and training of the semi-supervised K-means model. This patent can use this model to perform subsequent power system source-grid coordination performance anomaly detection tasks and realize real-time monitoring of the status of hydropower units.
[0032] Figure 3 This is the model training result for the data sample of the same unit control parameter. When the model is trained on the data sample of the same unit control parameter, the data sample only contains five dimensions of data: steady-state value, rise time, peak time, overshoot and adjustment time. Figure 2 It contains 300 data samples, of which 50 are labeled data. After the model training is completed, the 5-dimensional data is reduced to 3-dimensional data through principal component analysis (PCA) for visualization. The training results show that semi-supervised K-means clustering has good data partitioning capabilities.
[0033] Semi-supervised K-means clustering model for anomaly detection of power system source-grid coordination performance: Semi-supervised K-means clustering combines the ideas of K-means clustering algorithm and semi-supervised learning. When the model is applied to the task of anomaly detection of power system source-grid coordination performance, it is necessary to extract features related to the power system source-grid coordination performance, including steady-state value, rise time, peak time, overshoot and adjustment time, etc. At the same time, sufficient training data is required to ensure the accuracy of the model. The present invention successfully extracts features from the data generated by the power system source-grid coordination, prepares sufficient data samples, divides labeled data into unlabeled data, and determines the training set and data set, effectively training the anomaly detection model.
[0034] 3. Optimization model: After completing the construction and training of the semi-supervised K-means model, this patent will verify the accuracy of the trained model through a test data set and optimize the model by adjusting parameters.
[0035] Specifically, this patent will input the test set samples into the trained model to obtain the model's output results. Then, the output results are compared with the labels of the test samples to evaluate the accuracy of the model. If a large number of sample outputs do not match the labels, it means that the performance of the model needs to be improved.
[0036] In order to improve the accuracy of the model, the present invention adjusts the parameters of the model, including the number of clusters, distance measurement method, etc. By adjusting the parameters and retraining the model, the model can be continuously optimized until a satisfactory accuracy is achieved. Through this process, the performance of the model can be verified and the model can be continuously optimized to ensure that it has high accuracy and generalization ability in practical applications.
[0037] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.
Claims
1. A method for detecting abnormality in power system source-grid coordination performance, characterized in that: It includes the following steps: S1. Source network coordinated frequency response curve collection; S2, source network coordination signal preprocessing; S3, divide the data set into labeled and unlabeled data sets; S4, divide the training set and test set; S5. Build a semi-supervised K-means clustering model and perform training; S6. Optimize model.
2. A method for detecting abnormality in power system source-grid coordination performance according to claim 1, characterized in that: Specifically, S1 is as follows: when the hydropower unit adjusts its output to adjust the grid frequency, a corresponding frequency response curve is generated, and these frequency response curves are collected and transmitted to the control workstation.
3. A method for detecting abnormality in power system source-grid coordination performance according to claim 1, characterized in that: Specifically, S2 is as follows: after receiving these frequency response curve signals, the control workstation performs feature extraction to obtain steady-state value, rise time, peak time, overshoot and adjustment time values; these values together with the parameters for controlling the hydropower unit at this time form a vector, which becomes a data sample in the subsequent training data set.
4. A method for detecting abnormality in power system source-grid coordination performance according to claim 1, characterized in that: S3 is specifically as follows: during the data collection process, a vector, i.e., a data sample, is obtained after each frequency modulation of the hydropower unit; a group of data samples are collected through multiple frequency modulations, and the hydropower unit data previously recorded in the hydropower station are integrated to extract vector data of the same dimension to form a data set for model training; in the data set, labels are added to the data that already contains the unit status; If it can be determined at this time that the unit status is normal, a "normal" label is added to the data sample; through the preliminary label adding work, the accuracy of the semi-supervised learning model in subsequent training can be increased; after completing the label adding work, the data set can be divided into labeled and unlabeled parts.
5. A method for detecting abnormality in power system source-grid coordination performance according to claim 1, characterized in that: In the process of dividing the training set and the test set in S4, the training data set consists of some labeled data and all unlabeled data, while the test data set only contains a small amount of labeled data.
6. A method for detecting abnormality in power system source-grid coordination performance according to claim 1, characterized in that: The S5 is specifically: 5.
1. Initialization: Use labeled data points to initialize the cluster centers; the cluster center of each labeled point is its true label; 5.
2. Assignment: Assign the unlabeled data points to the nearest cluster center; calculate the distance from each data point to all cluster centers, the distance formula uses Euclidean distance or cosine distance, and then select the cluster center with the minimum distance; 5.
3. Update: Use the distribution results of all labeled and unlabeled data to update the cluster centers; the cluster centers of labeled data remain unchanged, and the cluster centers of unlabeled data are determined by the mean of the data points assigned to them; 5.
4. Iteration: Repeat the assignment and update until the cluster center no longer changes or the predetermined number of iterations is reached.
7. A method for detecting abnormality in power system source-grid coordination performance according to claim 1, characterized in that: The S6 is specifically as follows: after completing the construction and training of the semi-supervised K-means model, the accuracy of the trained model is verified through a test data set, and the model is optimized by adjusting parameters.
8. A method for detecting abnormality in power system source-grid coordination performance according to claim 7, characterized in that: The specific process of verifying the accuracy of the trained model through the test data set and optimizing the model by adjusting the parameters is as follows: Input the test set samples into the trained model to obtain the model output results; then, compare the output results with the labels of the test samples to evaluate the accuracy of the model; if a large number of sample outputs do not match the labels, it means that the performance of the model needs to be improved; In order to improve the accuracy of the model, adjust the parameters of the model, including the number of clusters and the distance measurement method; by adjusting the parameters and retraining the model, continuously optimize the model until a satisfactory accuracy is achieved; through this process, the performance of the model can be verified and the model can be continuously optimized.