CN112821559B - Non-invasive household appliance load depth re-identification method - Google Patents
Non-invasive household appliance load depth re-identification method Download PDFInfo
- Publication number
- CN112821559B CN112821559B CN202110089840.XA CN202110089840A CN112821559B CN 112821559 B CN112821559 B CN 112821559B CN 202110089840 A CN202110089840 A CN 202110089840A CN 112821559 B CN112821559 B CN 112821559B
- Authority
- CN
- China
- Prior art keywords
- sample
- formula
- value
- identification
- load
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
- H02J13/00002—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by monitoring
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2310/00—The network for supplying or distributing electric power characterised by its spatial reach or by the load
- H02J2310/10—The network having a local or delimited stationary reach
- H02J2310/12—The local stationary network supplying a household or a building
- H02J2310/14—The load or loads being home appliances
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02B—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO BUILDINGS, e.g. HOUSING, HOUSE APPLIANCES OR RELATED END-USER APPLICATIONS
- Y02B70/00—Technologies for an efficient end-user side electric power management and consumption
- Y02B70/30—Systems integrating technologies related to power network operation and communication or information technologies for improving the carbon footprint of the management of residential or tertiary loads, i.e. smart grids as climate change mitigation technology in the buildings sector, including also the last stages of power distribution and the control, monitoring or operating management systems at local level
- Y02B70/3225—Demand response systems, e.g. load shedding, peak shaving
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S20/00—Management or operation of end-user stationary applications or the last stages of power distribution; Controlling, monitoring or operating thereof
- Y04S20/20—End-user application control systems
- Y04S20/222—Demand response systems, e.g. load shedding, peak shaving
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Power Engineering (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a non-invasive household appliance load depth re-identification method, which is implemented according to the following steps: denoising the acquired user high frequency data; carrying out event detection on the data; extracting multidimensional load characteristics from the detected event variable points; screening relevant features aiming at the extracted multidimensional features; and taking the obtained characteristics as load marks, identifying the working state of the internal household appliance load of the user by using the maximum membership degree through establishing a load characteristic library, training the parameter smoothness factors of the GRNN network, and carrying out deep re-identification of the sample to be tested by using the trained GRNN network to finish the final non-invasive household appliance load identification. The problem that the load identification accuracy is low due to a single algorithm in the prior art is solved.
Description
Technical Field
The invention belongs to the technical field of household appliance load identification, and relates to a non-invasive household appliance load deep re-identification method.
Background
The load monitoring technology is one of basic technologies for constructing ubiquitous power Internet of things and transparent power grids, and can help users to save up to 14% of electric energy expenditure by knowing the electricity utilization rule, but the non-invasive load monitoring (NILM) technology is an object of attention of power grid service providers and users due to the low cost.
The non-invasive load monitoring is to install monitoring equipment at a power inlet, obtain the type and the running condition of a single load in a load cluster by monitoring voltage, current and other signal analysis, and the non-invasive household appliance load identification is a non-invasive load monitoring technology facing to a user side, and the process is concentrated in three steps: event detection, feature extraction and load identification. In the aspect of event detection, the generalized likelihood estimation (GLR) mainly considers the mean value, so that misjudgment is easy to occur, and the F test can reflect the sample variance; in the aspect of feature extraction, the type of the feature is usually determined subjectively according to experience, the existing information is underutilized, and the known label information can be fully utilized for quantifying the feature by the filtering feature selection of the combination of the semi-supervised Relief-F and the maximum correlation and the minimum redundancy (mRmR); in addition, the prior art has the problem that the load identification accuracy is low due to power overlapping and a single algorithm.
Disclosure of Invention
The invention aims to provide a non-invasive household appliance load depth re-identification method, which solves the problem of low load identification accuracy caused by a single algorithm in the prior art.
The technical scheme adopted by the invention is that a non-invasive household appliance load depth re-identification method is implemented according to the following steps:
step 1, acquiring user high-frequency data, and denoising the acquired user high-frequency data;
step 2, carrying out event detection on the data in the step 1 through improved generalized likelihood ratio detection, if an event is detected, executing the step 3, otherwise returning to the step 1;
step 3, extracting multidimensional load characteristics from the detected event variable points;
step 4, screening relevant features by using a semi-supervised algorithm combining the Relief-F and the mRmR aiming at the multidimensional features extracted in the step 3;
step 5, taking the characteristics obtained in the step 4 as load marks, establishing a load characteristic library through a self-adaptive FCM algorithm, identifying the load working state of the household appliance in the user by using the maximum membership degree, and if the identification results of the two FCM algorithms are consistent, ending, and if the identification results are inconsistent, executing the step 6;
and 6, training the parameter smoothing factors of the GRNN network by adopting an SA-BAS (simulated annealing-longhorn beetle whisker) algorithm, and performing deep re-identification of the sample to be tested by using the trained GRNN network to finish final identification of the non-invasive household appliance load.
The invention is also characterized in that:
the step 1 is specifically implemented according to the following steps:
step 1.1, acquiring high-frequency household appliance load data containing electrical parameters, wherein the high-frequency household appliance load data comprise voltage, current and corresponding power;
step 1.2, denoising processing of the power signal, wherein isolated noise points are easy to be mistakenly identified as events by an event detection algorithm, so that a median filtering method is selected to process the original power signal, and edge information is not changed while noise is eliminated: suppose there is a digital signal sequence x j (-∞<j<+ -infinity), when it is subjected to a filtering process, a window of odd length L is first defined, l=2n+1, n being a positive integer, assuming that at a certain instant i, the signal samples within the window are x i-N ,…,x i ,…,x i+N Wherein x is i Is the signal sample value at the center of the window, and after rearranging the L signals from small to large, the value is defined as the output value of the median filter.
The step 2 is specifically implemented according to the following steps:
step 2.1, calculating fundamental wave active power P according to the step (1) 1 The combined active power P is used as a two-dimensional power time sequenceAccording to formula (2), binary hypothesis testing is proposed;
wherein V is 1 Is the fundamental voltage, I 1 As a fundamental current, a current is supplied,a phase difference between the two. n is n c For the time of occurrence of the change point, k is the total length of the window, n is the last sample time in the window, μ 0 ,∑ 0 To hypothesis test H 0 Gaussian distribution mean, covariance matrix, mu under the condition a ,∑ a Is H 1 Multi-dimensional signal mean value and multi-dimensional covariance matrix, mu before occurrence of change point under condition b ,∑ b Is H 1 A multidimensional signal mean value and a multidimensional covariance matrix after the occurrence of the variable points under the condition;
step 2.2 defining two consecutive windows W in the time sequence a And W is b Samples in two windows are X n ＝{x m M=n-k+1, …, n }, both window lengths are k/2, μ and Σ in both windows are calculated according to equation (3) and equation (4), respectively, and then the decision function g is calculated according to equation (5) n ；
Step 2.3, g n And threshold h 1 Comparing, searching for a suspicious point of occurrence of the event: when the decision function value is greater than h 1 When rejecting H 0 The data distribution in the two windows is inconsistent, and the time n is the variable point time c There is a possibility of an event occurring; when the decision function is less than h 1 When rejecting H 1 Number of two windowsAccording to the distribution, no event occurs. And when the sampling points of the detected events are continuous or spaced within 3, it is regarded as having only the maximum g n The corresponding event occurs;
and 2.4, taking the event point as a base point, and performing F test screening on the event of false detection. First, assume that the variance values of window data before and after the base point are equal:calculating the value of the statistic F according to formula (6), given the significance level α, rejecting the hypothesis H if the value of F satisfies formula (7) 0 Considering that the two parent variances have significant differences, and judging that an event occurs at the point;
in the method, in the process of the invention,the sample capacity and variance of the windows before and after the base point, respectively.
Step 3 is specifically implemented according to the following steps: the power characteristics of the variable points are extracted, and the method specifically comprises the following steps: active power, fundamental active power, reactive power, fundamental reactive power, apparent power, distortion power, power factor angle, fundamental power factor; extracting harmonic characteristics at the variable points, wherein the harmonic characteristics comprise voltage, amplitude of each subharmonic of one to nine times of voltage, content of each subharmonic, difference of content of each subharmonic and total harmonic distortion; the current waveform characteristics comprise a wave peak value, an average value and a wave crest coefficient; extracting the V-I track characteristics at the variable points, wherein the method specifically comprises the following steps: symmetry, wrapping direction, wrapping area, number of intersections, Y-axis intercept, Y-axis span, centerline curvature, track mid-portion peak, left and right portion area, mid-portion shape, instantaneous admittance standard deviation.
Step 4 is specifically implemented according to the following steps:
step 4.1, labeling part of the multidimensional feature sample as a labeled sample S 1 The remainder is marked as unlabeled sample S 2 The label is C, and the multidimensional feature set is a= (a) 1 ,A 2 ,…,A N )；
Step 4.2, slave S 1 Randomly extracting a sample s of the class C q (C q E C); from S 2 D neighbor samples are selected and marked asAt S 1 Middle-slave division C q Each other class C p E, in C (p is not equal to q), respectively solving a nearest neighbor sample x of s; and at S 2 D neighbor samples of x are solved, denoted +.>Wherein the neighbor formula is shown as formula (8); updating all feature weights based on equation (9);
wherein M represents the iteration times, d is the number of neighbor samples,for the t nearest neighbor sample in class q to which sample s belongs,>representing the p-th class different from the sample s classT-th neighbor sample in (C), P (C) p ) Representing the probability of a class p object, A k For the kth feature>Representing sample s and sample->With respect to feature A k Is a distance of (2);
step 4.3, circularly executing the step 4.2 for M times, and obtaining the final output characteristic weight omega of the semi-supervision Relief-F k 'the features with weight coefficients smaller than theta are removed from the features (k=1, 2, … and w'), a candidate feature subset is obtained, and semi-supervised mRmR feature selection is further carried out on the candidate feature subset;
step 4.4, in the labeled sample S according to equation (11) 1 The correlation degree of each feature and the sample label is calculated, and the sample S is not marked according to the formula (12) 2 Calculating redundancy of each feature;
in the method, in the process of the invention,is shown in labeled sample S 1 Middle calculation A k Mutual information with tag C, R (A k ,S m-1 ) Representing an existing feature subset S m-1 Comprising m-1 features, not in S m-1 Feature A in the subset k Redundancy with selected features;
step 4.5, establishing a feature candidate set H, and selecting the maximum correlation degree D max Corresponding features as candidate set leader H 1 Sequentially selecting the kth feature A according to equation (13) k Placing the specific characteristics into H until the specific characteristic number w is selected;
and 4.6, calculating the weight of each characteristic according to the formula (14).
Step 5 is specifically implemented according to the following steps:
step 5.1, setting an initial cluster number b, determining an initial cluster center based on a maximum and minimum distance algorithm, and resetting a counter k to zero;
a. calculating an average value of a data setThe sample point farthest from the average value is recorded as V 1j (j=1, 2, …, w); b. calculating the minimum distance D of each data point from the selected cluster center according to the formula (15) x Select D x The maximum point is used as a new clustering center; c. repeating the step b until b initial clustering centers are selected;
D x ＝mind(x i ,Z' k )k'＝1,…,kselected (15)
step 5.2, constructing a loss function L of the FCM1 algorithm 1 As shown in formula (17), for u ij ，λ j And v ik Combining the partial derivatives and performing u by using the derived formula (18) and formula (19) ij And v ik Is updated iteratively;
wherein n is the number of samples, b is the number of clusters, w is the number of features,for the sample weight, u ij The j-th sample belongs to the membership degree of the i-th class, v is a clustering center matrix, T u E (0, infinity) replaces the fuzzy index m, is used for controlling the entropy value, and introduces the maximum entropy as regularization;
step 5.3, calculate J 1 If the value of (1)The clustering of the initial clustering number b is completed, iteration is terminated, otherwise, k=k+1 is transferred to step 5.2 to continue iteration until the requirement is met;
step 5.4, calculating the cluster effectiveness index of the initial cluster number b, namely the contour coefficient S according to the formula (20) b If it meets S b-2 ＜S b-1 And S is b-1 ＞S b The self-adaptive FCM1 classification is finished, otherwise, whether the cluster number b reaches the maximum value is judgedIf the maximum value is reached, taking all S b B=b+1 if the maximum value of b, u, v is not reached, and the step 5.1 is executed again;
wherein n is the number of samples, a (i) is the average distance between the sample point i and the remaining sample objects in the same cluster, b (i) is the minimum value of the average distance between the sample point i and the remaining sample objects in each cluster, S b At [ -1,1]The greater the S value, the moreThe better the class effect;
step 5.5, setting the optimal cluster number b of the FCM1 as the cluster number of the FCM2, determining an initial cluster center by adopting a maximum and minimum distance algorithm, and resetting a counter k;
step 5.6, constructing a loss function J of the FCM2 algorithm 2 As shown in formula (21), for v ik Obtaining bias derivative, and v is carried out by using derived formula (22) ik Is updated by iteration of the method, and the sampling projection gradient descent method is used for u ij Iteration is performed: u (u) (k+1) ＝argmin:J 2 (u,v (k) )；
Wherein lambda is equal to or greater than 0 and is an orthogonal constraint parameter, and the second function term is to add an approximately orthogonal constraint to the membership matrix to balance the influence of unbalanced data, wherein u ip Representing the membership degree of the ith sample point belonging to the class cluster p, which is obtained by the original membership degree function u ip ' pass throughTransformed into the material;
step 5.7, calculate J 2 If the value of (1)The FCM2 clustering is completed and the iteration is terminated, otherwise, k=k+1, go to step 5.6 and continue the iteration until +.>The clustering number and the clustering center of the FCM algorithm in the load feature library are determined;
and 5.8, inputting samples to be detected, determining an initial clustering center by a load feature library, obtaining a membership function through FCM clustering, taking membership vectors corresponding to the samples to be identified, outputting the membership vectors, sequencing the membership, taking the category corresponding to the maximum membership, and marking the category as the category of the load to be identified. If the sample identification results of the FCM1 and the FCM2 are consistent, the algorithm identification is finished, the category corresponding to the maximum membership degree is the final sample identification result, and if the sample identification results are inconsistent, the step 6 is executed.
Step 6 is specifically implemented according to the following steps:
step 6.1, combining samples with consistent FCM identification results, and training an improved GRNN neural network as a training set;
step 6.2, setting parameters: the number of neurons of the input layer is the characteristic dimension w; the mode layer neuron is the sample number n of the sample set of the training set, and the transfer function of the ith neuron is shown as a formula (23); the summation layer comprises two types of neurons, one is to carry out arithmetic summation on the output of all the mode layer neurons, the transfer function of the two types of neurons is shown as a formula (24), and the other is to carry out weighted summation on the output of all the mode layer neurons, and the transfer function of the two types of neurons is shown as a formula (25); the number of the neurons of the output layer is equal to the dimension of the output vector in the training sample, namely the clustered electrical appliance class number b, and the network output of the output layer is divided by two types of neurons, as shown in a formula (26);
wherein X is an input sample, X i For the ith sample data of the training set, σ is the smoothing factor,Y ij for the ith output sample Y i The j-th element of (a);
step 6.3, the network parameters of GRNN are only smooth factors, so we use SA-BAS algorithm to find the optimal smooth factor sigma; firstly, training data are input, and parameters are initialized: longicorn initial position X 0 =1, temperature t=200, step factor α=0.95, maximum number of iterations n=200, annealing cycle number l=100, centroid to tentacle distance d=1.5, counter h is normalized to 1;
step 6.4, setting the step length S=T of the longicorn, and simultaneously setting the counter T to be 1;
step 6.5, updating the left whisker position and the right whisker position of the longicorn according to the formula (27), and updating the longicorn position X according to the formula (28) t+1 Calculating probability according to Metropolis criterion, judging whether X is accepted t+1 As a new solution, the step S is updated as shown in equation (29) t ＝αS t-1 Updating the distance from the centroid to the tentacle according to the formula (30), judging whether the counter t is more than or equal to the annealing cycle times L, if so, executing the step 6.6, otherwise, executing the step 6.5 in a circulating way, wherein t=t+1;
in the method, in the process of the invention,taking the direction of right whisker pointing to left whisker as the space of longicorn as the random unit vectorSign () is a sign function, the value is greater than zero, 1 is taken, less than zero, 1 is taken, equal to zero, 0 is taken, T is the current temperature, Δt=f (X t+1 )-f(X t )，Omega is an expression of the fitness function j As characteristic weight, x ij Output value for training sample network, +.>Expected output values for training samples;
and 6.6, updating the self-adaptive factor beta, T=beta T according to a formula (31), judging whether the counter h is larger than or equal to the iteration times N, if so, outputting the optimal longicorn position as an optimal smoothing factor, and ending training, otherwise, executing the step 6.4.
Wherein f h For the current fitness value, f min The historical optimal fitness value is used;
and (3) constructing a GRNN network by combining the trained smoothing factors with the conditions of the step 6.2, so that the deep re-identification of the sample to be detected is performed, and the final non-invasive household appliance load identification is completed.
The beneficial effects of the invention are as follows: the invention discloses a non-invasive home appliance load depth re-identification method, which solves the problem of low load identification accuracy caused by a single algorithm in the prior art. The event occurrence time is positioned through an event detection algorithm combining GLR and F detection, the characteristic with high correlation with an electric tag is screened out by using a semi-supervised algorithm combining Relief-F and mRmR, primary discrimination is carried out by combining the self-adaptive FCM1 and FCM2, secondary discrimination is carried out by using the improved GRNN, misdiscrimination and feature redundancy are reduced, and the method has the advantages of high recognition accuracy and high recognition rate.
Drawings
FIG. 1 is a flow chart of a non-invasive appliance load depth re-identification method of the present invention;
FIG. 2 is a flow chart of the GLR and F test based event detection for a non-invasive appliance load depth re-identification method according to the present invention;
FIG. 3 is a flow chart of feature selection based on semi-supervised Relief-F and mRmr for a non-invasive appliance load depth re-identification method of the present invention;
FIG. 4 is a load identification flow chart of a non-invasive home appliance load depth re-identification method based on one-time identification of the adaptive FCM1 algorithm according to the present invention;
FIG. 5 is a load identification flow chart of a non-invasive home appliance load depth re-identification method based on one-time identification of the adaptive FCM2 algorithm;
FIG. 6 is a load identification flow chart of a non-invasive home appliance load depth re-identification method based on the secondary discrimination of the improved GRNN neural network.
Detailed Description
The invention will be described in detail below with reference to the drawings and the detailed description.
The invention discloses a non-invasive household appliance load depth re-identification method, which is implemented according to the following steps as shown in fig. 1:
step 1, acquiring user high-frequency data, and denoising the acquired user high-frequency data;
the step 1 is specifically implemented according to the following steps:
step 1.1, acquiring high-frequency household appliance load data containing electrical parameters, wherein the high-frequency household appliance load data comprise voltage, current and corresponding power;
step 1.2, denoising processing of the power signal, wherein isolated noise points are easy to be mistakenly identified as events by an event detection algorithm, so that a median filtering method is selected to process the original power signal, and edge information is not changed while noise is eliminated: suppose there is a digital signal sequence x j (-∞<j<+ -infinity), when it is subjected to a filtering process, a window of odd length L is first defined, l=2n+1, n being a positive integer, assuming that at a certain instant i, the signal samples within the windowThe cost is x i-N ,…,x i ,…,x i+N Wherein x is i Is the signal sample value at the center of the window, and after rearranging the L signals from small to large, the value is defined as the output value of the median filter.
Step 2, carrying out event detection on the data in the step 1 through improved generalized likelihood ratio detection, if an event is detected, executing the step 3, otherwise returning to the step 1;
as shown in fig. 2, the step 2 is specifically implemented according to the following steps:
step 2.1, calculating fundamental wave active power P according to the step (1) 1 The combined active power P is used as a two-dimensional power time sequenceAccording to formula (2), binary hypothesis testing is proposed;
wherein V is 1 Is the fundamental voltage, I 1 As a fundamental current, a current is supplied,a phase difference between the two. n is n c For the time of occurrence of the change point, k is the total length of the window, n is the last sample time in the window, μ 0 ,∑ 0 To hypothesis test H 0 Gaussian distribution mean, covariance matrix, mu under the condition a ,∑ a Is H 1 Multi-dimensional signal mean value and multi-dimensional covariance matrix, mu before occurrence of change point under condition b ,∑ b Is H 1 A multidimensional signal mean value and a multidimensional covariance matrix after the occurrence of the variable points under the condition;
step 2.2 defining two consecutive windows W in the time sequence a And W is b Two windowsIntraoral sample X n ＝{x m M=n-k+1, …, n }, both window lengths are k/2, μ and Σ in both windows are calculated according to equation (3) and equation (4), respectively, and then the decision function g is calculated according to equation (5) n ；
Step 2.3, g n And threshold h 1 Comparing, searching for a suspicious point of occurrence of the event: when the decision function value is greater than h 1 When rejecting H 0 The data distribution in the two windows is inconsistent, and the time n is the variable point time c There is a possibility of an event occurring; when the decision function is less than h 1 When rejecting H 1 The two windows are consistent in data distribution, and no event occurs. And when the sampling points of the detected events are continuous or spaced within 3, it is regarded as having only the maximum g n The corresponding event occurs;
and 2.4, taking the event point as a base point, and performing F test screening on the event of false detection. First, assume that the variance values of window data before and after the base point are equal:calculating the value of the statistic F according to formula (6), given the significance level α, rejecting the hypothesis H if the value of F satisfies formula (7) 0 Considering that the two parent variances have significant differences, and judging that an event occurs at the point;
in the method, in the process of the invention,the sample capacity and variance of the windows before and after the base point, respectively.
Step 3, extracting multidimensional load characteristics from the detected event variable points;
step 3 is specifically implemented according to the following steps: the power characteristics of the variable points are extracted, and the method specifically comprises the following steps: active power, fundamental active power, reactive power, fundamental reactive power, apparent power, distortion power, power factor angle, fundamental power factor; extracting harmonic characteristics at the variable points, wherein the harmonic characteristics comprise voltage, amplitude of each subharmonic of one to nine times of voltage, content of each subharmonic, difference of content of each subharmonic and total harmonic distortion; the current waveform characteristics comprise a wave peak value, an average value and a wave crest coefficient; extracting the V-I track characteristics at the variable points, wherein the method specifically comprises the following steps: symmetry, wrapping direction, wrapping area, number of intersections, Y-axis intercept, Y-axis span, centerline curvature, track mid-portion peak, left and right portion area, mid-portion shape, instantaneous admittance standard deviation. Step 4, screening relevant features by using a semi-supervised algorithm combining the Relief-F and the mRmR aiming at the multidimensional features extracted in the step 3;
as shown in fig. 3, the step 4 is specifically implemented according to the following steps:
step 4.1, labeling part of the multidimensional feature sample as a labeled sample S 1 The remainder is marked as unlabeled sample S 2 The label is C, and the multidimensional feature set is a= (a) 1 ,A 2 ,…,A N )；
Step 4.2, slave S 1 Randomly extracting a sample s of the class C q (C q E C); from S 2 D neighbor samples are selected and marked asAt S 1 Middle-slave division C q Each other thanClass C p E, in C (p is not equal to q), respectively solving a nearest neighbor sample x of s; and at S 2 D neighbor samples of x are solved, denoted +.>Wherein the neighbor formula is shown as formula (8); updating all feature weights based on equation (9);
wherein M represents the iteration times, d is the number of neighbor samples,for the t nearest neighbor sample in class q to which sample s belongs,>represents the t-th neighbor sample in the P-th class different from the sample s class, P (C p ) Representing the probability of a class p object, A k For the kth feature>Representing sample s and sample->With respect to feature A k Is a distance of (2);
step 4.3, circularly executing the step 4.2 for M times, and obtaining the final output characteristic weight omega of the semi-supervision Relief-F k 'features with weight coefficients smaller than θ are rejected (k=1, 2, …, w'), so that a candidate feature subset is obtained, and the process is performedSemi-supervised mRmR characteristic selection is carried out on the model;
step 4.4, in the labeled sample S according to equation (11) 1 The correlation degree of each feature and the sample label is calculated, and the sample S is not marked according to the formula (12) 2 Calculating redundancy of each feature;
in the method, in the process of the invention,is shown in labeled sample S 1 Middle calculation A k Mutual information with tag C, R (A k ,S m-1 ) Representing an existing feature subset S m-1 Comprising m-1 features, not in S m-1 Feature A in the subset k Redundancy with selected features;
step 4.5, establishing a feature candidate set H, and selecting the maximum correlation degree D max Corresponding features as candidate set leader H 1 Sequentially selecting the kth feature A according to equation (13) k Placing the specific characteristics into H until the specific characteristic number w is selected;
and 4.6, calculating the weight of each characteristic according to the formula (14).
Step 5, taking the characteristics obtained in the step 4 as load marks, establishing a load characteristic library through a self-adaptive FCM algorithm, identifying the load working state of the household appliance in the user by using the maximum membership degree, and if the identification results of the two FCM algorithms are consistent, ending, and if the identification results are inconsistent, executing the step 6;
as shown in fig. 4 and 5, the step 5 is specifically implemented as follows:
step 5.1, setting an initial cluster number b, determining an initial cluster center based on a maximum and minimum distance algorithm, and resetting a counter k to zero;
a. calculating an average value of a data setThe sample point farthest from the average value is recorded as V 1j (j=1, 2, …, w); b. calculating the minimum distance D of each data point from the selected cluster center according to the formula (15) x Select D x The maximum point is used as a new clustering center; c. repeating the step b until b initial clustering centers are selected;
D x ＝mind(x i ,Z' k )k'＝1,…,kselected (15)
step 5.2, constructing a loss function L of the FCM1 algorithm 1 As shown in formula (17), for u ij ，λ j And v ik Combining the partial derivatives and performing u by using the derived formula (18) and formula (19) ij And v ik Is updated iteratively;
/>
wherein n is the number of samples, b is the number of clusters, w is the number of features,for the sample weight, u ij The j-th sample belongs to the membership degree of the i-th class, v is a clustering center matrix, T u E (0, infinity) replaces the fuzzy index m, is used for controlling the entropy value, and introduces the maximum entropy as regularization;
step 5.3, calculate J 1 If the value of (1)The clustering of the initial clustering number b is completed, iteration is terminated, otherwise, k=k+1 is transferred to step 5.2 to continue iteration until the requirement is met;
step 5.4, calculating the cluster effectiveness index of the initial cluster number b, namely the contour coefficient S according to the formula (20) b If it meets S b-2 ＜S b-1 And S is b-1 ＞S b The self-adaptive FCM1 classification is finished, otherwise, whether the cluster number b reaches the maximum value is judgedIf the maximum value is reached, taking all S b B=b+1 if the maximum value of b, u, v is not reached, and the step 5.1 is executed again;
wherein n is the number of samples, a (i) is the average distance between the sample point i and the remaining sample objects in the same cluster, b (i) is the minimum value of the average distance between the sample point i and the remaining sample objects in each cluster, S b At [ -1,1]The larger the S value is, the better the clustering effect is;
step 5.5, setting the optimal cluster number b of the FCM1 as the cluster number of the FCM2, determining an initial cluster center by adopting a maximum and minimum distance algorithm, and resetting a counter k;
step 5.6, constructing a loss function J of the FCM2 algorithm 2 As shown in formula (21), for v ik Deviation determination using derived(22) Proceeding v ik Is updated by iteration of the method, and the sampling projection gradient descent method is used for u ij Iteration is performed: u (u) (k+1) ＝argmin:J 2 (u,v (k) )；
Wherein lambda is equal to or greater than 0 and is an orthogonal constraint parameter, and the second function term is to add an approximately orthogonal constraint to the membership matrix to balance the influence of unbalanced data, wherein u ip Representing the membership degree of the ith sample point belonging to the class cluster p, which is obtained by the original membership degree function u ip ' pass throughTransformed into the material;
step 5.7, calculate J 2 If the value of (1)The FCM2 clustering is completed and the iteration is terminated, otherwise, k=k+1, go to step 5.6 and continue the iteration until +.>The clustering number and the clustering center of the FCM algorithm in the load feature library are determined;
and 5.8, inputting samples to be detected, determining an initial clustering center by a load feature library, obtaining a membership function through FCM clustering, taking membership vectors corresponding to the samples to be identified, outputting the membership vectors, sequencing the membership, taking the category corresponding to the maximum membership, and marking the category as the category of the load to be identified. If the sample identification results of the FCM1 and the FCM2 are consistent, the algorithm identification is finished, the category corresponding to the maximum membership degree is the final sample identification result, and if the sample identification results are inconsistent, the step 6 is executed.
And 6, training the parameter smoothing factors of the GRNN by adopting an SA-BAS algorithm, and carrying out secondary identification of a sample to be tested by utilizing the trained GRNN to finish non-invasive secondary identification of the household appliance load.
As shown in fig. 6, step 6 is specifically performed according to the following steps:
step 6.1, combining samples with consistent FCM identification results, and training an improved GRNN neural network as a training set;
step 6.2, setting parameters: the number of neurons of the input layer is the characteristic dimension w; the mode layer neuron is the sample number n of the sample set of the training set, and the transfer function of the ith neuron is shown as a formula (23); the summation layer comprises two types of neurons, one is to carry out arithmetic summation on the output of all the mode layer neurons, the transfer function of the two types of neurons is shown as a formula (24), and the other is to carry out weighted summation on the output of all the mode layer neurons, and the transfer function of the two types of neurons is shown as a formula (25); the number of the neurons of the output layer is equal to the dimension of the output vector in the training sample, namely the clustered electrical appliance class number b, and the network output of the output layer is divided by two types of neurons, as shown in a formula (26);
wherein X is an input sample, X i For the ith sample data of the training set, sigma is a smoothing factor, Y ij For the ith output sample Y i The j-th element of (a);
step 6.3, network parameters of GRNNSeveral smoothing factors are present, so we use the SA-BAS algorithm to find the optimal smoothing factor σ; firstly, training data are input, and parameters are initialized: longicorn initial position X 0 =1, temperature t=200, step factor α=0.95, maximum number of iterations n=200, annealing cycle number l=100, centroid to tentacle distance d=1.5, counter h is normalized to 1;
step 6.4, setting the step length S=T of the longicorn, and simultaneously setting the counter T to be 1;
step 6.5, updating the left whisker position and the right whisker position of the longicorn according to the formula (27), and updating the longicorn position X according to the formula (28) t+1 Calculating probability according to Metropolis criterion, judging whether X is accepted t+1 As a new solution, the step S is updated as shown in equation (29) t ＝αS t-1 Updating the distance from the centroid to the tentacle according to the formula (30), judging whether the counter t is more than or equal to the annealing cycle times L, if so, executing the step 6.6, otherwise, executing the step 6.5 in a circulating way, wherein t=t+1;
/>
in the method, in the process of the invention,taking the direction of right whisker pointing to left whisker as the orientation of longicorn in space, sign () as a sign function, the value being greater than zero, taking 1, less than zero, taking-1, equal to zero, taking 0, T as the current temperature, deltaT=f (X t+1 )-f(X t )，Omega is an expression of the fitness function j As characteristic weight, x ij Output value for training sample network, +.>Expected output values for training samples;
and 6.6, updating the self-adaptive factor beta, T=beta T according to a formula (31), judging whether the counter h is larger than or equal to the iteration times N, if so, outputting the optimal longicorn position as an optimal smoothing factor, and ending training, otherwise, executing the step 6.4.
Wherein f h For the current fitness value, f min The historical optimal fitness value is used;
and (3) constructing a GRNN network by combining the trained smoothing factors with the conditions of the step 6.2, so as to carry out deep re-identification of the sample to be detected and finish non-invasive secondary identification of the household appliance load.
According to the non-invasive household appliance load depth re-identification method, the event detection algorithm combined with GLR and F detection is used for locating the occurrence time of an event, the semi-supervised algorithm combined with Relief-F and mRmR is used for screening out the characteristics with high correlation with an electric tag, the self-adaption FCM1 and FCM2 are combined for one-time identification, the improved GRNN is used for final identification, misidentification and feature redundancy are reduced, and compared with other literature methods, the identification accuracy is high and the identification rate is high.
Claims (7)
1. A non-invasive household appliance load depth re-identification method is characterized by comprising the following steps:
step 1, acquiring user high-frequency data, and denoising the acquired user high-frequency data;
step 2, carrying out event detection on the data in the step 1 through improved generalized likelihood ratio detection, if an event is detected, executing the step 3, otherwise returning to the step 1;
step 3, extracting multidimensional load characteristics from the detected event variable points;
step 4, screening relevant features by using a semi-supervised algorithm combining the Relief-F and the mRmR aiming at the multidimensional features extracted in the step 3;
step 5, taking the characteristics obtained in the step 4 as load marks, establishing a load characteristic library through a self-adaptive FCM algorithm, identifying the load working state of the household appliance in the user by using the maximum membership degree, and if the identification results of the two FCM algorithms are consistent, ending, and if the identification results are inconsistent, executing the step 6;
and 6, training the parameter smoothing factors of the GRNN by adopting an SA-BAS algorithm, and performing secondary identification of a sample to be tested by using the trained GRNN to finish non-invasive type household appliance load deep re-identification.
2. The non-invasive home appliance load depth re-identification method according to claim 1, wherein the step 1 is specifically implemented according to the following steps:
step 1.1, acquiring high-frequency household appliance load data containing electrical parameters, wherein the high-frequency household appliance load data comprise voltage, current and corresponding power;
step 1.2, denoising processing of the power signal, wherein isolated noise points are easy to be mistakenly identified as events by an event detection algorithm, so that a median filtering method is selected to process the original power signal, and edge information is not changed while noise is eliminated: suppose there is a digital signal sequence x j ，-∞<j<When carrying out filtering treatment, firstly, a window with the length of an odd number L is defined, l=2n+1, n being a positive integer, assuming that at a certain instant i, the signal samples within the window are x i-N ,…,x i ,…,x i+N Wherein x is i Is the signal sample value at the center of the window, and after rearranging the L signals from small to large, the value is defined as the output value of the median filter.
3. The non-invasive home appliance load depth re-identification method according to claim 1, wherein the step 2 is specifically implemented according to the following steps:
step 2.1, calculating fundamental wave active power P according to the step (1) 1 The combined active power P is used as a two-dimensional power time sequenceAccording to formula (2), binary hypothesis testing is proposed;
wherein V is 1 Is the fundamental voltage, I 1 As a fundamental current, a current is supplied,for the phase difference of the two, n c For the time of occurrence of the change point, k is the total length of the window, n is the last sample time in the window, μ 0 ,∑ 0 To hypothesis test H 0 Gaussian distribution mean, covariance matrix, mu under the condition a ,∑ a Is H 1 Multi-dimensional signal mean value and multi-dimensional covariance matrix, mu before occurrence of change point under condition b ,∑ b Is H 1 A multidimensional signal mean value and a multidimensional covariance matrix after the occurrence of the variable points under the condition;
step 2.2 defining two consecutive windows W in the time sequence a And W is b Samples in two windows are X n ＝{x m M=n-k+1, …, n }, both window lengths are k/2, μ and Σ in both windows are calculated according to equation (3) and equation (4), respectively, and then the decision function g is calculated according to equation (5) n ；
Step 2.3, g n And threshold h 1 Comparing, searching for a suspicious point of occurrence of the event: when the decision function value is greater than h 1 When rejecting H 0 The data distribution in the two windows is inconsistent, and the time n is the variable point time c There is a possibility of an event occurring; when the decision function is less than h 1 When rejecting H 1 The two windows are consistent in data distribution, and no event occurs; and when the sampling points of the detected events are continuous or spaced within 3, it is regarded as having only the maximum g n The corresponding event occurs;
step 2.4, taking the event point as a base point, and carrying out F test screening on the event of false detection: first, assume that the variance values of window data before and after the base point are equal:calculating the value of the statistic F according to formula (6), given the significance level α, rejecting the hypothesis H if the value of F satisfies formula (7) 0 Considering that the two parent variances have significant differences, and judging that an event occurs at the point;
or->
Wherein n is 1 ,n 2 ,/>The sample capacity and variance of the windows before and after the base point, respectively.
4. The non-invasive home appliance load depth re-identification method according to claim 1, wherein the step 3 is specifically implemented as follows: extracting power characteristics of the variable points, including active power, fundamental wave active power, reactive power, fundamental wave reactive power, apparent power, distortion power, power factor angle and fundamental wave power factor; extracting harmonic characteristics at the variable points, including voltage, amplitude of each subharmonic of one to nine times of voltage, content of each subharmonic, difference of content of each subharmonic and total harmonic distortion; the current waveform characteristics comprise a wave peak value, an average value and a wave crest coefficient; and extracting V-I track characteristics at the variable points, wherein the V-I track characteristics comprise symmetry, surrounding direction, surrounding area, number of intersection points, Y-axis intercept, Y-axis span, center line curvature, track middle part peak value, left and right part area, middle part shape and instantaneous admittance standard deviation.
5. The non-invasive home appliance load depth re-identification method according to claim 1, wherein the step 4 is specifically implemented according to the following steps:
step 4.1, labeling part of the multidimensional feature sample as a labeled sample S 1 The remainder is marked as unlabeled sample S 2 The label is C, and the multidimensional feature set is a= (a) 1 ,A 2 ,…,A N )；
Step 4.2, slave S 1 Randomly extracting a sample s of the class C q ，C q E C; from S 2 D neighbor samples are selected and marked asAt S 1 Middle-slave division C q Each other class C p E, in C and p is not equal to q, respectively solving a nearest neighbor sample x of s; and at S 2 D neighbor samples of x are solved, denoted +.>Wherein the neighbor formula is shown as formula (8); updating all feature weights based on equation (9);
wherein M represents the iteration times, d is the number of neighbor samples,for the t-th neighbor sample in the q class to which sample s belongs,represents the t-th neighbor sample in the P-th class different from the sample s class, P (C p ) Representing the probability of a class p object, A k For the kth feature>Representing sample s and sample->With respect to feature A k Is a distance of (2);
step 4.3, circularly executing the step 4.2 for M times, and obtaining the final output characteristic weight omega of the semi-supervision Relief-F k ′，k＝1，2，…，w′，Removing the features with the weight coefficients smaller than theta to obtain a candidate feature subset, and further performing semi-supervised mRmR feature selection on the candidate feature subset;
step 4.4, in the labeled sample S according to equation (11) 1 The correlation degree of each feature and the sample label is calculated, and the sample S is not marked according to the formula (12) 2 Calculating redundancy of each feature;
in the method, in the process of the invention,is shown in labeled sample S 1 Middle calculation A k Mutual information with tag C, R (A k ,S m-1 ) Representing an existing feature subset S m-1 Comprising m-1 features, not in S m-1 Feature A in the subset k Redundancy with selected features;
step 4.5, establishing a feature candidate set H, and selecting the maximum correlation degree D max Corresponding features as candidate set leader H 1 Sequentially selecting the kth feature A according to equation (13) k Placing the specific characteristics into H until the specific characteristic number w is selected;
step 4.6, calculating the weight of each feature according to the formula (14)
6. The non-invasive home appliance load depth re-identification method according to claim 1, wherein the step 5 is specifically implemented according to the following steps:
step 5.1, setting an initial cluster number b, determining an initial cluster center based on a maximum and minimum distance algorithm, and resetting a counter k to zero;
a. calculating an average value of a data setThe sample point farthest from the average value is recorded as V 1j (j＝1,2,…,w)；
b. Calculating the minimum distance D of each data point from the selected cluster center according to the formula (15) x Select D x The maximum point is used as a new clustering center;
c. repeating the step b until b initial clustering centers are selected;
D x ＝min d(x i ,Z' k ) k'＝1,…,kselected (15)
step 5.2, constructing a loss function L of the FCM1 algorithm 1 As shown in formula (17), for u ij ，λ j And v ik Combining the partial derivatives and performing u by using the derived formula (18) and formula (19) ij And v ik Is updated iteratively;
wherein n is the number of samples, b is the number of clusters, w is the number of features,for the sample weight, u ij The j-th sample belongs to the membership degree of the i-th class, v is a clustering center matrix, T u E (0, infinity) replaces the fuzzy index m, is used for controlling the entropy value, and introduces the maximum entropy as regularization;
step 5.3, calculate J 1 If the value of (1)The clustering of the initial clustering number b is completed, iteration is terminated, otherwise, k=k+1 is transferred to step 5.2 to continue iteration until the requirement is met;
step 5.4, calculating the cluster effectiveness index of the initial cluster number b, namely the contour coefficient S according to the formula (20) b If it meets S b-2 <S b-1 And S is b-1 >S b The self-adaptive FCM1 classification is finished, otherwise, whether the cluster number b reaches the maximum value is judgedIf the maximum value is reached, taking all S b B=b+1 if the maximum value of b, u, v is not reached, and the step 5.1 is executed again;
wherein n is the number of samples, a (i) is the average distance between the sample point i and the remaining sample objects in the same cluster, b (i) is the minimum value of the average distance between the sample point i and the remaining sample objects in each cluster, S b At [ -1,1]The larger the S value is, the better the clustering effect is;
step 5.5, setting the optimal cluster number b of the FCM1 as the cluster number of the FCM2, determining an initial cluster center by adopting a maximum and minimum distance algorithm, and resetting a counter k;
step 5.6, constructing a loss function J of the FCM2 algorithm 2 As shown in formula (21), for v ik Obtaining bias derivative, and v is carried out by using derived formula (22) ik Is updated by iteration of the method, and the sampling projection gradient descent method is used for u ij Iteration is performed: u (u) (k+1) ＝argmin:J 2 (u,v (k) )；
Wherein lambda is equal to or greater than 0 and is an orthogonal constraint parameter, and the second function term is to add an approximately orthogonal constraint to the membership matrix to balance the influence of unbalanced data, wherein u ip Representing the membership degree of the ith sample point belonging to the class cluster p, which is obtained by the original membership degree function u ip ' pass throughTransformed into the material;
step 5.7, calculate J 2 If the value of (1)The FCM2 clustering is completed and the iteration is terminated, otherwise, k=k+1, go to step 5.6 and continue the iteration until +.>The clustering number and the clustering center of the FCM algorithm in the load feature library are determined;
step 5.8, inputting a sample to be detected, determining an initial clustering center by a load feature library, obtaining a membership function through FCM clustering, taking a membership vector corresponding to the sample to be identified, outputting the membership vector, sequencing the membership, taking a category corresponding to the maximum membership, and marking the category as the category of the load to be identified; if the sample identification results of the FCM1 and the FCM2 are consistent, the algorithm identification is finished, the category corresponding to the maximum membership degree is the final sample identification result, and if the sample identification results are inconsistent, the step 6 is executed.
7. The non-invasive home appliance load depth re-identification method according to claim 1, wherein the step 6 is specifically implemented according to the following steps:
step 6.1, combining samples with consistent FCM identification results, and training an improved GRNN neural network as a training set;
step 6.2, setting parameters: the number of neurons of the input layer is the characteristic dimension w; the mode layer neuron is the sample number n of the sample set of the training set, and the transfer function of the ith neuron is shown as a formula (23); the summation layer comprises two types of neurons, one is to carry out arithmetic summation on the output of all the mode layer neurons, the transfer function of the two types of neurons is shown as a formula (24), and the other is to carry out weighted summation on the output of all the mode layer neurons, and the transfer function of the two types of neurons is shown as a formula (25); the number of the neurons of the output layer is equal to the dimension of the output vector in the training sample, namely the clustered electrical appliance class number b, and the network output of the output layer is divided by two types of neurons, as shown in a formula (26);
wherein X is an input sample, X i For the ith sample data of the training set, sigma is a smoothing factor, Y ij For the ith output sample Y i The j-th element of (a);
step 6.3, the network parameters of GRNN are only smooth factors, so we use SA-BAS algorithm to find the optimal smooth factor sigma; head partFirstly, training data are input, and parameters are initialized: longicorn initial position X 0 =1, temperature t=200, step factor α=0.95, maximum number of iterations n=200, annealing cycle number l=100, centroid to tentacle distance d=1.5, counter h is normalized to 1;
step 6.4, setting the step length S=T of the longicorn, and simultaneously setting the counter T to be 1;
step 6.5, updating the left whisker position and the right whisker position of the longicorn according to the formula (27), and updating the longicorn position X according to the formula (28) t+1 Calculating probability according to Metropolis criterion, judging whether X is accepted t+1 As a new solution, the step S is updated as shown in equation (29) t ＝αS t-1 Updating the distance from the centroid to the tentacle according to the formula (30), judging whether the counter t is more than or equal to the annealing cycle times L, if so, executing the step 6.6, otherwise, executing the step 6.5 in a circulating way, wherein t=t+1;
in the method, in the process of the invention,taking the direction of right whisker pointing to left whisker as the orientation of longicorn in space, sign () as a sign function, the value being greater than zero, taking 1, less than zero, taking-1, equal to zero, taking 0, T as the current temperature, deltaT=f (X t+1 )-f(X t )，Omega is an expression of the fitness function j As characteristic weight, x ij Output value for training sample network, +.>Expected output values for training samples;
step 6.6, updating the self-adaptive factor beta, T=beta T according to a formula (31), judging whether the counter h is larger than or equal to the iteration times N, if so, outputting the optimal longicorn position as an optimal smoothing factor, and finishing training, otherwise, executing the step 6.4;
wherein f h For the current fitness value, f min The historical optimal fitness value is used;
and (3) constructing a GRNN network by combining the trained smoothing factors with the conditions of the step 6.2, so as to carry out deep re-identification of the sample to be detected and finish the final non-invasive household appliance load identification.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110089840.XA CN112821559B (en) | 2021-01-22 | 2021-01-22 | Non-invasive household appliance load depth re-identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110089840.XA CN112821559B (en) | 2021-01-22 | 2021-01-22 | Non-invasive household appliance load depth re-identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112821559A CN112821559A (en) | 2021-05-18 |
| CN112821559B true CN112821559B (en) | 2023-08-01 |
Family
ID=75858877
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110089840.XA Active CN112821559B (en) | 2021-01-22 | 2021-01-22 | Non-invasive household appliance load depth re-identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112821559B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113595242B (en) * | 2021-06-28 | 2024-02-20 | 湖南大学 | Non-invasive load identification method based on depth CNN-HMM |
| CN113723479A (en) * | 2021-08-18 | 2021-11-30 | 南京工程学院 | Non-invasive load identification method based on GRNN and mean shift algorithm |
| CN114384999B (en) * | 2021-11-19 | 2023-07-21 | 福州大学 | User-independent myoelectric gesture recognition system based on self-adaptive learning |
| CN114325081A (en) * | 2021-12-29 | 2022-04-12 | 润建股份有限公司 | Non-invasive load identification method based on multi-modal characteristics |
| CN116893314A (en) * | 2023-09-04 | 2023-10-17 | 国网浙江省电力有限公司余姚市供电公司 | Non-invasive power load monitoring method, device, equipment and storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20160141032A (en) * | 2015-05-27 | 2016-12-08 | 전자부품연구원 | Non-Intrusive Appliance Load Monitoring Method using a Switching Factorial Hidden Markov Model and System applying the same |
| CN108846530A (en) * | 2018-09-28 | 2018-11-20 | 国网上海市电力公司 | One kind being based on the short-term load forecasting method of " cluster-recurrence " model |
| CN108898154A (en) * | 2018-09-29 | 2018-11-27 | 华北电力大学 | A kind of electric load SOM-FCM Hierarchical clustering methods |
| CN110555369A (en) * | 2019-07-16 | 2019-12-10 | 浙江工业大学 | MLCDTL-based non-intrusive load identification method |
| CN110866841A (en) * | 2019-11-20 | 2020-03-06 | 江苏方天电力技术有限公司 | Power consumer industry dimension power consumption pattern identification analysis method and system based on double clustering method |
| CN110991818A (en) * | 2019-11-14 | 2020-04-10 | 广西电网有限责任公司电力科学研究院 | Load identification method integrating event detection and neural network |
| CN112101110A (en) * | 2020-08-13 | 2020-12-18 | 西安理工大学 | Non-invasive load identification method for user side of power system |
-
2021
- 2021-01-22 CN CN202110089840.XA patent/CN112821559B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20160141032A (en) * | 2015-05-27 | 2016-12-08 | 전자부품연구원 | Non-Intrusive Appliance Load Monitoring Method using a Switching Factorial Hidden Markov Model and System applying the same |
| CN108846530A (en) * | 2018-09-28 | 2018-11-20 | 国网上海市电力公司 | One kind being based on the short-term load forecasting method of " cluster-recurrence " model |
| CN108898154A (en) * | 2018-09-29 | 2018-11-27 | 华北电力大学 | A kind of electric load SOM-FCM Hierarchical clustering methods |
| CN110555369A (en) * | 2019-07-16 | 2019-12-10 | 浙江工业大学 | MLCDTL-based non-intrusive load identification method |
| CN110991818A (en) * | 2019-11-14 | 2020-04-10 | 广西电网有限责任公司电力科学研究院 | Load identification method integrating event detection and neural network |
| CN110866841A (en) * | 2019-11-20 | 2020-03-06 | 江苏方天电力技术有限公司 | Power consumer industry dimension power consumption pattern identification analysis method and system based on double clustering method |
| CN112101110A (en) * | 2020-08-13 | 2020-12-18 | 西安理工大学 | Non-invasive load identification method for user side of power system |
Non-Patent Citations (4)
| Title |
|---|
| Non-Intrusive Identification of Loads by Random Forest and Fireworks Optimization;PAULO RICARDO ZAMBELLI TAVEIRA et al.;《IEEE》;20200416;第75060-75072页 * |
| 利用快速S变换的电能质量扰动识别方法;满蔚仕 等;《西安交通大学学报》;20120831;第47卷(第8期);第133-140页 * |
| 基于广义回归神经网络的非侵入式负荷识别方法;江帆 等;《电测与仪表》;20200210;第27卷(第03期);第6-11页 * |
| 基于聚类特征及seq2seq深度CNN的家电负荷识别方法研究;汪繁荣 等;《电测与仪表》;20201016;第1-10页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112821559A (en) | 2021-05-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112821559B (en) | Non-invasive household appliance load depth re-identification method | |
| Ahmed et al. | Feature selection–based detection of covert cyber deception assaults in smart grid communications networks using machine learning | |
| CN111967294B (en) | Unsupervised domain self-adaptive pedestrian re-identification method | |
| Ferreira et al. | Time series clustering via community detection in networks | |
| Ikram et al. | Improving accuracy of intrusion detection model using PCA and optimized SVM | |
| CN112732748B (en) | Non-invasive household appliance load identification method based on self-adaptive feature selection | |
| Akbari et al. | Clustered K nearest neighbor algorithm for daily inflow forecasting | |
| CN112560596B (en) | Radar interference category identification method and system | |
| CN111553186A (en) | Electromagnetic signal identification method based on depth long-time and short-time memory network | |
| Liu et al. | Integrating multi-objective genetic algorithm and validity analysis for locating and ranking alternative clustering | |
| Wan et al. | Mining subsequent trend patterns from financial time series | |
| CN107909090A (en) | Learn semi-supervised music-book on pianoforte difficulty recognition methods based on estimating | |
| CN114124437B (en) | Encrypted flow identification method based on prototype convolutional network | |
| CN112465253B (en) | Method and device for predicting links in urban road network | |
| CN115643153A (en) | Alarm correlation analysis method based on graph neural network | |
| Handra et al. | Anomaly detection in data mining. Hybrid approach between filtering-and-refinement and DBSCAN | |
| KR101012848B1 (en) | Clustering method of complex networking and the method for grouping of clustered complex networking | |
| Alsharkawi et al. | Poverty classification using machine learning: The case of Jordan. Sustainability 2021, 13, 1412 | |
| Mehrmolaei et al. | Fuzzy-clustering time series: population-based an enhanced technique | |
| CN117675303A (en) | Cloud server DDoS attack detection method based on Stacking algorithm | |
| Lumauag | A Modified Memory-Based Collaborative Filtering Algorithm based on a New User Similarity Measure | |
| Wen et al. | Research on Automated Classification Method of Network Attacking Based on Gradient Boosting Decision Tree | |
| Chitraa et al. | Clustering of navigation patterns using Bolzwano_Weierstrass theorem | |
| RAJALAKSHMI et al. | Centroid Stabilized Fuzzy Tukey Quartile and Z Curve Neural Network based Outlier Detection | |
| Liu et al. | An Improvement of K-means Algorithm for Big Data Clustering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20230706 Address after: 518000 1002, Building A, Zhiyun Industrial Park, No. 13, Huaxing Road, Henglang Community, Longhua District, Shenzhen, Guangdong Province Applicant after: Shenzhen Wanzhida Technology Co.,Ltd. Address before: 710048 Shaanxi province Xi'an Beilin District Jinhua Road No. 5 Applicant before: XI'AN University OF TECHNOLOGY Effective date of registration: 20230706 Address after: 628, building 118, No.72 Guowei Road, Xianhu community, Liantang street, Luohu District, Shenzhen, Guangdong 518000 Applicant after: Wuxing Technology (Shenzhen) Co.,Ltd. Address before: 518000 1002, Building A, Zhiyun Industrial Park, No. 13, Huaxing Road, Henglang Community, Longhua District, Shenzhen, Guangdong Province Applicant before: Shenzhen Wanzhida Technology Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |