User side load abnormity situation awareness method and system based on radar map multi-dimensional evaluation

Abstract

The invention discloses a user side load abnormal situation awareness method and system based on radar map multi-dimensional evaluation. The method comprises the steps that power operation parameters and external environment parameters form a unified data input stream; performing feature extraction on the unified data input stream, and constructing a situation vector of a multi-dimensional index; and classifying the situation feature vectors, distinguishing normal and abnormal modes, and forming a multi-dimensional abnormal feature set. By means of the method, comprehensive modeling and abnormal accurate recognition of user side load multi-dimensional features are achieved, the visual expression ability and quantitative evaluation level of the load abnormal situation are remarkably improved, and therefore a rapid, visual and reliable decision basis is provided for an electric power operation and maintenance department.

Description

Technical Field

[0001] This invention belongs to the field of power system user-side load monitoring and energy management technology, specifically involving a user-side load anomaly situation perception method and system based on radar chart multidimensional assessment. Background Technology

[0002] With the rapid development of new power systems, the role of the user side in power system operation is becoming increasingly prominent. The large-scale integration of new loads such as distributed photovoltaics, energy storage devices, and electric vehicles has resulted in significant volatility, randomness, and diversity in user-side electricity consumption characteristics. This not only places higher demands on the safe and stable operation of the power grid but also makes the monitoring and assessment of abnormal user load conditions a crucial issue that power companies must address. Traditional technical solutions typically rely on single parameters such as current, voltage, or power, using threshold values ​​to determine anomalies. While this method is simple to implement, it is prone to misjudgment due to frequent load fluctuations or short-term surges in practical applications, failing to fully reveal the multidimensional characteristics of user loads. Another approach is based on statistical or cluster analysis, such as using algorithms like K-means, fuzzy clustering, or density clustering to classify and identify patterns in load data. This improves adaptability to complex loads to some extent, but the results usually remain at the numerical or categorical level, lacking intuitive and effective visualization methods, making it difficult for operation and maintenance personnel to quickly understand and assess the severity of anomalies.

[0003] Meanwhile, existing methods have significant limitations in feature construction. Most studies rely on only a few power parameters while neglecting the influence of external environmental factors. However, on the user side, load status is often affected by a combination of external conditions such as temperature, humidity, and light intensity. If these data cannot be integrated with power parameters, the resulting situational model often fails to accurately reflect the operating patterns of user loads. On the other hand, existing anomaly detection results are mostly presented in the form of numerical indicators or two-dimensional curves. For load data containing multi-dimensional features such as current, voltage, power fluctuation rate, and energy entropy, traditional methods struggle to simultaneously display the differences in anomalies across various dimensions. This results in anomaly identification results that are not intuitive, making it difficult for maintenance personnel to make timely and reasonable judgments when faced with complex load anomaly scenarios.

[0004] As power systems evolve towards digitalization and intelligence, operation and maintenance departments are placing higher demands on the intuitiveness, accuracy, and operability of load anomaly identification. Single-parameter determination methods are insufficient to meet the high-precision monitoring needs of modern power systems, while anomaly identification methods based on numerical results lack sufficient interpretability and visualization capabilities, failing to effectively support practical decision-making.

[0005] Therefore, how to construct a new method that can integrate multi-source data, extract multi-dimensional features, and combine intuitive graphical means to comprehensively assess load anomalies has become an urgent problem to be solved. Summary of the Invention

[0006] Therefore, the purpose of this invention is to provide a method and system for user-side load anomaly situation perception based on radar chart multidimensional assessment, so as to achieve comprehensive modeling and accurate anomaly identification of multidimensional characteristics of user-side load, and provide power operation and maintenance departments with fast, intuitive and reliable decision-making basis.

[0007] The technical solution provided by this invention is: a user-side load anomaly situation perception method based on radar chart multi-dimensional assessment, comprising:

[0008] The power operation parameters and external environmental parameters are integrated into a unified data input stream;

[0009] Feature extraction is performed on the unified data input stream to construct a situation vector of multi-dimensional indicators;

[0010] The situation feature vectors are classified to distinguish between normal and abnormal patterns, forming a multidimensional set of abnormal features.

[0011] Preferably, the power operating parameters include user-side voltage. Current Active power With reactive power The data is acquired in real time by the intelligent power acquisition terminal installed at the user access point;

[0012] The external environmental parameters include temperature. ,humidity and light intensity Data is collected through IoT sensors;

[0013] Power operating parameters and external environmental parameters are sampled at the same period. Record it and include a timestamp.

[0014] Preferably, the step of forming a unified data input stream from power operation parameters and external environmental parameters includes:

[0015] The power operation parameters and external environmental parameters are normalized, and different parameters are mapped to... The interval is determined using the following formula:

[0016]

[0017] in, Indicates the original parameter value. and These are the minimum and maximum values ​​of the parameter within the sampling period, respectively.

[0018] Multi-source fusion is performed on the normalized parameters, and weighting coefficients are introduced into them. Calculate the unified input signal :

[0019]

[0020] Detecting and correcting outliers in parameter data using the sliding window method:

[0021] Set the window length to k, and calculate the local mean at each time t. and standard deviation And calculate the standardized outlier factor:

[0022]

[0023] when Greater than the threshold This point was identified as an outlier and replaced with the nearest mean.

[0024] Wavelet transform is used to decompose the signal:

[0025]

[0026] in, For the first The low-frequency approximation components of the layer reflect the long-term trend of the signal; For high-frequency detail components of each layer;

[0027] Using threshold function High-frequency components are compressed to remove noise:

[0028]

[0029] The final denoised signal is obtained through reconstruction:

[0030]

[0031] Based on normalization, weighted fusion, outlier correction, and wavelet denoising, a unified input data stream is formed:

[0032]

[0033] in Indicates the total number of parameters.

[0034] Preferably, the step of extracting features from the unified data input stream and constructing a multi-dimensional feature vector includes:

[0035] The input stream is divided into several consecutive time segments using the sliding time window method. The window length is set to W. Then, at time t, the parameters within the window... The set of possible values ​​is:

[0036]

[0037] in, Indicates the first The parameters at time... The normalized and denoised values;

[0038] Within each window, extract the mean feature, volatility feature, peak-to-valley difference feature, and energy entropy feature;

[0039] The mean characteristic, volatility characteristic, peak-to-valley difference characteristic, and energy entropy characteristic are combined into a multi-dimensional feature vector:

[0040]

[0041] in, Indicates at time Within the corresponding window, there is a multidimensional feature set for each parameter.

[0042] Preferably, the formula for calculating the mean characteristic is:

[0043]

[0044] in, Indicates parameters In the window The average value within;

[0045] The formula for calculating volatility characteristics is:

[0046]

[0047] like A large value indicates that the load is in an abnormal fluctuation state;

[0048] The peak-valley difference characteristic is defined as:

[0049]

[0050] in, Indicates parameters The difference between the maximum and minimum values ​​within the window reflects the extreme variation in load during that time period;

[0051] Energy entropy characteristics: Definition of normalized energy distribution:

[0052]

[0053] in, Indicates parameters The energy percentage at the k-th time within the window;

[0054] Calculate energy entropy based on energy entropy characteristics:

[0055]

[0056] in, Indicates parameters Energy entropy within a window is used to measure the complexity of energy distribution.

[0057] Preferably, a feature reduction method is used to transform the multidimensional feature set into a high-dimensional feature vector. Mapping to a low-dimensional feature space:

[0058]

[0059] Among them, matrix For the dimension reduction mapping matrix, This represents a low-dimensional feature representation.

[0060] Preferably, an improved clustering algorithm is used to classify the situation feature vectors, distinguishing between normal and abnormal patterns, and then combined with a dynamic threshold discrimination mechanism to form a multi-dimensional abnormal feature set:

[0061] Let the set of eigenvectors be:

[0062]

[0063] in, Indicates at time The low-dimensional situational feature vector, where n represents the number of samples;

[0064] Density clustering is performed on the feature vector set for any sample point. Its neighborhood density is defined as:

[0065]

[0066] in, Indicates sample neighborhood density, Represents the neighborhood radius. As an indicator function, points with lower neighborhood density correspond to abnormal loads;

[0067] Based on the density clustering, fuzzy C-means is used to further distinguish the membership relationships of clusters, and its objective function is:

[0068]

[0069] in, Indicates sample For cluster center Membership degree, range of values And satisfy ;parameter >1 is the fuzzy weighting coefficient, used to control the degree of fuzziness;

[0070] Iterative update and cluster center To obtain clustering results;

[0071] A dynamic threshold mechanism based on time series statistical features is used to evaluate the feature sequence of a certain indicator. Calculate the mean within the window length 𝑊 and standard deviation The dynamic threshold is set as follows:

[0072]

[0073] in, This represents the dynamic threshold at time t. This is an adjustment coefficient; when the feature value of the sample exceeds... When this occurs, it is determined to be an abnormal state;

[0074] Combining the clustering results with the dynamic threshold discrimination results, a multidimensional set of anomaly features is obtained:

[0075]

[0076] in, Indicates at time The set of features that are determined to be abnormal For the first The dynamic threshold of each feature, if the feature vector Falling into an isolated cluster is also considered an anomaly.

[0077] Preferably, the method further includes: mapping the abnormal features to the coordinate system of the radar image, and calculating the area of ​​the radar image. Morphological differences With boundary volatility Maintenance personnel based on the area of ​​the radar chart Morphological differences With boundary volatility Based on the combined results, the severity of the abnormality is determined:

[0078] The radar image coordinate mapping: assuming at a certain moment... The set of abnormal features contains One dimension:

[0079]

[0080] in, Indicates the first The numerical values ​​of each abnormal feature, and each feature Mapped onto the polar coordinate axis of the radar chart, the angle is:

[0081]

[0082] The radius is:

[0083]

[0084] in, Indicates the first The angle at which each feature is located. This represents the projection value of the feature onto the radar image. As the indicator weight, satisfying ;

[0085] Calculating the area of ​​a radar image: The area of ​​a polygon in a radar image is defined as follows:

[0086]

[0087] in, Indicates at time The area of ​​the radar image, if Sometimes The larger the area, the more severe the multidimensional anomaly.

[0088] Calculate morphological difference: The morphological difference index is defined as follows:

[0089]

[0090] in, Represents the average radius, if A larger value indicates that certain abnormal characteristics are more prominent, suggesting a risk of bias in the load condition.

[0091] Calculating boundary volatility: Boundary volatility characterizes the dynamic changes in the shape of a radar chart, and is defined as follows:

[0092]

[0093] in, This represents the mean square error of the radar image boundary between adjacent time periods; a larger value indicates a more drastic evolution of the abnormal situation.

[0094] Secondly, this aspect also provides a user-side load anomaly situation awareness system based on radar chart multi-dimensional assessment, including:

[0095] Data preprocessing module: Combines power operation parameters and external environmental parameters into a unified data input stream;

[0096] The feature extraction module extracts features from a unified data input stream and constructs a situation vector for multi-dimensional indicators.

[0097] The classification module classifies the situation feature vectors, distinguishes between normal and abnormal patterns, and forms a multi-dimensional abnormal feature set.

[0098] Preferably, it also includes: a mapping module, which maps the set of abnormal features to the radar chart coordinate system, calculates the difference in graphic area and shape and the degree of abnormality based on the weight of each indicator, and intuitively displays the dynamic changes in the load situation in combination with the time series evolution.

[0099] This invention provides a method and system for user-side load anomaly situation perception based on radar chart multidimensional assessment. The method achieves complete acquisition and high-quality input of load information through multi-source data fusion and adaptive denoising, characterizes the user load operation status with multidimensional situation feature vectors, and identifies abnormal patterns by combining improved clustering algorithm and dynamic threshold mechanism. Finally, the multidimensional abnormal features are mapped to the radar chart coordinate system to realize intuitive display and quantitative assessment of abnormal situation.

[0100] This invention introduces radar charts, a multi-dimensional visualization tool, which can not only display anomalies of multiple indicators in the same graphic, but also quantify the severity of anomalies through differences in area and shape, thereby significantly improving the ability to identify and assess abnormal load conditions of users. This is of great significance for ensuring the safe and stable operation of the power grid and improving the decision-making efficiency of operation and maintenance personnel.

[0101] In summary, this invention effectively overcomes the shortcomings of existing technologies, such as inaccurate judgment of single indicators, lack of visual expression of anomaly identification results, and insufficient dynamic adaptability, thereby providing a comprehensive, intuitive, and highly reliable means for monitoring and supporting user load anomalies in power system operation and maintenance. Attached Figure Description

[0102] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0103] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0104] Figure 1The flowchart shows the user-side load anomaly situation perception method based on radar chart multidimensional assessment provided by the present invention. Detailed Implementation

[0105] The present invention will be further explained below with reference to specific implementation schemes, but this explanation does not limit the scope of the invention.

[0106] Given the trend towards digitalization and intelligentization in power systems, operation and maintenance departments are placing higher demands on the intuitiveness, accuracy, and operability of load anomaly identification, such as... Figure 1 As shown, this invention proposes a user-side load anomaly situation perception method based on radar chart multidimensional assessment, including:

[0107] S1: Collect user-side voltage, current, active power, reactive power, and external environment data, and use multi-source fusion and adaptive noise reduction algorithms to form a unified data input stream to ensure the integrity and reliability of load information;

[0108] Firstly, it is necessary to fully collect and fuse load data from multiple sources on the user side to ensure the quality of the foundational data for subsequent feature modeling and anomaly identification. The collected data mainly includes two categories: one is power operating parameters, such as voltage... Current Active power With reactive power Another category is external environmental parameters, such as temperature. ,humidity and light intensity Power parameters are acquired in real time by smart power acquisition terminals installed at user access points, while environmental parameters are collected via IoT sensors. All data are sampled at the same period. Record the data and attach a timestamp to ensure data alignment.

[0109] The aforementioned multi-source fusion includes weighted normalization of power parameters and environmental parameters, and suppression of transient interference through wavelet denoising or adaptive filtering algorithms;

[0110] S11: Data normalization processing, including: Due to the significant differences in the dimensions and value ranges of various parameters, direct use would lead to an imbalance in the influence between different features. Therefore, this invention first normalizes the collected data using the following formula:

[0111]

[0112] in, Indicates the original parameter value. and These are the minimum and maximum values ​​of the parameter within the sampling period, respectively. After this processing, different parameters are mapped to... The interval is used to ensure that all indicators are comparable in subsequent analysis.

[0113] S12: Weighted Multi-Source Fusion: In the multi-source fusion stage, this invention considers the differences in the importance of different parameters to load situation perception. For example, changes in current and active power more directly reflect load characteristics, while the influence of environmental factors such as humidity or light intensity is relatively minor. Therefore, weighting coefficients are introduced. Calculate the unified input signal :

[0114]

[0115] in, It can be set based on expert experience or learned automatically through methods such as information gain or mutual information analysis. This method not only ensures the comprehensive utilization of multi-source information but also highlights the role of key indicators in situation modeling.

[0116] S13: Outlier Detection and Correction: In the data quality control stage, this invention uses a sliding window method to detect and correct outliers. The window length is set to k, and at each time t, the local mean is calculated. and standard deviation And calculate the standardized outlier factor:

[0117]

[0118] when Greater than the threshold This point was identified as an outlier and replaced with the nearest mean. This avoids the interference of single-point noise on the overall result while maintaining the continuity and smoothness of the signal.

[0119] S14: Wavelet denoising and filtering. Considering the potential high-frequency interference in the acquired data, this invention employs wavelet transform to decompose the signal during the denoising stage.

[0120]

[0121] in, For the first The low-frequency approximation components of the layer reflect the long-term trend of the signal; These are the high-frequency detail components of each layer, mainly consisting of short-time fluctuations and noise. This invention utilizes a threshold function. High-frequency components are compressed to remove noise:

[0122]

[0123] Among them, threshold It can be adaptively determined based on the noise variance. The final denoised signal is obtained through reconstruction:

[0124]

[0125] S15: Forming a unified input stream: Through the above-mentioned normalization, weighted fusion, outlier correction, and wavelet denoising processes, this invention ultimately forms a unified input data stream:

[0126]

[0127] This input stream not only ensures the consistency and robustness of multi-source data under complex power operation environments, but also provides a high-quality input foundation for subsequent feature extraction, situation modeling, and anomaly identification.

[0128] S2: Based on the time series window, feature extraction is performed on the data to construct a situation vector containing multi-dimensional indicators such as mean, volatility, peak-to-valley difference, and energy entropy, which is uniformly mapped to the feature space; the situation feature vector further includes power factor, load fluctuation entropy, and frequency domain energy distribution to enhance feature distinguishability;

[0129] Based on the unified input stream obtained in step S1 The method employs time series windowing to extract features from load data, thereby constructing a multi-dimensional feature vector that comprehensively describes the load situation. The core of this step lies in forming a feature set that characterizes the user-side load operating status by performing statistical, dynamic, and complexity analyses on data within different time windows.

[0130] S21: The input stream is divided into several continuous time segments using a sliding time window method. Let the window length be W. Then, at time t, the parameters within the window... The set of possible values ​​is:

[0131]

[0132] in, Indicates the first The parameters at time... The normalized and denoised values. Indicates the total number of parameters.

[0133] Within each window, extract the following features:

[0134] (1) Mean characteristics

[0135] The mean reflects the overall load level over a given period of time, and its calculation formula is as follows:

[0136]

[0137] in, Indicates parameters In the window The average value within a given period reflects the overall level of the parameter during that time period.

[0138] (2) Volatility characteristics

[0139] Volatility describes the stability of data within a window, and the calculation formula is:

[0140]

[0141] like A large value indicates that the load is in an abnormal fluctuation state. A large fluctuation rate usually indicates that the load has strong random fluctuations, which may be related to frequent equipment start-ups and shutdowns or external environmental disturbances. It is an important reference indicator for identifying anomalies.

[0142] (3) Peak-valley difference characteristics

[0143] The peak-to-valley difference reflects the impact characteristics of the load within a window, and is defined as:

[0144]

[0145] in, Indicates parameters The difference between the maximum and minimum values ​​within the window reflects the extreme variation in load during that time period.

[0146] (4) Energy entropy characteristics

[0147] First, define the normalized energy distribution:

[0148]

[0149] in, Indicates parameters The percentage of energy at the k-th time within the window.

[0150] Based on this, calculate the energy entropy:

[0151]

[0152] in, Indicates parameters The energy entropy within the window is used to measure the complexity of the energy distribution. A higher energy entropy indicates a uniform energy distribution and complex signal patterns; a lower energy entropy indicates concentrated energy and a simple signal patterns.

[0153] S22: Constructing a situation feature vector: Combining the various indicators obtained through the above calculations into a multi-dimensional feature vector:

[0154]

[0155] in, This indicates that within the window corresponding to time t, the multidimensional feature set of each parameter can comprehensively depict the operating status of the user-side load.

[0156] S23: Feature Space Mapping: To reduce redundant features and improve computational efficiency, feature dimensionality reduction methods, such as Principal Component Analysis (PCA), are further employed to transform high-dimensional feature vectors... Mapping to a low-dimensional feature space:

[0157]

[0158] Among them, matrix For the dimension reduction mapping matrix, This is a low-dimensional feature representation. This process reduces computational complexity while preserving key information, providing efficient input for subsequent anomaly detection.

[0159] S3: An improved clustering algorithm is used to classify the situation feature vectors, distinguish between normal and abnormal patterns, and then combined with a dynamic threshold discrimination mechanism to form a multi-dimensional abnormal feature set, thereby realizing the typological identification of anomalies.

[0160] The improved clustering algorithm is a hybrid algorithm combining density clustering and fuzzy clustering to improve the accuracy of identifying nonlinear load anomalies; the situation feature vector obtained in step S2 or the dimensionality-reduced feature vector Based on this, an improved clustering algorithm is introduced for pattern classification to achieve automatic differentiation between normal and abnormal loads.

[0161] S31: Let the set of eigenvectors be:

[0162]

[0163] in, Indicates at time The low-dimensional situation feature vector, where n represents the number of samples.

[0164] S32: Density Clustering: Density clustering is used to identify unevenly distributed data clusters for any given sample point. Its neighborhood density is defined as:

[0165]

[0166] in, Indicates sample neighborhood density, Represents the neighborhood radius. This is an indicator function. Points with low neighborhood density are often isolated points and may correspond to abnormal loads.

[0167] S33: Fuzzy Clustering: Based on density clustering, the fuzzy C-means (FCM) algorithm is introduced to further distinguish the membership relationships of clusters. Its objective function is:

[0168]

[0169] in, Indicates sample For cluster center Membership degree, range of values And satisfy ;parameter >1 represents the fuzzy weighting coefficient, used to control the degree of fuzziness. Iterative updates are performed. and cluster center This allows for more stable clustering results.

[0170] S34: Dynamic Threshold Determination: To prevent fixed thresholds from failing under different load scenarios, a dynamic threshold mechanism based on time series statistical features is introduced. This mechanism is applied to a specific indicator feature sequence. Calculate the mean within the window length 𝑊 and standard deviation The dynamic threshold is set as follows:

[0171]

[0172] in, This represents the dynamic threshold at time t. This is the adjustment coefficient. When the feature value of the sample exceeds... When this occurs, it is determined to be an abnormal state.

[0173] S35: Forming an anomaly feature set: Combining the results of improved clustering identification and dynamic threshold discrimination, a multidimensional anomaly feature set is obtained.

[0174]

[0175] in, Indicates at time The set of features that are determined to be abnormal For the first Dynamic thresholds for each feature. If the feature vector... Falling into an isolated cluster is also considered an anomaly.

[0176] By using the above method, while maintaining the stability of clustering results, the adaptability to dynamic changes in load is enhanced, thereby achieving the typological identification of anomalies. This not only distinguishes between normal and abnormal states, but also reveals the specific dimensional characteristics of anomalies.

[0177] S4: Map the set of abnormal features to the radar chart coordinate system, calculate the area and shape difference of the graph according to the weight of each indicator, realize the quantification of the severity of the anomaly, and intuitively show the dynamic changes of the load situation by combining the time series evolution. The quantification result of the severity of the anomaly is calculated by the area of ​​the radar chart, the shape asymmetry, and the boundary volatility.

[0178] The abnormal feature set obtained in step S3 Based on this, abnormal features are mapped to the coordinate system of the radar chart to achieve intuitive visualization and quantitative evaluation of multi-dimensional features.

[0179] S41: Mapped radar map coordinates:

[0180] Suppose at a certain moment The set of abnormal features contains One dimension:

[0181]

[0182] in, Indicates the first The numerical values ​​of each anomalous feature. (The last part is incomplete and likely refers to a separate context.) Mapped onto the polar coordinate axis of the radar chart, the angle is:

[0183]

[0184] The radius is:

[0185]

[0186] in, Indicates the first The angle at which each feature is located. This represents the projection value of the feature onto the radar image. As the indicator weight, satisfying .

[0187] S42: Calculate the radar image area: To quantify the overall severity of the anomaly, the area of ​​the radar image polygon is calculated, defined as:

[0188]

[0189] in, This represents the radar map area at time t, if Sometimes The larger the area, the more severe the multidimensional anomaly.

[0190] S43: Calculation of Morphological Dissimilarity: In addition to the overall area, a morphological dissimilarity index is introduced to reflect whether the multidimensional abnormal distribution is balanced. It is defined as:

[0191]

[0192] in, This represents the average radius. If... A larger value indicates that certain abnormal characteristics are more prominent, suggesting a risk of bias in the load condition.

[0193] S44: Calculating Boundary Volatility: To characterize the dynamic changes in the shape of a radar chart, a boundary volatility index is proposed, defined as:

[0194]

[0195] in, This represents the mean square error of the radar image boundary between adjacent time periods; a larger value indicates a more drastic evolution of the abnormal situation.

[0196] S45: Situation Evolution Visualization: By overlaying and comparing radar charts from consecutive time points, the evolution trend of load anomalies over different time periods can be intuitively reflected. Maintenance personnel can analyze the area of ​​the radar chart... Morphological differences With boundary volatility The comprehensive results allow for a rapid assessment of the severity, primary sources, and development trends of the anomalies, enabling the formulation of more targeted control measures.

[0197] This invention proposes a user-side load anomaly situation perception method based on radar chart multidimensional assessment. It ensures the integrity and robustness of input information through multi-source data fusion and adaptive denoising. A situation vector is constructed by combining multidimensional features such as mean, volatility, peak-valley difference, and energy entropy within a time window. An improved clustering and dynamic threshold mechanism are used to accurately identify anomaly patterns. Finally, the anomaly results are mapped to the radar chart coordinate system, and quantitative and visual assessments are performed using graphic area, morphological differences, and boundary volatility. This enables comprehensive perception and dynamic tracking of user-side load anomalies, providing rapid, intuitive, and reliable decision support for power system operation and maintenance.

[0198] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0199] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A user-side load anomaly situation perception method based on radar chart multi-dimensional assessment, characterized in that, include: The power operation parameters and external environmental parameters are integrated into a unified data input stream; Feature extraction is performed on the unified data input stream to construct a situation vector of multi-dimensional indicators; The situation feature vectors are classified to distinguish between normal and abnormal patterns, forming a multidimensional set of abnormal features.

2. The user-side load anomaly situation perception method based on radar chart multidimensional assessment according to claim 1, characterized in that, The power operating parameters include user-side voltage. Current Active power With reactive power The data is acquired in real time by the intelligent power acquisition terminal installed at the user access point; The external environmental parameters include temperature. ,humidity and light intensity Data is collected through IoT sensors; Power operating parameters and external environmental parameters are sampled at the same period. Record it and include a timestamp.

3. The user-side load anomaly situation perception method based on radar chart multidimensional assessment according to claim 1, characterized in that, The process of forming a unified data input stream from power operation parameters and external environmental parameters includes: The power operation parameters and external environmental parameters are normalized, and different parameters are mapped to... The interval is determined using the following formula: ； in, Indicates the original parameter value. and These are the minimum and maximum values ​​of the parameter within the sampling period, respectively. Multi-source fusion is performed on the normalized parameters, and weighting coefficients are introduced into them. Calculate the unified input signal : ； Detecting and correcting outliers in parameter data using the sliding window method: Set the window length to k, and calculate the local mean at each time t. and standard deviation And calculate the standardized outlier factor: ； when Greater than the threshold This point was identified as an outlier and replaced with the nearest mean. Wavelet transform is used to decompose the signal: ； in, For the first The low-frequency approximation components of the layer reflect the long-term trend of the signal; For high-frequency detail components of each layer; Using threshold function High-frequency components are compressed to remove noise: ； The final denoised signal is obtained through reconstruction: ； Based on normalization, weighted fusion, outlier correction, and wavelet denoising, a unified input data stream is formed: ； in Indicates the total number of parameters.

4. The user-side load anomaly situation perception method based on radar chart multidimensional assessment according to claim 1, characterized in that, The step of extracting features from the unified data input stream and constructing a multidimensional feature vector includes: The input stream is divided into several consecutive time segments using the sliding time window method. The window length is set to W. Then, at time t, the parameters within the window... The set of possible values ​​is: ； in, Indicates the first The parameters at time... The normalized and denoised values; Within each window, extract the mean feature, volatility feature, peak-to-valley difference feature, and energy entropy feature; The mean characteristic, volatility characteristic, peak-to-valley difference characteristic, and energy entropy characteristic are combined into a multi-dimensional feature vector: ； in, Indicates at time Within the corresponding window, there is a multidimensional feature set for each parameter.

5. The user-side load anomaly situation perception method based on radar chart multidimensional assessment according to claim 1, characterized in that, The formula for calculating the mean characteristic is: ； in, Indicates parameters In the window The average value within; The formula for calculating volatility characteristics is: ； like A large value indicates that the load is in an abnormal fluctuation state; Peak-valley difference characteristics are defined as follows: ； in, Indicates parameters The difference between the maximum and minimum values ​​within the window reflects the extreme variation in load during that time period; Energy entropy characteristics: Definition of normalized energy distribution: ； in, Indicates parameters The energy percentage at the k-th time within the window; Calculate energy entropy based on energy entropy characteristics: ； in, Indicates parameters Energy entropy within a window is used to measure the complexity of energy distribution.

6. The user-side load anomaly situation perception method based on radar chart multidimensional assessment according to claim 1, characterized in that, The feature reduction method is used to transform the multidimensional feature set into high-dimensional feature vectors. Mapping to a low-dimensional feature space: ； Among them, matrix For the dimension reduction mapping matrix, This represents a low-dimensional feature representation.

7. The user-side load anomaly situation perception method based on radar chart multidimensional assessment according to claim 1, characterized in that, An improved clustering algorithm is used to classify the situation feature vectors, distinguishing between normal and abnormal patterns. This is then combined with a dynamic threshold discrimination mechanism to form a multi-dimensional set of abnormal features. Let the set of eigenvectors be: ； in, Indicates at time The low-dimensional situational feature vector, where n represents the number of samples; Density clustering is performed on the feature vector set for any sample point. Its neighborhood density is defined as: ； in, Indicates sample neighborhood density, Represents the neighborhood radius. As an indicator function, points with lower neighborhood density correspond to abnormal loads; Based on the density clustering, fuzzy C-means is used to further distinguish the membership relationships of clusters, and its objective function is: ； in, Indicates sample For cluster center Membership degree, range of values And satisfy ;parameter >1 is the fuzzy weighting coefficient, used to control the degree of fuzziness; Iterative update and cluster center To obtain clustering results; A dynamic threshold mechanism based on time series statistical features is used to evaluate the feature sequence of a certain indicator. Calculate the mean within the window length 𝑊 and standard deviation The dynamic threshold is set as follows: ； in, This represents the dynamic threshold at time t. This is an adjustment coefficient; when the feature value of the sample exceeds... When this occurs, it is determined to be an abnormal state; Combining the clustering results with the dynamic threshold discrimination results, a multidimensional set of anomaly features is obtained: ； in, Indicates at time The set of features that are determined to be abnormal For the first The dynamic threshold of each feature, if the feature vector Falling into an isolated cluster is also considered an anomaly.

8. The user-side load anomaly situation perception method based on radar chart multidimensional assessment according to claim 1, characterized in that, Also includes: Map the abnormal features to the coordinate system of the radar image and calculate the area of ​​the radar image. Morphological differences With boundary volatility Maintenance personnel based on the area of ​​the radar chart Morphological differences With boundary volatility Based on the combined results, the severity of the abnormality is determined: The radar image coordinate mapping: assuming at a certain moment... The set of abnormal features contains One dimension: ； in, Indicates the first The numerical values ​​of each abnormal feature, and each feature Mapped onto the polar coordinate axis of the radar chart, the angle is: ； The radius is: ； in, Indicates the first The angle at which each feature is located. This represents the projection value of the feature onto the radar image. As the indicator weight, satisfying ; Calculating the area of ​​a radar image: The area of ​​a polygon in a radar image is defined as follows: ； in, Indicates at time The area of ​​the radar image, if Sometimes The larger the area, the more severe the multidimensional anomaly. Calculate morphological difference: The morphological difference index is defined as follows: ； in, Represents the average radius, if A larger value indicates that certain abnormal characteristics are more prominent, suggesting a risk of bias in the load condition. Calculating boundary volatility: Boundary volatility characterizes the dynamic changes in the shape of a radar chart, and is defined as follows: ； in, This represents the mean square error of the radar image boundary between adjacent time periods; a larger value indicates a more drastic evolution of the abnormal situation.

9. A user-side load anomaly situation awareness system based on radar chart multi-dimensional assessment, characterized in that, include: Data preprocessing module: Combines power operation parameters and external environmental parameters into a unified data input stream; The feature extraction module extracts features from the unified data input stream and constructs a situation vector of multi-dimensional indicators. The classification module classifies the situation feature vectors, distinguishes between normal and abnormal patterns, and forms a multi-dimensional abnormal feature set.

10. The user-side load anomaly situation awareness system based on radar chart multidimensional assessment according to claim 9, characterized in that, Also includes: The mapping module maps the set of abnormal features to the radar chart coordinate system, calculates the graphic area and shape differences and the severity of anomalies based on the weight of each indicator, and intuitively displays the dynamic changes in the load situation by combining the time series evolution.

Images