Power utilization safety event high-sensitivity detection method and system based on multiple classifiers
By collecting and preprocessing electricity consumption information data, extracting features and using multiple classifiers for detection, combining D-S evidence theory and decision-making cascade technology, the accuracy and intelligence of existing electricity consumption safety event detection methods are solved, and more efficient electricity consumption safety detection and rating are achieved.
Patent Information
- Application Number
- CN202411844939.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-16
- Publication Date
- 2025-05-23
AI Technical Summary
The existing power safety event detection methods have problems such as low detection accuracy, lack of generalization capabilities, poor real-time performance, insufficient intelligence, and lack of coordination and sharing mechanisms, which leads to difficulty in time and accurate judgment and early warning, and thus difficult to effectively ensure the power safety of low-voltage users.
By collecting low-voltage user edge side power consumption data, preprocessing and feature extraction, Fourier transform extract features from time series data, and forming high-dimensional feature vectors. Then, the data set is divided for model training, the results are aggregated using D-S evidence theory, and the power consumption safety event detection results are output through decision-making cascade.
It improves the accuracy, stability and intelligence of power safety detection, improves the safety perception ability of low-voltage users, and helps improve the safety and stability of the power grid.
Smart Images
Figure CN120030401A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of electricity safety detection, and in particular to a highly sensitive detection method and system for electricity safety events based on multiple classifiers. Background Art
[0002] The low-voltage user side of the distribution network contains a large number of electrical products of various types, which poses a huge challenge to the safe and reliable operation of the power system. In order to improve the level of intelligent identification of low-voltage user power safety hazards, illegal charging and power theft, and improve the accuracy of high-sensitivity detection and hazard rating of power safety events, a more sensitive power safety event detection method is needed.
[0003] However, some of the current electricity safety event detection methods have problems such as low detection accuracy, lack of generalization ability, poor real-time performance, insufficient intelligence, and lack of collaboration and sharing mechanisms. As a result, it is difficult for such methods to make timely and accurate judgments and warnings on electricity safety, and thus it is difficult to effectively ensure the electricity safety of low-voltage users.
[0004] To this end, the present invention improves data processing quality, optimizes feature engineering, selects a suitable machine learning model, deploys multiple classifiers, and uses result aggregation and decision cascade to judge whether a power failure occurs and the type of power safety, thereby improving the accuracy, stability and intelligence of power safety detection. Summary of the invention
[0005] In view of the above-mentioned problems, the present invention is proposed.
[0006] Therefore, the problem to be solved by the present invention is that the existing power safety event detection method has the problems of low detection accuracy, lack of generalization ability, poor real-time performance, insufficient intelligence, lack of coordination and sharing mechanism, etc.
[0007] To solve the above technical problems, the present invention provides the following technical solutions: a highly sensitive detection method for power safety events based on multiple classifiers, which includes collecting power consumption information data from the edge side of low-voltage users and preprocessing the data; extracting relevant features according to the characteristics of power safety events, and using Fourier transform to extract features from time series data to form a high-dimensional feature vector; dividing the data set for model training, using DS evidence theory to aggregate results, and outputting power safety event detection results through decision cascade.
[0008] As a preferred solution of the highly sensitive detection method of power safety events based on multiple classifiers described in the present invention, the preprocessing includes data cleaning and standardization, and the standardization formula is expressed as follows:
[0009]
[0010]
[0011] Among them, x i (i=1, 2, n) is the i-th sample data, Z i is the standardized value, μ is the sample mean, and σ is the sample standard deviation.
[0012] As a preferred solution of the highly sensitive detection method of power safety events based on multiple classifiers described in the present invention, wherein: the extracted related features include time domain and frequency domain features of voltage, current and power;
[0013] The Fourier transform is used to extract features from time series data to form a high-dimensional feature vector F(ω), which is expressed as follows:
[0014]
[0015] Among them, w represents frequency, t represents time, e -iwt It is expressed as a complex function; principal component analysis is applied to the high-dimensional feature vector to extract the principal component features that best represent the data variance.
[0016] As a preferred solution of the highly sensitive detection method for power safety incidents based on multiple classifiers described in the present invention, wherein: model training is performed based on the divided data set, including dividing the data set into a training set, a validation set and a test set; the training set is used to train model parameters, accounting for 70% of the total data set; the validation set is used to adjust model hyperparameters, accounting for 15% of the total data set; the test set is used to finally evaluate model performance, accounting for 15% of the total data set.
[0017] As a preferred solution of the highly sensitive detection method of power safety events based on multiple classifiers described in the present invention, wherein: the dividing of the data set for model training also includes training using a random forest algorithm, specifically including: determining information entropy, an indicator for measuring the quality of feature segmentation, the formula is expressed as,
[0018]
[0019] Among them, Y is the category set, p(y) is the probability of category y; determine the conditional information entropy, which is used to measure the quality indicator of the category probability after given feature segmentation, and the formula is expressed as,
[0020]
[0021] Determine the information gain, which is used to measure the reduction of information entropy due to feature segmentation. The formula is expressed as:
[0022] IG(X,Y)=H(Y)-H(Y|X)
[0023] Construct a forecasting performance metric based on the forecast error measurement model and test data.
[0024] As a preferred solution of the highly sensitive detection method for power safety events based on multiple classifiers described in the present invention, the method of aggregating results using the DS evidence theory includes converting the identification result of each classifier into a BPA. If the prediction result of a classifier is positive, it is converted into a BPA, in which the confidence of the positive class is 1 and the confidence of the negative class is 0; the BPAs from different classifiers are combined using the evidence combination rules of the DS evidence theory; if it is determined to be necessary, the aggregated BPA is subjected to evidence reduction to simplify the result.
[0025] As a preferred solution of the highly sensitive detection method of power safety incidents based on multiple classifiers described in the present invention, wherein: the output of power safety incident detection results through decision cascade includes: building a decision cascade model based on the XGBoost model, and inputting the aggregated output results of multiple classifiers into the XGBoost model as input features of the decision cascade model; using a data set to train the decision cascade model, with the goal of maximizing the accuracy of the final output result, training the decision cascade model on the training set, and evaluating the performance on the validation set; in the deployment stage, inputting new power safety features into the trained decision cascade model, the decision cascade model comprehensively considers the identification results of each classifier, and outputs the final power safety type prediction.
[0026] Another object of the present invention is to provide a highly sensitive detection system for power safety events based on multiple classifiers, which can detect power safety events of multiple classifiers.
[0027] To solve the above technical problems, the present invention provides the following technical solutions: a system for a highly sensitive detection method for power safety events based on multiple classifiers, comprising: a data acquisition module, a feature extraction module and a detection result output module; the data acquisition module collects power consumption information data on the edge side of low-voltage users and pre-processes the data; the feature extraction module extracts relevant features according to the characteristics of power safety events, and uses Fourier transform to extract features from time series data to form a high-dimensional feature vector; the detection result output module divides the data set for model training, uses DS evidence theory for result aggregation, and outputs power safety event detection results through decision cascade.
[0028] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for highly sensitive detection of power safety events based on multiple classifiers as described above are implemented.
[0029] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for highly sensitive detection of power safety events based on multiple classifiers as described above.
[0030] The beneficial effects of the present invention are as follows: the present invention extracts characteristic indicators related to power safety as input, selects machine learning models suitable for power safety issues and deploys multiple classifiers. The identification results of different classifiers are aggregated by using DS evidence theory, and the final power safety event type is comprehensively judged by decision cascade, thereby improving the accuracy of power safety event detection and type judgment, improving the power safety perception ability of low-voltage users, and helping to improve the safety and stability of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:
[0032] Figure 1 This is a flow chart of the highly sensitive detection method for power safety events based on multiple classifiers in Example 1.
[0033] Figure 2 This is a module structure diagram of the highly sensitive detection system for power safety events based on multiple classifiers in Example 1. DETAILED DESCRIPTION
[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.
[0035] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0036] Example 1, reference Figure 1 , which is the first embodiment of the present invention, and which provides a highly sensitive detection method for power safety events based on multiple classifiers, including: Figure 1 As shown:
[0037] S1. Collect power consumption information data of low-voltage users at the edge and pre-process the data.
[0038] Preprocessing includes data cleaning and standardization. The standardization formula is expressed as:
[0039]
[0040] Among them, x i (i=1, 2, n) is the i-th sample data, Z i is the standardized value, μ is the sample mean, and σ is the sample standard deviation.
[0041] S2. According to the characteristics of power safety events, relevant features are extracted, and Fourier transform is used to extract features from time series data to form a high-dimensional feature vector.
[0042] The relevant features extracted include time domain and frequency domain features of voltage, current and power.
[0043] Fourier transform is used to extract features from time series data to form a high-dimensional feature vector F(ω), which can be expressed as follows:
[0044]
[0045] Among them, w represents frequency, t represents time, e -iwt Expressed as a complex function.
[0046] The principal component analysis method is applied to the high-dimensional feature vector to extract the principal component features that best represent the data variance.
[0047] Principal component analysis is a method of reducing the dimension of original variables. Its basic principle is to reduce the number of related indicators that need to be studied.
[0048] Through principal component analysis, it is recombined into a smaller number of independent comprehensive indicators F N To replace the original index. The original index is comprehensively expressed so that the new variable can best represent the original variable X. p The information represented can ensure that the new indicators remain independent of each other and the data information they contain does not overlap.
[0049] Assume that there are n sample data, each of which has a common attribute variable, and form them into an n×p order sample data matrix:
[0050]
[0051] According to the basic knowledge in mathematical theory, the information content extracted by the principal components is measured by their variance. When the variance Var(F i ) is larger, F i The more information it contains, the more information it contains. Generally, they are arranged according to the size of their variance, so the first principal component F 1The amount of information contained is the largest, F 1 Yes [X 1 ,X 2 ,X 3 ,X 4 ,…,X p ] Among all the indicators that make up the linear equation of the comprehensive indicator, the one with the largest variance is F 1 Recorded as the first principal component.
[0052] Combining the safety characteristics of low-voltage electricity use, the most relevant features are selected. This patent focuses on the following aspects, as shown in Table 1:
[0053] Table 1. Feature selection for low voltage power safety event detection
[0054]
[0055] S3. Divide the data set for model training, use DS evidence theory to aggregate the results, and output the power safety event detection results through decision cascade.
[0056] The dataset is divided into training set, validation set and test set; the training set is used to train model parameters, accounting for 70% of the total dataset; the validation set is used to adjust model hyperparameters, accounting for 15% of the total dataset; the test set is used for the final evaluation of model performance, accounting for 15% of the total dataset.
[0057] The training using random forest algorithm specifically includes:
[0058] Determine the information entropy, which is used to measure the quality of feature segmentation. The formula is expressed as:
[0059]
[0060] Where Y is the set of categories, p(y) is the probability of category y;
[0061] Determine the conditional information entropy, which is used to measure the quality index of the category probability after given feature segmentation. The formula is expressed as:
[0062]
[0063] Determine the information gain, which is used to measure the reduction of information entropy due to feature segmentation. The formula is expressed as:
[0064] IG(X,Y)=H(Y)-H(Y|X)
[0065] Calculate the prediction error as a measure of the model's predictive performance on the test data.
[0066] The specific method of using DS evidence theory for result aggregation is to convert the identification result of each classifier into BPA. If the prediction result of a classifier is positive, it is converted into a BPA, where the confidence of the positive class is 1 and the confidence of the negative class is 0.
[0067] The evidence combination rules of DS evidence theory are used to combine BPAs from different classifiers. If it is determined to be necessary, evidence reduction is performed on the aggregated BPA to simplify the result.
[0068] DS evidence theory (Dempster-Shafer evidence theory) is an uncertain reasoning theory that can be used to deal with incompletely certain or ambiguous information. DS theory can handle the uncertainty information of multiple classifiers well and integrate it into the final decision result. The basic concept of DS evidence theory is basic probability assignment (BPA). BPA is a function that maps each possible state to a closed interval [0, 1]. The value of BPA represents the confidence of the state. DS evidence theory is mainly composed of identification framework, basic probability assignment function, trust function and likelihood function, as follows:
[0069] The identification framework Ω is an exhaustive set of all hypotheses of the problem, and all hypotheses are mutually exclusive. Suppose Ω contains N elements, Ω can be expressed as,
[0070] Ω={H 1 ,H 2 ,…,H N}
[0071] A subset of Ω is called a proposition, and the power set of Ω is 2 Ω It consists of all subsets of Ω, including 2 N elements, 2 Ω It can be expressed as,
[0072]
[0073] The basic probability distribution function is derived from 2 Ω The mapping from 0 to [0,1] is expressed as:
[0074] m:2 Ω →[0,1]
[0075] It meets the following two conditions:
[0076]
[0077] Among them, m(A) represents the degree of support of the evidence for proposition A.
[0078] The trust function is expressed as,
[0079]
[0080] Among them, Bel(A) represents the overall trust level of A, and B is a subset of A.
[0081] The likelihood function table is set as,
[0082]
[0083] in, The likelihood function represents the degree of confidence that A is not rejected.
[0084] Outputting the detection results of power safety events through decision cascade specifically includes selecting an XGBoost model as a decision cascade model, and inputting the aggregated output results of multiple classifiers into the XGBoost model as input features of the decision cascade model.
[0085] The decision cascade model is trained using the dataset, with the goal of maximizing the accuracy of the final output result. The decision cascade model is trained on the training set and the performance is evaluated on the validation set.
[0086] During the deployment phase, new electricity safety features are input into the trained decision cascade model, which comprehensively considers the identification results of each classifier and outputs the final electricity safety type prediction.
[0087] Example 2 is the second example of the present invention, which is different from the first example in that: the highly sensitive detection method for power safety events based on multiple classifiers also includes: in order to verify and illustrate the technical effects adopted in this method, this example adopts a traditional technical solution and the method of the present invention for comparative testing, and compares the test results by means of scientific argumentation to verify the actual effect of this method.
[0088] This patent uses power system simulation data to compare the detection effect of this method and traditional detection methods. The data set contains 10,000 power safety event samples, including leakage, overload, short circuit, fire and other types. The experimental data is shown in Table 2:
[0089] Table 2 Comparison between the patented electricity safety detection method and the traditional detection method
[0090]
[0091] As can be seen from Table 2, the power safety detection method based on multiple classifiers in this patent is superior to the traditional power safety detection method in all indicators except computing resource consumption, especially the accuracy, false alarm rate, detection timeliness and other indicators have improved significantly. This shows that the power safety detection method based on multiple classifiers can more effectively identify various types of power safety events, has higher sensitivity and stability, and can reduce the false alarm rate.
[0092] Example 3, reference Figure 2 , which is the third embodiment of the present invention, and is different from the previous two embodiments in that: a system for a high-sensitivity detection method for power safety events based on multiple classifiers, comprising a data acquisition module 100, a feature extraction module 200 and a detection result output module 300; the data acquisition module 100 collects power consumption information data of the edge side of low-voltage users and pre-processes the data; the feature extraction module 200 extracts relevant features according to the characteristics of power safety events, and uses Fourier transform to extract features from time series data to form a high-dimensional feature vector; the detection result output module 300 divides the data set for model training, uses DS evidence theory for result aggregation, and outputs the power safety event detection results through decision cascade.
[0093] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.
[0094] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.
[0095] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.
[0096] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0097] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A highly sensitive detection method for power safety events based on multiple classifiers, characterized by: Collect power consumption information data of low-voltage users at the edge and pre-process the data; According to the characteristics of power safety events, relevant features are extracted, and Fourier transform is used to extract features from time series data to form a high-dimensional feature vector; The data set is divided for model training, DS evidence theory is used for result aggregation, and the power safety event detection results are output through decision cascade.
2. The highly sensitive detection method for power safety events based on multiple classifiers according to claim 1, characterized in that: The preprocessing includes data cleaning and standardization. The standardization formula is expressed as: Among them, x i (i=1, 2, n) is the i-th sample data, Z i is the standardized value, μ is the sample mean, and σ is the sample standard deviation.
3. The highly sensitive detection method for power safety events based on multiple classifiers according to claim 2, characterized in that: The extraction-related features include time domain and frequency domain features of voltage, current and power; The Fourier transform is used to extract features from time series data to form a high-dimensional feature vector F(ω), which is expressed as follows: Among them, w represents frequency, t represents time, e -iwt Expressed as a complex function; The principal component analysis method is applied to the high-dimensional feature vector to extract the principal component features that best represent the data variance.
4. The highly sensitive detection method for power safety events based on multiple classifiers according to claim 3 is characterized in that: Performing model training based on the divided data set, including dividing the data set into a training set, a validation set, and a test set; The training set is used to train model parameters and accounts for 70% of the total data set; The validation set is used to adjust the model hyperparameters and accounts for 15% of the total dataset; The test set is used for the final evaluation of model performance and accounts for 15% of the total dataset.
5. The highly sensitive detection method for power safety events based on multiple classifiers according to claim 4 is characterized in that: The dividing of the data set for model training also includes using a random forest algorithm for training, specifically including: Determine the information entropy, which is used to measure the quality of feature segmentation. The formula is expressed as: Where Y is the set of categories, p(y) is the probability of category y; Determine the conditional information entropy, which is used to measure the quality index of the category probability after given feature segmentation. The formula is expressed as: Determine the information gain, which is used to measure the reduction of information entropy due to feature segmentation. The formula is expressed as: IG(X,Y)=H(Y)-H(Y|X) Construct a forecasting performance metric based on the forecast error measurement model and test data.
6. The highly sensitive detection method for power safety events based on multiple classifiers according to claim 5, characterized in that: The result aggregation using DS evidence theory includes converting the identification result of each classifier into a BPA. If the prediction result of a classifier is positive, it is converted into a BPA, where the confidence of the positive class is 1 and the confidence of the negative class is 0; Combine BPAs from different classifiers using evidence combination rules from DS evidence theory; If it is determined to be necessary, evidence reduction is performed on the aggregated BPA to simplify the result.
7. The highly sensitive detection method for power safety events based on multiple classifiers according to claim 6, characterized in that: Outputting the power safety event detection results through decision cascade includes building a decision cascade model based on the XGBoost model, and inputting the aggregated output results of multiple classifiers into the XGBoost model as input features of the decision cascade model; Use the dataset to train the decision cascade model, with the goal of maximizing the accuracy of the final output results. Train the decision cascade model on the training set and evaluate the performance on the validation set. During the deployment phase, new electricity safety features are input into the trained decision cascade model, which comprehensively considers the identification results of each classifier and outputs the final electricity safety type prediction.
8. A system using the highly sensitive detection method for power safety events based on multiple classifiers as claimed in any one of claims 1 to 7, characterized in that: It comprises a data acquisition module (100), a feature extraction module (200) and a detection result output module (300); The data collection module (100) collects power consumption information data of the edge side of low-voltage users and pre-processes the data; The feature extraction module (200) extracts relevant features according to the characteristics of the power safety event, and uses Fourier transform to extract features from the time series data to form a high-dimensional feature vector; The detection result output module (300) divides the data set for model training, uses DS evidence theory to aggregate results, and outputs power safety event detection results through decision cascade.
9. A computer device comprising a memory and a processor, which respectively implement data storage and processing, characterized in that: When the processor executes the computer program, the steps of the highly sensitive detection method for power safety events based on multiple classifiers described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the highly sensitive detection method for power safety events based on multiple classifiers described in any one of claims 1 to 7 are implemented.