CN116720127A - Non-invasive load monitoring method based on LMD and k nearest neighbor algorithm - Google Patents
Non-invasive load monitoring method based on LMD and k nearest neighbor algorithm Download PDFInfo
- Publication number
- CN116720127A CN116720127A CN202310485699.4A CN202310485699A CN116720127A CN 116720127 A CN116720127 A CN 116720127A CN 202310485699 A CN202310485699 A CN 202310485699A CN 116720127 A CN116720127 A CN 116720127A
- Authority
- CN
- China
- Prior art keywords
- sample
- data
- lmd
- samples
- category
- 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.)
- Pending
Links
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 61
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000012544 monitoring process Methods 0.000 title claims abstract description 24
- 238000012360 testing method Methods 0.000 claims abstract description 21
- 230000008569 process Effects 0.000 claims abstract description 9
- 238000013480 data collection Methods 0.000 claims abstract description 8
- 230000008859 change Effects 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 37
- 238000000354 decomposition reaction Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 8
- 238000005259 measurement Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 125000004122 cyclic group Chemical group 0.000 claims description 3
- 230000006872 improvement Effects 0.000 claims description 3
- 238000012804 iterative process Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 27
- 238000004458 analytical method Methods 0.000 description 4
- 230000005611 electricity Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a non-invasive load monitoring method based on LMD and k neighbor algorithms, which comprises the steps of adaptively decomposing data into multiple product components by the LMD algorithm on the premise of keeping the amplitude and frequency change characteristics of original signals, and meanwhile decomposing low-dimensional active power characteristics in a data set into multidimensional characteristic components to obtain detailed load characteristics of electrical equipment corresponding to the power data; an improved k-nearest neighbor algorithm model is established, a given test sample is calculated, euclidean distance between a sample to be tested and a plurality of adjacent test samples is calculated, the label with the largest category number is identified as the label of test set data, and the distance weight value is added to reclassify so as to reduce load identification errors caused by error disturbance in the data collection process; acquiring power data of a user side, decomposing the power data characteristics, and carrying the power data into a trained model; and (3) judging the approximation degree of the classification result by adopting a cosine function, comparing the approximation degree with a similarity threshold value, and finally obtaining an identification result.
Description
Technical Field
The application relates to the technical field of non-invasive load decomposition, in particular to a non-invasive load monitoring method based on an LMD and k nearest neighbor algorithm.
Background
Electric power is ubiquitous in everyday life and industrial systems and is an important factor for social stability and happiness. The intelligent power system can timely acquire the user demand and the power decomposition condition. Load monitoring is a key step in achieving intelligent power splitting, so the importance of load monitoring is self-evident. The methods of load monitoring can be divided into two types: invasive and non-invasive. The intrusive load monitoring method is to install a sensor on each user's electrical equipment to acquire operation data. Non-invasive load monitoring (Non-Intrusive Load Monitoring, NILM) does not require the installation of sensors. The invasive measurement can truly reflect the operation data, but monitoring equipment needs to be installed for all users, so that the cost is high and the users are difficult to accept. The NILM can implement power analysis such as fault analysis without installing a sensor, and identification of the access electrical device is achieved by measuring electrical elements of the access point and analyzing load characteristics of individual electrical devices. The NILM does not make any changes to the user equipment and ensures normal electricity usage to achieve electricity usage resolution. Thus, as smart grids evolve, NILM becomes increasingly important. The system and the method enable the user to better know the load characteristics of the user, guide the user to change the electricity utilization habit, and achieve the purpose of energy conservation. The load monitoring technology can predict the power demand, better perform power dispatching, reduce unnecessary energy consumption and realize good interaction between the power generation end and the power utilization end.
The NILM technology has been developed for many years at home and abroad, and the research mainly comprises two parts: firstly, researching sample feature extraction, and obtaining feature signals by using different data processing; the other is the study of the machine learning method. Although researchers have conducted intensive studies on these parts, incomplete samples or insignificant feature signals may lead to poor machine learning models, making the diagnosis inaccurate. The load identification algorithm is richer in theory, but has not been promoted on a large scale in practical application, and the load characteristic database with complete classification and multiple models is difficult to establish, and meanwhile, the existing load identification algorithm has higher requirements on the data sampling frequency, so the load identification algorithm is still to be enhanced in practical application.
Disclosure of Invention
1. The technical problems to be solved are as follows:
aiming at the technical problems, the application provides a non-invasive load monitoring method based on LMD and k nearest neighbor algorithm, which improves the non-invasive load decomposition precision on the basis of not increasing the training time cost.
2. The technical scheme is as follows:
a non-invasive load monitoring method based on LMD and k neighbor algorithm is characterized in that: the method comprises the following steps:
step one, acquiring a given test sample which is power data acquired by a user side; the power data comprise time sequence data of current and voltage;
the data obtained in the first step is adaptively decomposed into multiple product components on the premise of keeping the amplitude and frequency change characteristics of original signals through an LMD algorithm, and meanwhile, the low-dimensional active power characteristics in the data set are decomposed into multi-dimensional characteristic components, so that the detailed load characteristics of the electrical equipment corresponding to the power data are obtained;
step three, an improved k-nearest neighbor algorithm model is established, a test sample is given, euclidean distance between the sample to be tested obtained in the step two and a plurality of adjacent test samples is calculated, the label with the largest category number is identified as the label of test set data, and the distance weight value is added for reclassifying so as to reduce load identification errors caused by error disturbance in the data collection process;
step four: acquiring power data of a user side, decomposing the power data based on the characteristics of the second step, and carrying the power data into a model trained in the third step;
and fifthly, performing approximation degree judgment on the classification result obtained in the fourth step by adopting a cosine function, calculating approximation degree of the feature vector of the sample to be detected and the mean value point of the same class sample of the classification result in the feature library, comparing the approximation degree with a similarity threshold value, and finally obtaining an identification result.
Further, in the fourth step, the power data of the user side is collected through the data collection equipment; the data acquisition equipment comprises intelligent electric equipment and an intelligent ammeter.
Further, the second step specifically includes the following steps:
s21: processing the data, decomposing the collected active power characteristics of the electrical equipment by analyzing the training sample data set, and calculating the local average function m of the power signal P (t) of each acquisition period 11 (t) and envelope estimation function a 11 (t); wherein the local average value m of the jth acquisition period j And local amplitude a j This can be obtained by the following equation:
(1) Wherein n is j ，n j+1 Respectively representing local extreme points of power data of j and j+1 acquisition periods;
sequentially connecting the local mean values m according to the time sequence j And local amplitude a j To obtain a local mean function m j (t) and envelope estimation function a j (t)；
S22, calculating the signal S as in formula (2) 11 (t) and S 11 (t) repeating the above-mentioned process as an original signal to iteratively calculate S 12 (t)……S 1n (t) up to S 1n (t) is a pure FM signal; in an iterative process, a series of FM signals as in equation (2) will be generated;
s23 by dividing the envelope signal a 1 (t) and FM signal S 1n (t) multiplying to obtain a first pf component p 1 (t); as shown in the following formula:
P 1 (t)＝a 1 (t)S 1n (t) (3)
and (3) sequentially obtaining the PF component k times of loops based on the formula: p is p 2 (t)、p 3 (t)…p k (t) and its corresponding balancing function u (t);
after k cyclic iterations of the decomposition, the original power signal is decomposed into k PF components and a margin u k (t), i.e
Thereby realizing the decomposition of the low-dimensional active power characteristic in the data set into the multidimensional characteristic component based on the LMD decomposition.
Further, the third step comprises the following steps:
s31, dividing a given test sample into a training set and a verification set, and establishing a traditional k-nearest neighbor algorithm model; for the samples to be tested in the verification set, analyzing the samples to be tested and all samples z in the training set j Is a distance of (2); the Euclidean distance is used as a distance measurement formula of a k-nearest neighbor algorithm:
(6) Wherein d euc Representing the distances between the sample to be tested and all samples in the training set; z is Z l Representing an L-th feature of a sample to be classified; z l j Representing the first characteristic of the jth training sample, and m represents the characteristic number of the training sample;
training all samples to be testedThe calculation results of the distances of all samples are concentrated and arranged according to increasing order, k points are selected in the training set, the distance between the k points and the sample Z to be measured is minimum, and the neighborhood of the sample Z to be measured containing the k points is recorded as N k (Z); finally at N k (Z) identifying the label with the largest number of categories as the category c of the sample Z to be tested Z The classification voting function is as follows;
in the formula (7), G is the sum of all sample types; class (c) zj ) For training sample z j The species of (2);returning to 0 when the value is false, otherwise returning to 1, determining the most category in k neighbor data through superposition calculation, and considering the sample to be detected as the category;
s32, selecting the first k samples according to the distance increment sequence, and calculating the distance weight W of the k samples j The method comprises the steps of carrying out a first treatment on the surface of the Selecting the category with the largest overlapping distance weight index, namely the improved category c of the sample to be detected, according to the principle of classified overlapping on the calculated distance weight index z ' its improvement classification voting function improves as:
wherein G is a set of all sample categories; class (c) zj ) For training sample z j Is a category of (2);as a logic function, returns 0 when its value is false, otherwise returns 1.
Further, the fifth step is specifically: selecting the category with the maximum weight index through an improved classification voting function, and calculating the approximation degree of the characteristic vector of the sample to be detected and the mean value point of the sample with the same category of the classification result in the characteristic library; if the approximation degree value of the two is larger than alpha, the classification is considered to be correct, and a classification result is obtained; if the cosine function similarity value is smaller than alpha, the load is not the database load and is regarded as strange equipment; alpha is a preset approximation value.
3. The beneficial effects are that:
the application provides a non-invasive load monitoring method combining an LMD algorithm and a k nearest neighbor algorithm, wherein a local mean decomposition algorithm can gradually decompose a complex multi-component signal into the sum of a plurality of product functions and a residual component in a multi-loop iteration mode according to the complexity degree and the change rule of the signal, so that characteristic information can be extracted in different frequency bands of an original signal; in order to reduce load identification deviation caused by error disturbance in the data collection process, an improved k-nearest neighbor algorithm is provided; and finally, the cosine similarity judging principle can identify strange equipment, so that the accuracy of the NILM is improved, the calculated amount is reduced, and the convergence speed of the model is improved.
Drawings
FIG. 1 is a schematic diagram of the overall flow of the present application;
fig. 2 is a schematic diagram of the k-nearest neighbor algorithm in the present application.
Detailed Description
The present application will be described in detail with reference to the accompanying drawings.
As shown in fig. 1 to fig. 2, a non-invasive load monitoring method based on LMD and k nearest neighbor algorithm is characterized in that: the method comprises the following steps:
step one, acquiring a given test sample which is power data acquired by a user side; the power data comprise time sequence data of current and voltage;
the data obtained in the first step is adaptively decomposed into multiple product components on the premise of keeping the amplitude and frequency change characteristics of original signals through an LMD algorithm, and meanwhile, the low-dimensional active power characteristics in the data set are decomposed into multi-dimensional characteristic components, so that the detailed load characteristics of the electrical equipment corresponding to the power data are obtained;
step three, an improved k-nearest neighbor algorithm model is established, a test sample is given, euclidean distance between the sample to be tested obtained in the step two and a plurality of adjacent test samples is calculated, the label with the largest category number is identified as the label of test set data, and the distance weight value is added for reclassifying so as to reduce load identification errors caused by error disturbance in the data collection process;
step four: acquiring power data of a user side, decomposing the power data based on the characteristics of the second step, and carrying the power data into a model trained in the third step;
and fifthly, performing approximation degree judgment on the classification result obtained in the fourth step by adopting a cosine function, calculating approximation degree of the feature vector of the sample to be detected and the mean value point of the same class sample of the classification result in the feature library, comparing the approximation degree with a similarity threshold value, and finally obtaining an identification result.
Further, in the fourth step, the power data of the user side is collected through the data collection equipment; the data acquisition equipment comprises intelligent electric equipment and an intelligent ammeter.
Further, the second step specifically includes the following steps:
s21: processing the data, decomposing the collected active power characteristics of the electrical equipment by analyzing the training sample data set, and calculating the local average function m of the power signal P (t) of each acquisition period 11 (t) and envelope estimation function a 11 (t); wherein the local average value m of the jth acquisition period j And local amplitude a j This can be obtained by the following equation:
(1) Wherein n is j ，n j+1 Respectively representing local extreme points of power data of j and j+1 acquisition periods;
sequentially connecting the local mean values m according to the time sequence j And local amplitude a j To obtain a local mean function m j (t) and envelope estimation function a j (t)；
S22, calculating the signal S as in formula (2) 11 (t) and S 11 (t) repeating the above-mentioned process as an original signal to iteratively calculate S 12 (t)……S 1n (t) up to S 1n (t) is a pure FM signal; in an iterative process, a series of FM signals as in equation (2) will be generated;
s23 by dividing the envelope signal a 1 (t) and FM signal S 1n (t) multiplying to obtain a first pf component p 1 (t); as shown in the following formula:
P 1 (t)＝a 1 (t)S 1n (t) (3)
and (3) sequentially obtaining the PF component k times of loops based on the formula: p is p 2 (t)、p 3 (t)…p k (t) and its corresponding balancing function u (t);
after k cyclic iterations of the decomposition, the original power signal is decomposed into k PF components and a margin u k (t), i.e
Thereby realizing the decomposition of the low-dimensional active power characteristic in the data set into the multidimensional characteristic component based on the LMD decomposition.
Further, the third step comprises the following steps:
s31, dividing a given test sample into a training set and a verification set, and establishing a traditional k-nearest neighbor algorithm model; for the samples to be tested in the verification set, analyzing the samples to be tested and all samples z in the training set j Is a distance of (2); the Euclidean distance is used as a distance measurement formula of a k-nearest neighbor algorithm:
(6) Wherein d euc Representing the distances between the sample to be tested and all samples in the training set; z is Z l Representing an L-th feature of a sample to be classified;representing the first characteristic of the jth training sample, and m represents the characteristic number of the training sample;
the calculation results of the distances between all samples to be measured and all samples in the training set are arranged according to increasing order, k points are selected in the training set, the distance between the k points and the Z of the samples to be measured is the smallest, and the neighborhood of the Z of the samples to be measured containing the k points is recorded as N k (Z); finally at N k (Z) identifying the label with the largest number of categories as the category c of the sample Z to be tested Z The classification voting function is as follows;
in the formula (7), G is the sum of all sample types; class (c) zj ) For training sample z j The species of (2);returning to 0 when the value is false, otherwise returning to 1, determining the most category in k neighbor data through superposition calculation, and considering the sample to be detected as the category;
s32, selecting the first k samples according to the distance increment sequence, and calculating the distance weight W of the k samples j The method comprises the steps of carrying out a first treatment on the surface of the Selecting the category with the largest overlapping distance weight index, namely the improved category c of the sample to be detected, according to the principle of classified overlapping on the calculated distance weight index z ' its improvement classification voting function improves as:
wherein G is a set of all sample categories; class (c) zj ) For training sample z j Is a category of (2);as a logic function, returns 0 when its value is false, otherwise returns 1.
Further, the fifth step is specifically: selecting the category with the maximum weight index through an improved classification voting function, and calculating the approximation degree of the characteristic vector of the sample to be detected and the mean value point of the sample with the same category of the classification result in the characteristic library; if the approximation degree value of the two is larger than alpha, the classification is considered to be correct, and a classification result is obtained; if the cosine function similarity value is smaller than alpha, the load is not the database load and is regarded as strange equipment; alpha is a preset approximation value.
The similarity is generally expressed by a cosine function similarity value, and the cosine similarity calculation formula is as follows:
wherein A is a sample characteristic vector to be classified, and B is a sample central value vector of a classification result.
The improved k-nearest neighbor algorithm is combined with a similarity recognition mode to meet the requirement of being capable of recognizing non-database loads.
If the cosine function similarity value of the two is larger than alpha, the classification is considered to be correct, and a classification result is obtained; if the cosine function similarity value is smaller than alpha, the load is not the database load, and is regarded as strange equipment, and the output result is other. The identification result judgment threshold alpha based on the similarity of the cosine function is set to be 0.89 according to experience.
Verification example:
1. simulation process
In order to verify the effectiveness of the algorithm, a public data set REIFT is selected as experimental data, a non-invasive load monitoring method model based on LMD and k neighbor algorithms is built on a MATLAB platform, and the hardware platform is a 64-bit Windows system computer with an Inter (R) Core (TM) i7-8700 CPU (3.20 GHz) and a 16GB memory. A part of appliances and a part of models in the REIFT data set were selected, and the processed appliance data are shown in table 1.
TABLE 1
And carrying out algorithm identification verification on 11 types of household appliances in the table, randomly extracting 100 groups of acquired data in each large type of appliance to be used as a test set, and dividing a training set and the test set according to the number of characteristic samples of each type of electric equipment in equal proportion, wherein the training set accounts for 30%. The content shown in table 2 is the result of the conventional kNN algorithm after identification, and the overall identification accuracy can reach 93.5% after calculation; table 3 shows the identification result obtained by the method, and the overall accuracy can reach 96.3%.
TABLE 2kNN identification results
Table 3 identification results based on KMD and improved kNN algorithm
In order to further verify the superiority of the algorithm, various common classification algorithms are compared with the algorithm of the application in terms of identification accuracy, the experimental data adopts part of electrical appliance data in the REIFT public data set, each algorithm is repeatedly tested 100 times, and the arithmetic average value of the overall identification accuracy is taken, and the result is shown in figure 1.
TABLE 4 identification accuracy of different algorithms
| Contrast algorithm | The algorithm of the application | SVM | BP | LSTM |
| Training time | / | 0.052 | 5.62 | 183.9 |
| Predicting time consumption/s | 0.076 | 0.045 | 0.012 | 0.52 |
| Accuracy/% | 96.3 | 94.5 | 94.65 | 96.0 |
The analysis from the above graph shows that the algorithm provided by the application has higher reliability of identification accuracy, smaller fluctuation of results under 100 tests, higher credibility than other algorithms and better reproducibility in practical application. The algorithm has the highest identification accuracy and better identification performance.
2. Analysis of results
As can be seen from table 4, the KMD and kNN-modified algorithm incorporates a plurality of different basic algorithms and is modified, the average value of the accuracy is as high as 96.3%, which is slightly higher than the above comparative algorithm and 96.0% of the LSTM algorithm, but the model of LSTM requires a lot of time for training; in addition, the prediction consumption of the algorithm is more compared with that of the SVM and BP algorithms, but the BP and SVM neural networks require more training time, and the model collocation is complex. The implementation proves that the household load identification accuracy is effectively improved by the algorithm.
The application discloses a non-invasive load monitoring method based on LMD and k neighbor algorithms, which comprises the following steps: collecting load power data; collecting data such as load voltage and current; processing the data based on an LMD algorithm, decomposing the active power load characteristic, and analyzing the decomposed components to obtain the load characteristic of the electrical equipment; establishing a k-nearest neighbor algorithm model, adopting Euclidean distance as a distance measurement formula of a sample to be tested and all samples in a training set, and improving the k-nearest neighbor algorithm model by adding a distance weight index to reduce errors caused by error disturbance; and finally, calculating the approximation degree of the feature vector of the sample to be detected and the mean value point of the sample of the same category of the classification result in the feature library, and combining the k nearest neighbor algorithm with a cosine similarity judging mechanism to achieve the aim of identifying strange load. The application can improve the accuracy of non-invasive load monitoring, reduce the calculated amount and improve the convergence rate of the model.
While the application has been described with reference to the preferred embodiments, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the application, and it is intended that the scope of the application shall be defined by the appended claims.
Claims (5)
1. A non-invasive load monitoring method based on LMD and k neighbor algorithm is characterized in that: the method comprises the following steps:
step one, acquiring a given test sample which is power data acquired by a user side; the power data comprise time sequence data of current and voltage;
the data obtained in the first step is adaptively decomposed into multiple product components on the premise of keeping the amplitude and frequency change characteristics of original signals through an LMD algorithm, and meanwhile, the low-dimensional active power characteristics in the data set are decomposed into multi-dimensional characteristic components, so that the detailed load characteristics of the electrical equipment corresponding to the power data are obtained;
step three, an improved k-nearest neighbor algorithm model is established, a test sample is given, euclidean distance between the sample to be tested obtained in the step two and a plurality of adjacent test samples is calculated, the label with the largest category number is identified as the label of test set data, and the distance weight value is added for reclassifying so as to reduce load identification errors caused by error disturbance in the data collection process;
step four: acquiring power data of a user side, decomposing the power data based on the characteristics of the second step, and carrying the power data into a model trained in the third step;
and fifthly, performing approximation degree judgment on the classification result obtained in the fourth step by adopting a cosine function, calculating approximation degree of the feature vector of the sample to be detected and the mean value point of the same class sample of the classification result in the feature library, comparing the approximation degree with a similarity threshold value, and finally obtaining an identification result.
2. A non-invasive load monitoring method based on LMD and k-nearest neighbor algorithm according to claim 1, characterized in that: step four, the power data of the user side is collected through the data collection equipment; the data acquisition equipment comprises intelligent electric equipment and an intelligent ammeter.
3. A non-invasive load monitoring method based on LMD and k-nearest neighbor algorithm according to claim 1, characterized in that: the second step specifically comprises the following steps:
s21: processing the data, decomposing the collected active power characteristics of the electrical equipment by analyzing the training sample data set, and calculating the local average function m of the power signal P (t) of each acquisition period 11 (t) and envelope estimation function a 11 (t); wherein the local average value m of the jth acquisition period j And local amplitude a j This can be obtained by the following equation:
(1) Wherein n is j ，n j+1 Respectively representing local extreme points of power data of j and j+1 acquisition periods;
sequentially connecting the local mean values m according to the time sequence j And local amplitude a j To obtain a local mean function m j (t) and envelope estimation function a j (t)；
S22, calculating the signal S as in formula (2) 11 (t) and S 11 (t) repeating the above-mentioned process as an original signal to iteratively calculate S 12 (t)……S 1n (t) up to S 1n (t) is a pure FM signal; in an iterative process, a series of FM signals as in equation (2) will be generated;
s23 by dividing the envelope signal a 1 (t) and FM signal S 1n (t) multiplying to obtain a first pf component p 1 (t); as shown in the following formula:
P 1 (t)＝a 1 (t)S 1n (t) (3)
and (3) sequentially obtaining the PF component k times of loops based on the formula: p is p 2 (t)、p 3 (t)…p k (t) and its corresponding balancing function u (t);
after k cyclic iterations of the decomposition, the original power signal is decomposed into k PF components and a margin u k (t), i.e
Thereby realizing the decomposition of the low-dimensional active power characteristic in the data set into the multidimensional characteristic component based on the LMD decomposition.
4. A non-invasive load monitoring method based on LMD and k-nearest neighbor algorithm according to claim 1, characterized in that: the third step comprises the following steps:
s31, dividing a given test sample into a training set and a verification set, and establishing a traditional k-nearest neighbor algorithm model; for the samples to be tested in the verification set, analyzing the samples to be tested and all samples z in the training set j Is a distance of (2); the Euclidean distance is used as a distance measurement formula of a k-nearest neighbor algorithm:
(6) Wherein d euc Representing the distances between the sample to be tested and all samples in the training set; z is Z l Representing an L-th feature of a sample to be classified;representing the first characteristic of the jth training sample, and m represents the characteristic number of the training sample;
arranging the calculation results of the distances between all the samples to be tested and all the samples in the training set according to increasing orderSelecting k points in the training set, wherein the distance between the k points and the sample Z to be detected is the smallest, and marking the neighborhood of the sample Z to be detected containing the k points as N k (Z); finally at N k (Z) identifying the label with the largest number of categories as the category c of the sample Z to be tested Z The classification voting function is as follows;
in the formula (7), G is the sum of all sample types; class (c) zj ) For training sample z j The species of (2); i (v=class (c) zj ) If the value is false, returning to 0, otherwise returning to 1, determining the most category in k neighbor data through superposition calculation, and considering the sample to be detected as the category;
s32, selecting the first k samples according to the distance increment sequence, and calculating the distance weight W of the k samples j The method comprises the steps of carrying out a first treatment on the surface of the Selecting the category with the largest overlapping distance weight index, namely the improved category c of the sample to be detected, according to the principle of classified overlapping on the calculated distance weight index z ' its improvement classification voting function improves as:
wherein G is a set of all sample categories; class (c) zj ) For training sample z j Is a category of (2);as a logic function, returns 0 when its value is false, otherwise returns 1.
5. A non-invasive load monitoring method based on LMD and k-nearest neighbor algorithm according to claim 1, characterized in that: the fifth step is specifically as follows: selecting the category with the maximum weight index through an improved classification voting function, and calculating the approximation degree of the characteristic vector of the sample to be detected and the mean value point of the sample with the same category of the classification result in the characteristic library; if the approximation degree value of the two is larger than alpha, the classification is considered to be correct, and a classification result is obtained; if the cosine function similarity value is smaller than alpha, the load is not the database load and is regarded as strange equipment; alpha is a preset approximation value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310485699.4A CN116720127A (en) | 2023-04-28 | 2023-04-28 | Non-invasive load monitoring method based on LMD and k nearest neighbor algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310485699.4A CN116720127A (en) | 2023-04-28 | 2023-04-28 | Non-invasive load monitoring method based on LMD and k nearest neighbor algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116720127A true CN116720127A (en) | 2023-09-08 |
Family
ID=87872315
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310485699.4A Pending CN116720127A (en) | 2023-04-28 | 2023-04-28 | Non-invasive load monitoring method based on LMD and k nearest neighbor algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116720127A (en) |
-
2023
- 2023-04-28 CN CN202310485699.4A patent/CN116720127A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112699913B (en) | Method and device for diagnosing abnormal relationship of household transformer in transformer area | |
| Wang et al. | A novel weighted sparse representation classification strategy based on dictionary learning for rotating machinery | |
| CN109271975B (en) | Power quality disturbance identification method based on big data multi-feature extraction collaborative classification | |
| CN109543943B (en) | Electric price checking execution method based on big data deep learning | |
| CN102435910B (en) | Power electronic circuit health monitoring method based on support vector classification | |
| CN110336534A (en) | A kind of method for diagnosing faults based on photovoltaic array electric parameter time series feature extraction | |
| CN113962259B (en) | Multi-mode double-layer fault diagnosis method for fuel cell system | |
| CN109359665B (en) | Household appliance load identification method and device based on support vector machine | |
| CN110991263B (en) | Non-invasive load identification method and system for resisting background load interference | |
| CN104966161A (en) | Electric energy quality recording data calculating analysis method based on Gaussian mixture model | |
| CN111680726B (en) | Transformer fault diagnosis method and system based on neighbor component analysis and k neighbor learning fusion | |
| CN111639586B (en) | Non-invasive load identification model construction method, load identification method and system | |
| CN111275069B (en) | Non-invasive load monitoring method | |
| CN111177216A (en) | Association rule generation method and device for behavior characteristics of comprehensive energy consumer | |
| CN111311434B (en) | Method and device for separating loads of electric equipment, computer equipment and storage medium | |
| CN114881077A (en) | Voltage sag source classification method and system based on time sequence trajectory characteristics | |
| CN111090679B (en) | Time sequence data representation learning method based on time sequence influence and graph embedding | |
| CN113762355A (en) | User abnormal electricity consumption behavior detection method based on non-invasive load decomposition | |
| CN112163595B (en) | Method and device for acquiring typical electricity utilization mode of user and electronic equipment | |
| CN113327008A (en) | Electricity stealing detection method, system and medium based on time sequence automatic encoder | |
| Yu et al. | Rolling bearing degradation state identification based on LCD relative spectral entropy | |
| CN113076354A (en) | User electricity consumption data analysis method and device based on non-invasive load monitoring | |
| Jazizadeh et al. | Unsupervised clustering of residential electricity consumption measurements for facilitated user-centric non-intrusive load monitoring | |
| Mariño et al. | NILM: Multivariate DNN performance analysis with high frequency features | |
| CN116720127A (en) | Non-invasive load monitoring method based on LMD and k nearest neighbor algorithm |
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 |