Electricity load anomaly detection method and system based on multi-granularity fuzzy rough set

Abstract

The application discloses a power load anomaly detection method and system based on a multi-granularity fuzzy rough set, and relates to the technical field of power data analysis. The method comprises the following steps: obtaining a high-dimensional time sequence feature matrix from power load time sequence data to be detected by using a long short-term memory network; gathering the high-dimensional time sequence feature matrix into a plurality of multi-granularity particle balls by a multi-granularity particle ball generation method; calculating multi-granularity fuzzy relationships between the multi-granularity particle balls, and constructing a multi-granularity fuzzy rough set model; calculating the fuzzy rough density of each multi-granularity particle ball and the multi-granularity fuzzy entropy of each attribute based on the model; calculating the anomaly scores of the multi-granularity particle balls, mapping the anomaly scores of the multi-granularity particle balls to corresponding samples in the power load time sequence data, and judging the corresponding samples for abnormal loads based on the anomaly scores. The method can effectively utilize multi-granularity information of load data, enhance noise resistance, and improve the accuracy of anomaly detection by processing data uncertainty.

Description

Technical Field

[0001] This invention relates to the field of power data analysis technology, and more specifically, to a method and system for detecting abnormal power load based on multi-granularity fuzzy rough sets. Background Technology

[0002] Anomaly detection of electricity load is a crucial aspect of power system operation and management. It enables the timely detection of abnormal loads, which helps improve the stability and reliability of the power system, optimize energy distribution efficiency, and promote the development of smart grids and new energy integration. Effective anomaly detection methods are of great significance for ensuring power supply security, improving demand response capabilities, and supporting the optimization of the electricity market.

[0003] Existing methods for detecting abnormal power loads mainly include those based on traditional statistical analysis, those based on standard LSTM, and those based on threshold rules. Statistical analysis methods typically identify anomalies by calculating the mean, variance, or correlation of load data. However, this approach relies on the assumption that the data conforms to a specific distribution and has limited capability in processing non-stationary or nonlinear load data, especially prone to misjudgment in complex noise environments. Standard LSTM networks, based on fine-grained samples, detect abnormal loads by capturing long-term dependencies in time series. While offering advantages in time series modeling, their single-sample processing method makes it difficult to fully utilize multi-granularity data characteristics, and the model exhibits poor stability under noise interference. Threshold rule-based methods also rely on fine-grained samples, judging anomalies by setting pre-defined load thresholds. However, this method is highly dependent on experience in threshold selection, underutilizes multi-granularity information, and has limited noise resistance. Furthermore, these methods generally fail to consider uncertainties in load data, such as ambiguity, making it difficult to uncover potential anomalies, particularly when dealing with time-varying and complex load patterns, thus failing to meet the high-precision anomaly detection requirements of modern power systems.

[0004] Therefore, how to research and design a method for detecting abnormal power load that can overcome the above-mentioned defects has become a technical problem that urgently needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for detecting abnormal power load based on multi-granularity fuzzy rough sets. It combines time-series feature extraction, multi-granularity particle calculation and fuzzy rough set modeling, which can effectively utilize the multi-granularity information of load data, enhance noise resistance, and improve the accuracy of abnormal detection by processing the uncertainty of data.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0007] Firstly, this application provides a method for detecting abnormal power load based on multi-granularity fuzzy rough sets, including the following specific steps:

[0008] Long Short-Term Memory (LSTM) networks are used to extract features from the time-series data of the electricity load to be detected, resulting in a high-dimensional time-series feature matrix.

[0009] The high-dimensional temporal feature matrix is ​​aggregated into multiple multi-granularity spheres by a multi-granularity sphere generation method;

[0010] Calculate the multi-granularity fuzzy relationships between each multi-granularity particle, and construct a multi-granularity fuzzy rough set model based on the multi-granularity fuzzy relationships;

[0011] Based on the multi-granularity fuzzy rough set model, the fuzzy roughness density of each multi-granularity particle is calculated, and the multi-granularity fuzzy entropy of each attribute in the high-dimensional temporal feature matrix is ​​calculated.

[0012] Based on the multi-granularity fuzzy entropy of each attribute and the fuzzy roughness density of each multi-granularity particle, the anomaly score of each multi-granularity particle is calculated. The anomaly score of each multi-granularity particle is mapped to the corresponding sample in the electricity load time series data, and the abnormal load of the corresponding sample is judged based on the anomaly score.

[0013] Based on the above technical solution, the present invention can be further improved as follows.

[0014] Furthermore, the aforementioned high-dimensional time-series feature matrix was obtained in the following way:

[0015] The time series data of electricity load is input into the Long Short-Term Memory network;

[0016] The temporal dependencies of the load data are learned through the hidden layers of the Long Short-Term Memory network, and the corresponding high-dimensional temporal feature matrix is ​​generated.

[0017] Furthermore, the above-mentioned method of generating multi-granularity spheres aggregates the high-dimensional temporal feature matrix into multiple multi-granularity spheres, specifically as follows:

[0018] Normalization is performed on each data point within the high-dimensional time-series feature matrix. The rows within the matrix are treated as samples, and the columns as attributes. Each column is then processed using... k - The mean clustering algorithm generates multiple data clusters;

[0019] The center and radius of each data cluster were calculated and determined.

[0020] For each data point within a data cluster, multiple data points whose Euclidean distance from the center of the data cluster does not exceed the radius are identified as multi-granularity spheres.

[0021] Furthermore, the above calculation of the multi-granularity fuzzy relationships between various multi-granularity spheres, and the construction of a multi-granularity fuzzy rough set model based on these relationships, specifically:

[0022] Calculate the multi-granularity fuzzy relationship between any two multi-granularity spheres, and construct a multi-granularity fuzzy relationship matrix using the obtained multi-granularity fuzzy relationships;

[0023] Based on the multi-granularity fuzzy relation matrix, multiple multi-granularity spheres are transformed into multi-granularity fuzzy information particles, forming a family of granularity fuzzy information particles;

[0024] A multi-granularity fuzzy rough set model is formed by using a family of granularity fuzzy information particles, and the upper and lower approximate fuzzy sets of each multi-granularity fuzzy information particle in the family of granularity fuzzy information particles with respect to the multi-granularity fuzzy relations.

[0025] Furthermore, the calculation of the fuzzy roughness density of each multi-grained sphere is as follows:

[0026] ,in:

[0027] ;

[0028] In the formula, Indicates multi-sized particles Regarding arbitrary attribute subsets in a high-dimensional time-series feature matrix fuzzy roughness density, Indicates multi-sized particles density, Represents a subset of attributes The number of attributes in Indicates the number of multi-grained particles. Indicates in the attribute Below is about multi-size particles Multi-granularity fuzzy information particles, Indicates multi-sized particles With multi-sized particles Inter-attribute Fuzzy similarity under the following conditions Indicates in the attribute Multi-granularity fuzzy relationships.

[0029] Furthermore, the above calculation yields the multi-granularity fuzzy entropy of each attribute in the high-dimensional time-series feature matrix, specifically:

[0030] ;

[0031] In the formula, Represents a subset of attributes Multi-granularity fuzzy entropy, Indicates the number of multi-grained particles. Indicates in the attribute Below is about multi-size particles The cardinality of multi-granularity fuzzy information particles.

[0032] Furthermore, the above calculation of the anomaly score for each of the multi-grained particles is specifically as follows:

[0033] ,in:

[0034] ;

[0035] In the formula, Indicates multi-sized particles Abnormal scores, Indicates multi-sized particles Regarding the attributes in the high-dimensional time series feature matrix fuzzy roughness density, Represents the attributes in the high-dimensional time series feature matrix Multi-granularity fuzzy entropy, This represents the sum of the multi-granularity fuzzy entropies of each attribute in the high-dimensional time-series feature matrix. The attribute set representing the high-dimensional time series feature matrix Medium attributes The weight.

[0036] Secondly, this application provides a power load anomaly detection system based on multi-granularity fuzzy rough sets, applied to the power load anomaly detection method based on multi-granularity fuzzy rough sets according to any one of the first aspects, including:

[0037] The feature matrix extraction module is used to extract features from the time series data of the electricity load to be detected using a long short-term memory network, and obtain a high-dimensional time series feature matrix.

[0038] The granularity particle generation module is used to aggregate a high-dimensional temporal feature matrix into multiple multi-granularity particles using a multi-granularity particle generation method.

[0039] The rough set model construction module is used to calculate the multi-granularity fuzzy relationships between various multi-granularity particles and construct a multi-granularity fuzzy rough set model based on the multi-granularity fuzzy relationships.

[0040] The multi-parameter calculation module is used to calculate the fuzzy roughness density of each multi-granularity particle based on the multi-granularity fuzzy rough set model, and to calculate the multi-granularity fuzzy entropy of each attribute in the high-dimensional temporal feature matrix.

[0041] The anomaly score judgment module is used to calculate the anomaly score of each multi-granularity particle based on the multi-granularity fuzzy entropy of each attribute and the fuzzy roughness density of each multi-granularity particle, map the anomaly score of each multi-granularity particle to the corresponding sample in the electricity load time series data, and judge the abnormal load of the corresponding sample based on the anomaly score.

[0042] Thirdly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method of any one of the first aspects.

[0043] Fourthly, this application provides a non-transitory computer-readable storage medium that stores computer instructions that cause a computer to perform any of the methods in the first aspect.

[0044] Compared with the prior art, the present invention has at least the following beneficial effects:

[0045] 1. This invention provides a method for detecting abnormal power loads, combining LSTM for feature extraction, multi-granularity sphere computation, and fuzzy rough set modeling. This effectively handles non-stationary and complex load data. Multi-granularity sphere computation significantly enhances the clarity of feature hierarchy by decomposing load data into multiple sphere units, laying a solid foundation for anomaly detection. Compared to traditional methods, this invention overcomes the limitations of single fine-grained sample processing through multi-granularity sphere computation, enabling a more comprehensive capture of the diverse characteristics of load data, thereby greatly improving the applicability and reliability of anomaly detection.

[0046] 2. The synergistic effect of multi-granularity particle sphere computation and temporal feature extraction in this invention is significant: The combination of temporal features extracted by LSTM and multi-granularity representation of particles enables the extraction of richer information from multiple granular levels, thereby improving the accuracy of anomaly detection. Traditional methods often rely solely on single-granularity data analysis, ignoring the complex correlations of load data across different time scales and feature dimensions. In contrast, this invention, through the generation of multi-granularity particles, can simultaneously focus on both local and global characteristics of the data. Combined with LSTM's ability to model temporal dependencies, it further uncovers hidden anomaly patterns, making the detection results more targeted and accurate.

[0047] 3. This invention achieves significant improvements in robustness and accuracy: the noise resistance of multi-granularity sphere computation combined with the uncertainty handling capabilities of fuzzy rough sets makes the system more robust to high-noise and non-stationary load data; simultaneously, the innovative definitions of new multi-granularity fuzzy roughness density and multi-granularity fuzzy entropy further enhance detection accuracy. Compared to traditional methods, this invention effectively reduces noise interference in detection results through granularity processing of spheres, while the multi-granularity fuzzy rough set model can better handle the fuzziness and uncertainty in load data, reducing false positives and false negatives. Furthermore, the innovatively proposed multi-granularity fuzzy roughness density and multi-granularity fuzzy entropy provide a more scientific quantitative basis for anomaly detection, making the detection process more detailed and controllable.

[0048] 4. This invention excels in its flexible handling of complex loads: By decomposing complex load signals into multi-granularity spheres and analyzing their uncertainties using a fuzzy rough set model, this decomposition-modeling process allows the method to flexibly adapt to various types of abnormal load patterns. Traditional methods often fail when processing complex load signals due to information confounding, while this invention decomposes complex signals into units of different granularities through multi-granularity sphere calculations, reducing signal complexity. Furthermore, the fuzzy rough set model can deeply mine the potential abnormal features of each sphere unit, thereby achieving accurate detection of diverse load patterns and significantly improving the adaptability and practicality of the method in real-world applications.

[0049] 5. Optimization of computational efficiency in this invention: Multi-granularity sphere computation significantly reduces the computational load of subsequent processing by aggregating data into sphere units, overcoming the inefficiency problem caused by fine-grained sample processing in traditional methods. Traditional methods often require processing each sample individually when facing large-scale datasets, resulting in high computational complexity. In contrast, this invention, through sphere granulation, compresses the data volume into a smaller number of sphere units, significantly reducing the computational cost of anomaly detection and thus improving the real-time performance and scalability of the method.

[0050] 6. Innovative Breakthrough in Uncertainty Handling: By proposing a multi-granularity fuzzy rough set model, this invention effectively addresses the uncertainty in load data and uncovers deeper anomaly features. Traditional methods often ignore uncertain information in the data, resulting in incomplete detection results. This invention, however, models uncertainty using a multi-granularity fuzzy rough set model and dynamically adjusts attribute weights using multi-granularity fuzzy entropy, further improving the depth and breadth of anomaly detection and providing a novel solution for anomaly detection in complex power systems. Attached Figure Description

[0051] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0052] Figure 1 This is a flowchart of the detection method in an embodiment of the present invention;

[0053] Figure 2 This is a flowchart illustrating the electrical load anomaly detection method in an embodiment of the present invention;

[0054] Figure 3 This is a schematic diagram of the connection of the detection system in an embodiment of the present invention;

[0055] Figure 4 This is a schematic diagram of the connection of an electronic device in an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0057] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0058] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0059] In the description of the embodiments of the present invention, "multiple" means at least two.

[0060] Example 1: To address the problem that traditional detection methods generally fail to consider uncertainties in load data, such as fuzziness, making it difficult to uncover potential anomalies, especially when dealing with time-varying and complex load patterns, and thus failing to meet the high-precision anomaly detection requirements of modern power systems, this invention considers that actual load data typically exhibits time-varying, nonlinear, and uncertain characteristics. Traditional methods, using fine-grained samples as basic units, struggle to effectively utilize multi-granularity information and are sensitive to noise. Therefore, this invention first extracts the time-series features of load data using LSTM, then aggregates the data into multiple granular units using multi-granularity sphere computation, and finally processes the uncertainty of the data using a multi-granularity fuzzy rough set model to construct an anomaly detection framework, ultimately achieving efficient and accurate anomaly load detection. The power load anomaly detection method based on multi-granularity fuzzy rough sets provided in this example is as follows: Figure 1 and Figure 2 As shown, the specific steps include the following:

[0061] S1. Long Short-Term Memory (LSTM) network is used to extract features from the time series data of the electricity load to be detected, resulting in a high-dimensional time series feature matrix.

[0062] Optionally, the above high-dimensional time-series feature matrix is ​​obtained in the following way:

[0063] S11, input the time series data of electricity load into the long short-term memory network.

[0064] S12 learns the temporal dependencies of the load data through the hidden layers of the Long Short-Term Memory network and generates the corresponding high-dimensional temporal feature matrix.

[0065] Specifically, the time series data of the electricity load to be detected can be defined as follows: ,in The length of the data is given; this data is input into an LSTM, which extracts temporal features through its hidden layer update mechanism; the specific feature extraction process is based on the following formula:

[0066] Input gate The hidden state from the previous time step is processed through an activation function. and current input The combination of the Forgotten Gate The decision to forget the cell's state information from the previous moment is made through the output gate. The hidden state output, cell state, determines the current moment. Update via forget gate and input gate, hidden state. The output gate and the current cell state are jointly determined. Ultimately, the LSTM generates a high-dimensional temporal feature matrix through multiple time-step iterations. each of the rows Corresponding to time step eigenvectors, matrices The dimension is ( The number of hidden layer units. (where the time step is 1). This high-dimensional time series feature matrix provides a data foundation for the subsequent generation of multi-granularity spheres, and at the same time, the dynamic change characteristics of the load data are effectively captured through the gating mechanism of LSTM.

[0067] S2 aggregates the high-dimensional temporal feature matrix into multiple multi-granularity spheres using a multi-granularity sphere generation method.

[0068] Optionally, the high-dimensional temporal feature matrix is ​​aggregated into multiple multi-granularity spheres using the multi-granularity sphere generation method, specifically as follows:

[0069] S21, normalize the data within the high-dimensional time-series feature matrix. The behavior samples and columns within the high-dimensional time-series feature matrix are treated as attributes, and each column is processed through... k - The means clustering algorithm generates multiple data clusters.

[0070] S22, by calculating and determining the center and radius of each data cluster.

[0071] S23, for each data within a data cluster, multiple data points whose Euclidean distance from the center of the data cluster does not exceed the radius are identified as multi-granularity spheres.

[0072] Specifically, based on the high-dimensional temporal feature matrix obtained in step S1 As input, using k - The mean clustering method groups data into multiple spheres; the specific process can be: the high-dimensional feature matrix in step S1. An information system can be imported, where each row represents a sample and each column represents an attribute of the sample; the information system uses binary tuples. express, Denotes a non-empty finite sample set, This represents a non-empty finite set of attributes. Furthermore, to eliminate the influence of differences in dimensions and orders of magnitude on the analysis results, the min-max method is used to normalize the original data, and the range of values ​​for numerical data is adjusted to... For real numbers, nominal data remains unchanged, and the calculation formula is as follows:

[0073] ;

[0074] in, Represents the normalization function. Indicates sample x In attributes aThe value on the feature matrix. In information systems, the feature matrix... Each column Clustering algorithms are applied to generate several data clusters; each data cluster is then defined as a sphere, and the center of the sphere is determined by calculating the mean of the data within the cluster. The radius of the sphere is determined by the maximum distance within the cluster. Finally, the distance between the data point and the center of the sphere is calculated according to the Euclidean distance formula. If the distance is less than the radius, the point belongs to the sphere. The multi-granularity sphere calculation preserves the multi-granularity characteristics of the data, while reducing the computational complexity and improving the noise resistance.

[0075] S3 calculates the multi-granularity fuzzy relationship between each multi-granularity particle and constructs a multi-granularity fuzzy rough set model based on the multi-granularity fuzzy relationship.

[0076] Optionally, the above calculation of the multi-granularity fuzzy relationships between various multi-granularity spheres, and the construction of a multi-granularity fuzzy rough set model based on these relationships, specifically:

[0077] S31, calculate the multi-granularity fuzzy relationship between any two multi-granularity spheres, and construct a multi-granularity fuzzy relationship matrix using the obtained multi-granularity fuzzy relationships.

[0078] Among them, fuzzy relation refers to... A fuzzy set on For any and Membership function Representative object With object Having a relationship To what extent, Fuzzy Relationships Using a fuzzy matrix To represent any subset of attributes It can induce a fuzzy relationship The formula for calculating its membership degree is as follows:

[0079] ;

[0080] in, Indicates an adjustable parameter. , Indicates sample of Attribute values; calculated using formulas from preprocessed data. U Any two samples and fuzzy relations membership degree .

[0081] Specifically, based on the multi-granularity spheres generated in step S2, the multi-granularity fuzzy relation can be defined as follows: , It can be expressed as in The collection of particles generated on the surface, Represents a specific sphere; for any subset of attributes It can induce the generation of a multi-granularity fuzzy relation The membership degree is calculated using the following formula:

[0082] ;

[0083] in, Indicates granules The center of the sphere. The preprocessed data is calculated using a formula. Any two balls Multi-granularity fuzzy relationship membership degree The multi-granularity fuzzy relation matrix of each multi-granularity sphere is obtained. .

[0084] S32, based on the multi-granularity fuzzy relation matrix, transforms multiple multi-granularity spheres into multi-granularity fuzzy information particles, and forms a family of granularity fuzzy information particles.

[0085] S33, a multi-granularity fuzzy rough set model is formed by the granularity fuzzy information particle family, and the upper approximate fuzzy set and lower approximate fuzzy set of each multi-granularity fuzzy information particle in the granularity fuzzy information particle family with respect to the multi-granularity fuzzy relation.

[0086] Among them, multi-granularity fuzzy relations The granules can be collected Granulation into a family of multi-granularity fuzzy information particles It contains multiple multi-granularity fuzzy information particles , The cardinality expression is as follows: .

[0087] For any subset of attributes Multi-granularity fuzzy information Regarding multi-granularity fuzzy relations upper approximation and the lower approximation These are two important fuzzy sets, and their membership function expressions are as follows:

[0088] ;

[0089] Therefore, a new fuzzy rough set model is defined, called the multi-granularity fuzzy rough set model.

[0090] S4, based on the multi-granularity fuzzy rough set model, calculates the fuzzy roughness density of each multi-granularity particle and obtains the multi-granularity fuzzy entropy of each attribute in the high-dimensional temporal feature matrix.

[0091] Specifically, this embodiment proposes a novel multi-granularity fuzzy coarse density calculation method. By integrating the multi-granularity fuzzy relationships between particles, it defines the density value of each particle under multi-granularity attributes. The specific calculation is based on the coverage area of ​​the particle and the distribution characteristics of its internal data points to generate a density index for each particle. This method innovatively quantifies the density distribution of particles, laying the foundation for anomaly detection; that is, for any subset of attributes… and any ball , about The multi-grained fuzzy roughness density is as follows:

[0092] ,in:

[0093] ;

[0094] In the formula, Indicates multi-sized particles Regarding arbitrary attribute subsets in a high-dimensional time-series feature matrix fuzzy roughness density, Indicates multi-sized particles density, Represents a subset of attributes The number of attributes in Indicates the number of multi-grained particles. Indicates in the attribute Below is about multi-size particles Multi-granularity fuzzy information particles, Indicates multi-sized particles With multi-sized particles Inter-attribute Fuzzy similarity under the following conditions Indicates in the attribute Multi-granularity fuzzy relationships.

[0095] Furthermore, the multi-granularity fuzzy entropy provided in this embodiment is based on the uncertainty distribution of the sphere at different granularities, and the entropy value of each attribute is calculated by weighted averaging; then, attribute weights are assigned according to the entropy values, and the weights are applied to the fuzzy coarse density to generate a weighted fuzzy coarse density value. This innovative definition enhances the model's adaptability to uncertainty; that is, for any subset of attributes... and any ball Regarding the attributes in the high-dimensional time series feature matrix The multi-granularity fuzzy entropy is as follows:

[0096] ;

[0097] In the formula, Represents a subset of attributes Multi-granularity fuzzy entropy, Indicates the number of multi-grained particles. Indicates in the attribute Below is about multi-size particles The cardinality of multi-granularity fuzzy information particles; for , The larger the value, the more disordered its value range distribution, which means that it has greater uncertainty; based on the above idea, and combining multi-granularity fuzzy roughness density and multi-granularity fuzzy entropy, a weighted multi-granularity fuzzy roughness anomaly score can be constructed.

[0098] S5. Based on the multi-granularity fuzzy entropy of each attribute and the fuzzy roughness density of each multi-granularity particle, calculate the anomaly score of each multi-granularity particle, map the anomaly score of each multi-granularity particle to the corresponding sample in the electricity load time series data, and judge the abnormal load of the corresponding sample based on the anomaly score.

[0099] Specifically, the above calculation of the anomaly score for each multi-grain size particle is as follows:

[0100] ,in:

[0101] ;

[0102] In the formula, Indicates multi-sized particles Abnormal scores, Indicates multi-sized particles Regarding the attributes in the high-dimensional time series feature matrix fuzzy roughness density, Represents the attributes in the high-dimensional time series feature matrix Multi-granularity fuzzy entropy, This represents the sum of the multi-granularity fuzzy entropies of each attribute in the high-dimensional time-series feature matrix. The attribute set representing the high-dimensional time series feature matrix Medium attributes The weight.

[0103] Specifically, based on the correspondence between multi-granularity spheres and samples, the anomaly scores of multi-granularity spheres can be mapped to samples to obtain the anomaly score of each sample. Finally, by setting an anomaly judgment threshold, sample instances with anomaly scores greater than the threshold are judged as anomalies.

[0104] Specifically, the present invention will be further illustrated by the following examples:

[0105] 1) Given a high-dimensional temporal feature matrix extracted by LSTM, as shown in Table 1.

[0106] Table 1

[0107] U a1 a2 a3 a4 x1 3 120 3 280 x2 2 20 2 180 x3 2 180 5 150 x4 1 160 6 50 x5 2 130 4 80 x6 3 30 1 100 x7 1 40 3 200 x8 3 90 5 140

[0108] 2) Formation of multi-sized granules:

[0109] Perform column analysis on the high-dimensional time series feature matrix k -means clustering is used to generate multi-granularity spheres, and the resulting set of spheres can be... ,in Includes samples and , Includes samples and , Includes samples , Includes samples , and Further calculations revealed that the center of each sphere was... , , .

[0110] 3) Construction of multi-granularity fuzzy relation matrix:

[0111] Based on the set of particles generated in step 2), set As a parameter of multi-granularity fuzzy relations, it can be calculated using the formula mentioned above. Multi-granularity fuzzy relation matrix of each multi-granularity sphere:

[0112] , ,

[0113] , .

[0114] 4) Calculate multi-grained fuzzy roughness density:

[0115] Based on the multi-granularity fuzzy relation matrix calculated in step 3), the multi-granularity fuzzy roughness density is further calculated:

[0116] ;

[0117] ;

[0118] ;

[0119] ;

[0120] Furthermore, it can be calculated that... , , , ; , , , ; , , , .

[0121] Then perform multi-granularity fuzzy entropy calculation:

[0122] Based on the calculated multi-granularity fuzzy coarseness density, the following multi-granularity fuzzy entropy can be calculated:

[0123] ; ;

[0124] .

[0125] 5) Calculate the anomaly score:

[0126] Based on the calculated multi-granularity fuzzy coarseness density and multi-granularity fuzzy entropy, the anomaly score of each particle can be calculated, and then its anomaly score can be mapped to the internal samples; first, the anomaly score of each particle is calculated:

[0127] ;

[0128]

[0129] Furthermore, based on the samples contained in the sphere set, anomaly scores are assigned to the samples sequentially: Includes samples and , Includes samples and , Includes samples , Includes samples , and ;thus:

[0130] ;

[0131] ;

[0132] ;

[0133] .

[0134] An anomaly detection threshold can be set to 0.5, which indicates that... Therefore, the sample It was detected as abnormal.

[0135] In summary, by extracting the temporal features of the load data using LSTM and decomposing the complex load data into multiple sphere units using multi-granularity sphere computation, the clarity and noise resistance of the feature hierarchy are significantly enhanced. On this basis, the multi-granularity fuzzy rough set model further mines potential abnormal features by handling the uncertainty of the data, thereby improving the accuracy and robustness of anomaly detection.

[0136] Example 2: This application provides an electricity load anomaly detection system based on multi-granularity fuzzy rough sets, applied to the electricity load anomaly detection method based on multi-granularity fuzzy rough sets in Example 1, such as... Figure 3 As shown, it includes:

[0137] The feature matrix extraction module is used to extract features from the time series data of the electricity load to be detected using a long short-term memory network, and obtain a high-dimensional time series feature matrix.

[0138] The granularity particle generation module is used to aggregate a high-dimensional temporal feature matrix into multiple multi-granularity particles using a multi-granularity particle generation method.

[0139] The rough set model construction module is used to calculate the multi-granularity fuzzy relationships between various multi-granularity particles and construct a multi-granularity fuzzy rough set model based on the multi-granularity fuzzy relationships.

[0140] The multi-parameter calculation module is used to calculate the fuzzy roughness density of each multi-granularity particle based on the multi-granularity fuzzy rough set model, and to calculate the multi-granularity fuzzy entropy of each attribute in the high-dimensional temporal feature matrix.

[0141] The anomaly score judgment module is used to calculate the anomaly score of each multi-granularity particle based on the multi-granularity fuzzy entropy of each attribute and the fuzzy roughness density of each multi-granularity particle, map the anomaly score of each multi-granularity particle to the corresponding sample in the electricity load time series data, and judge the abnormal load of the corresponding sample based on the anomaly score.

[0142] Example 3: This application provides an electronic device, such as... Figure 4As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method of Embodiment 1.

[0143] Example 4: This application provides a non-transitory computer-readable storage medium that stores computer instructions that cause a computer to execute the method of Example 1.

[0144] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0145] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0146] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0147] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0148] Those skilled in the art will understand that all or part of the steps in the above facts and methods can be implemented by a program instructing related hardware. The program or the program described therein can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: at this time, the corresponding method steps are introduced. The storage medium can be ROM / RAM, magnetic disk, optical disk, etc.

[0149] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting abnormal power load based on multi-granularity fuzzy rough sets, characterized in that, The specific steps include the following: Long Short-Term Memory (LSTM) networks are used to extract features from the time-series data of the electricity load to be detected, resulting in a high-dimensional time-series feature matrix. The high-dimensional temporal feature matrix is ​​aggregated into multiple multi-granularity spheres using a multi-granularity sphere generation method; Calculate the multi-granularity fuzzy relationship between each of the multi-granularity particles, and construct a multi-granularity fuzzy rough set model based on the multi-granularity fuzzy relationship; Based on the multi-granularity fuzzy rough set model, the fuzzy roughness density of each multi-granularity particle is calculated, and the multi-granularity fuzzy entropy of each attribute in the high-dimensional temporal feature matrix is ​​calculated. Based on the multi-granularity fuzzy entropy of each attribute and the fuzzy roughness density of each multi-granularity particle, the anomaly score of each multi-granularity particle is calculated, the anomaly score of each multi-granularity particle is mapped to the corresponding sample in the electricity load time series data, and the abnormal load of the corresponding sample is judged based on the anomaly score.

2. The method for detecting abnormal power load based on multi-granularity fuzzy rough sets according to claim 1, characterized in that, The high-dimensional time-series feature matrix is ​​obtained through the following method: The electricity load time series data is input into the long short-time memory network; The hidden layers of the Long Short-Term Memory network learn the temporal dependencies of the load data and generate the corresponding high-dimensional temporal feature matrix.

3. The method for detecting abnormal power load based on multi-granularity fuzzy rough sets according to claim 1, characterized in that, The high-dimensional temporal feature matrix is ​​aggregated into multiple multi-granularity spheres using a multi-granularity sphere generation method, specifically as follows: The data within the high-dimensional time-series feature matrix are normalized. The rows in the high-dimensional time-series feature matrix are considered as samples, and the columns as attributes. Each column is then processed through... k - The mean clustering algorithm generates multiple data clusters; The center and radius of each data cluster were calculated and determined. For each data point within each data cluster, multiple data points whose Euclidean distance from the center of the data cluster does not exceed the radius are identified as multi-grained spheres.

4. The method for detecting abnormal power load based on multi-granularity fuzzy rough sets according to claim 1, characterized in that, The calculation of the multi-granularity fuzzy relationships between the various multi-granularity spheres, and the construction of a multi-granularity fuzzy rough set model based on these relationships, specifically involves: Calculate the multi-granularity fuzzy relationship between any two multi-granularity spheres, and construct a multi-granularity fuzzy relationship matrix using the obtained multi-granularity fuzzy relationships; Based on the multi-granularity fuzzy relation matrix, multiple multi-granularity spheres are transformed into multi-granularity fuzzy information particles, forming a family of granularity fuzzy information particles; A multi-granularity fuzzy rough set model is formed by the aforementioned granularity fuzzy information particle family, and the upper approximate fuzzy set and lower approximate fuzzy set of each multi-granularity fuzzy information particle in the granularity fuzzy information particle family with respect to the multi-granularity fuzzy relation.

5. The method for detecting abnormal power load based on multi-granularity fuzzy rough sets according to claim 3, characterized in that, The calculation of the fuzzy roughness density of each multi-grained sphere is specifically as follows: ,in: ； In the formula, Indicates multi-sized particles Regarding arbitrary attribute subsets in a high-dimensional time-series feature matrix fuzzy roughness density, Indicates multi-sized particles density, Represents a subset of attributes The number of attributes in Indicates the number of multi-grained particles. Indicates in the attribute Below is about multi-size particles Multi-granularity fuzzy information particles, Indicates multi-sized particles With multi-sized particles Inter-attribute Fuzzy similarity under the following conditions Indicates in the attribute Multi-granularity fuzzy relationships.

6. The method for detecting abnormal power load based on multi-granularity fuzzy rough sets according to claim 1, characterized in that, The calculation yields the multi-granularity fuzzy entropy of each attribute in the high-dimensional temporal feature matrix, specifically as follows: ； In the formula, Represents a subset of attributes Multi-granularity fuzzy entropy, Indicates the number of multi-grained particles. Indicates in the attribute Below is about multi-size particles The cardinality of multi-granularity fuzzy information particles.

7. The method for detecting abnormal power load based on multi-granularity fuzzy rough sets according to claim 1, characterized in that, The calculation of the anomaly score for each of the multi-grained particles is specifically as follows: ,in: ； In the formula, Indicates multi-sized particles Abnormal scores, Indicates multi-sized particles Regarding the attributes in the high-dimensional time series feature matrix fuzzy roughness density, Represents the attributes in the high-dimensional time series feature matrix Multi-granularity fuzzy entropy, This represents the sum of the multi-granularity fuzzy entropies of each attribute in the high-dimensional time-series feature matrix. The attribute set representing the high-dimensional time series feature matrix Medium attributes The weight.

8. A power load anomaly detection system based on multi-granularity fuzzy rough sets, characterized in that, include: The feature matrix extraction module is used to extract features from the time series data of the electricity load to be detected using a long short-term memory network, and obtain a high-dimensional time series feature matrix. The granularity particle generation module is used to aggregate the high-dimensional temporal feature matrix into multiple multi-granularity particles using a multi-granularity particle generation method. The rough set model construction module is used to calculate the multi-granularity fuzzy relationship between each of the multi-granularity particles, and to construct a multi-granularity fuzzy rough set model based on the multi-granularity fuzzy relationship; The multi-parameter calculation module is used to calculate the fuzzy roughness density of each multi-granularity particle based on the multi-granularity fuzzy rough set model, and to calculate the multi-granularity fuzzy entropy of each attribute in the high-dimensional temporal feature matrix. The anomaly score judgment module is used to calculate the anomaly score of each multi-granularity particle based on the multi-granularity fuzzy entropy of each attribute and the fuzzy roughness density of each multi-granularity particle, map the anomaly score of each multi-granularity particle to the corresponding sample in the electricity load time series data, and judge the abnormal load of the corresponding sample based on the anomaly score.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the method of any one of claims 1-7.

Images