CN107330517B - Non-invasive resident load identification method based on S _ Kohonen - Google Patents
Non-invasive resident load identification method based on S _ Kohonen Download PDFInfo
- Publication number
- CN107330517B CN107330517B CN201710448867.7A CN201710448867A CN107330517B CN 107330517 B CN107330517 B CN 107330517B CN 201710448867 A CN201710448867 A CN 201710448867A CN 107330517 B CN107330517 B CN 107330517B
- Authority
- CN
- China
- Prior art keywords
- load
- layer
- network
- electric appliance
- kohonen
- 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
Images
Classifications
-
- 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
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
- G01R19/0092—Arrangements for measuring currents or voltages or for indicating presence or sign thereof measuring current only
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R23/00—Arrangements for measuring frequencies; Arrangements for analysing frequency spectra
- G01R23/16—Spectrum analysis; Fourier analysis
-
- 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
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a non-invasive resident load identification method based on S _ Kohonen, which comprises the following steps: the method comprises the following steps: judging a switching event according to the change of active power at a home power inlet, and collecting an electric appliance current sample of the switching event at the home power inlet when the switching event occurs; step two: carrying out frequency domain analysis on the collected electric appliance current samples, and extracting frequency domain harmonic amplitudes of the electric appliance current samples as load characteristics of each electric appliance to form a load characteristic library; step three: designing an S _ Kohonen neural network suitable for identifying the load of residents, determining the number of neurons of an input layer and an output layer of the S _ Kohonen neural network and the scale of a competition layer, and determining a selection mode of winning neurons and a learning algorithm for weight adjustment; step four: initializing parameters; step five: training the S _ Kohonen network by using a training set, and testing by using a testing set; step six: network parameters are adjusted to achieve optimal network performance.
Description
Technical Field
The invention relates to the technical field of power grid load monitoring, in particular to a non-intrusive resident load identification method based on S _ Kohonen.
Background
The development of smart grid technology enables power consumers to coordinate with the grid. The intelligent power utilization is one of important links of an intelligent power grid and is the core of an interactive service system. To realize intelligent power utilization, power consumers need to better understand the power utilization characteristics of the power consumers and timely obtain the energy consumption information of the power utilization equipment. The load monitoring is a key technology for realizing intelligent power utilization, and the power utilization information of various power utilization equipment can be analyzed through the load monitoring, so that power utilization habits of power consumers are guided to be changed, power utilization behaviors are optimized, and the purpose of energy conservation is achieved. For the power grid, load monitoring can help the power grid to know load composition and master the power consumption behavior of power consumers, provide guidance for planning and designing the power grid and power generation scheduling, and support bidirectional interactive service and intelligent energy utilization service.
Load monitoring includes intrusive load monitoring and non-intrusive load monitoring. The method has high accuracy, but has high cost and difficult maintenance. The non-invasive load monitoring is that monitoring equipment is installed at an electric power inlet of an electric power user, and the electricity utilization information of the electric power user is analyzed through a monitoring algorithm, so that the electricity utilization condition of each device in the electric power user is obtained, the hardware structure is greatly simplified, the economic cost is reduced, and the non-invasive load monitoring is suitable for a large number of independent user installation modes.
The existing non-invasive load identification algorithm has the problems that the identification accuracy is low in load monitoring, and correct identification is difficult for small-power electric appliances, multi-state electric appliances and electric appliances with similar characteristics, so that implementation of non-invasive load monitoring and realization of intelligent power utilization are seriously influenced.
It is therefore desirable to have a non-intrusive resident load identification method that overcomes or at least alleviates the problem of low identification accuracy in load monitoring of the prior art non-intrusive load identification algorithms.
Disclosure of Invention
The invention aims to provide a non-invasive resident load identification method to solve the problems that the identification accuracy rate of household appliances is not high in non-invasive load monitoring, and small-power appliances, multi-state appliances and appliances with similar characteristics are difficult to correctly identify.
The invention provides a non-invasive resident load identification method, which comprises the following steps:
the method comprises the following steps: judging a switching event according to the change of active power at a home power inlet, and collecting an electric appliance current sample of the switching event at the home power inlet when the switching event occurs;
step two: carrying out frequency domain analysis on the collected electric appliance current samples, and extracting frequency domain harmonic amplitudes of the electric appliance current samples as load characteristics of each electric appliance to form a load characteristic library;
step three: designing an S _ Kohonen neural network suitable for identifying the load of residents, determining the number of neurons of an input layer and an output layer of the S _ Kohonen neural network and the scale of a competition layer, and determining a selection mode of winning neurons and a learning algorithm for weight adjustment;
step four: initializing parameters, wherein the parameters comprise: connection weight omega between input layer and competition layerijAnd the connection weight omega between the output layer and the competition layerjkNeighborhood radius r, ωijLearning rate η of1And ωjkLearning rate η of2；
Step five: the load characteristic vectors of the electric appliances are used as network input, the electric appliance categories are used as network output, an S _ Kohonen network is trained through a training set, a network is used for testing a test set sample after the training is finished, an identification result is obtained, and the identification accuracy and the overall identification accuracy of the electric appliances are calculated to test the network performance;
step six: and adjusting the scale of a competition layer of the S _ Kohonen neural network, a termination threshold or the maximum iteration number, researching the relation between the network parameters and the network performance, and selecting proper parameters to realize the optimal network performance.
Preferably, in the first step, the voltage, the current and the active power value are collected by a collecting device installed at the household power inlet.
Preferably, in the first step, according to a rule judgment method, when the change of the active power value at the home power inlet exceeds 20W, a switching event is judged to occur, a difference between a steady-state current value before the switching event occurs and a steady-state current value after the switching event occurs is used as a current sample of a load where the switching event occurs, and a type of an electrical appliance where the switching event occurs is recorded as an event tag.
Preferably, the second step is further to perform frequency domain analysis on the collected steady-state currents of the electrical appliances, extract frequency domain features of the electrical appliances through discrete fourier series expansion, project each harmonic in one current period into a frequency domain, and take the harmonic with a larger amplitude as the load feature of the electrical appliances.
Preferably, in the third step, the S _ Kohonen neural network includes an input layer, a competition layer and an output layer; the number of neurons in the input layer is the dimension of the load characteristic, and the competitionThe neuron of the contention layer is arranged in a two-dimensional array, the scale of the S _ Kohonen neural network is determined according to the load clustering number, and the number of the neuron of the output layer is the number of load categories to be identified in the resident home; the selection mode of the competitive layer winning neuron is an Euclidean distance method, namely, an input vector X is calculated to be (X)1,x2,…xm) Distance d to competition layer neuron jjM is the dimension of the input load feature vector, and the distance calculation formula is as follows:
selecting the competition layer neuron c with the minimum distance with the input vector X as a winning neuron, namely realizing the mapping between the input vector X and the competition layer neuron c; in addition, the weights of the winning neuron and its surrounding neighborhood neurons are adjusted, and the neighborhood is defined as:
Nc(j)＝(j|find(norm(posj,posc)＜r))
j＝1,2,…,l
posj、poscthe positions of neurons j and c, respectively; norm is the Euclidean distance between two neurons; r is the selected neighborhood radius, l is the number of competition layer neurons; best matching neuron c and its neighborhood Nc(j) Weight ω between inner neuron and input layerijAccording to XiAdjust to gradually approach XiAnd the weight ω of the output layerjkThen Y is output according to the load expectationkAdjusting:
ωij＝ωij+y(j)*η1(Xi-ωij)
ωjk＝ωjk+y(j)*η2(Yk-ωjk)
wherein y (j) is a learning algorithm for weight adjustment, and a chef cap learning function is adopted here:
that is, the weight of the neuron in the neighborhood of radius r and the winning neuron are adjusted in the same way, while the weight of the neuron outside the neighborhood is not adjusted, so that the neuron c and the neighborhood N are respectively adjustedc(j) Weight coefficient omega between included node and input layer nodeijAnd a weight coefficient omega with the output layer nodejk；η1And η2Are respectively the weight value omegaijAnd ωjkThe learning rate of (d); r and eta1Decreases with increasing evolution time, eta2The evolution time increases along with the increase of the evolution time, wherein n is the iteration time, and the change rules of the three are respectively as follows:
preferably, in step four, ωijInitialisation to a random number, omegajkInitialized to 0, η1max，η1min，η2maxAnd η2minTaking the value of 0-1, selecting the neighborhood radius r according to the scale of the competition layer, wherein rminThe value is 0-1.
Preferably, in the fifth step, the designed network is trained by using a training set sample, and the training is stopped when the variation of the overall recognition accuracy rate is smaller than a threshold value or the number of iterations is reached; and (3) carrying out identification test on the test set sample by using the trained network, and checking the identification accuracy and the overall identification accuracy of each electric appliance, wherein the accuracy is defined as follows:
the identification accuracy rate of a single electric appliance is equal to the correct identification number of the electric appliances/the number of the electric appliance samples,
and the total identification accuracy rate is the correct identification number of all the electric appliances/the total sample number of all the electric appliances.
Preferably, in the sixth step, the relationship between the network parameter and the network performance is training time, single event recognition time and overall recognition accuracy of the network under different parameters, so as to improve working efficiency and select a proper parameter setting for online load monitoring.
The non-invasive resident load identification method of the invention carries out discrete Fourier transform on the steady-state working current of the electric appliance, extracts the amplitude of the harmonic current in the frequency domain as the load characteristic, reduces the requirement on the sampling frequency of the acquisition device and further reduces the investment of hardware equipment. The non-invasive resident load identification method projects the load characteristics to a two-dimensional competition layer plane, enables the intra-class area of the competition layer projected by the load characteristics to be continuously reduced according to the weight adjustment algorithm, and simultaneously gradually enlarges the distance between classes, so that the load clustering area with similar characteristics is gradually far away. The invention can improve the identification accuracy of similar electrical appliances, and simultaneously, the neurons in each type of load region have the same output by adjusting the weight of the output layer, so that the classification of the load has fault tolerance, namely, the load characteristics of the same type are allowed to slightly fluctuate, thereby still realizing accurate identification of the load when the voltage of a power grid fluctuates slightly, and enhancing the practicability of the algorithm. The invention can project different working states of the electric appliance to different areas, and the areas correspond to the same output, thus realizing the identification of the multi-state electric appliance.
In conclusion, the non-invasive resident load identification method has good stability, fault tolerance and practicability, can solve the problems that the identification accuracy rate of household appliances is not high in non-invasive load monitoring, and small-power appliances, multi-state appliances and appliances with similar characteristics are difficult to correctly identify, realizes effective identification on resident appliances, and can be widely applied to resident power load identification.
Drawings
FIG. 1 is a diagram of an S _ Kohonen neural network model.
Fig. 2 is a flowchart of a non-intrusive load identification method according to the present invention.
FIG. 3 is a schematic diagram of the aggregation of the competition layers of electrical appliances with similar characteristics according to an embodiment of the present invention.
FIG. 4 is a schematic diagram of training set sample competition layer clustering according to an embodiment of the present invention.
FIG. 5 is a diagram illustrating the test set sample competition layer clustering according to an embodiment of the present invention.
Detailed Description
In order to make the implementation objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings in the embodiments of the present invention. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are only some, but not all embodiments of the invention. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention. Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
As shown in fig. 1-2, the present invention provides a non-intrusive resident load recognition method based on S _ Kohonen, comprising the steps of:
the method comprises the following steps: judging a switching event according to the change of active power at a home power inlet, collecting a current sample of an electric appliance with the switching event at the home power inlet when the switching event occurs, collecting voltage, current and active power values by using a collecting device arranged at the home power inlet, judging the switching event to occur according to a rule judgment method when the change of the active power value at the home power inlet exceeds 20W, taking the difference between a steady-state current value before the switching event occurs and a steady-state current value after the switching event occurs as a current sample of a load with the switching event, and recording the type of the electric appliance with the switching event as an event label;
step two: carrying out frequency domain analysis on the collected current samples of the electric appliances, simultaneously carrying out frequency domain analysis on the collected steady-state currents of the electric appliances, extracting frequency domain characteristics of the electric appliances through discrete Fourier series expansion, projecting each harmonic in a current period into a frequency domain, and taking the harmonic with larger amplitude as a load characteristic to form a load characteristic library of each electric appliance;
step three: designing an S _ Kohonen neural network suitable for identifying the load of residents, determining the number of neurons of an input layer and an output layer of the S _ Kohonen neural network and the scale of a competition layer to determine a selection mode of winning neurons and a learning algorithm for weight adjustment, wherein the S _ Kohonen neural network comprises an input layer, a competition layer and an output layer; the number of neurons in the input layer is the dimension of the load characteristic, the neurons in the competition layer are arranged in a two-dimensional array, the scale of the S _ Kohonen neural network is determined according to the number of load clusters, and the number of neurons in the output layer is the number of load categories to be identified in residents; the selection mode of the competitive layer winning neuron is an Euclidean distance method, namely, an input vector X is calculated to be (X)1,x2,…xm) Distance d to competition layer neuron jjM is the dimension of the input load feature vector, and the distance calculation formula is as follows:
selecting the competition layer neuron c with the minimum distance with the input vector X as a winning neuron, namely realizing the mapping between the input vector X and the competition layer neuron c; in addition, the weights of the winning neuron and its surrounding neighborhood neurons are adjusted, and the neighborhood is defined as:
Nc(j)＝(j|find(norm(posj,posc)＜r))
j＝1,2,…,l
posj、poscthe positions of neurons j and c, respectively; norm is the Euclidean distance between two neurons; r is the selected neighborhood radius, l is the number of competition layer neurons; best matching neuron c and its neighborhood Nc(j) Weight ω between inner neuron and input layerijAccording to XiThe adjustment is carried out by the following steps,gradually trend towards XiAnd the weight ω of the output layerjkThen Y is output according to the load expectationkAdjusting:
ωij＝ωij+y(j)*η1(Xi-ωij)
ωjk＝ωjk+y(j)*η2(Yk-ωjk)
wherein y (j) is a learning algorithm for weight adjustment, and a chef cap learning function is adopted here:
that is, the weight of the neuron in the neighborhood of radius r and the winning neuron are adjusted in the same way, while the weight of the neuron outside the neighborhood is not adjusted, so that the neuron c and the neighborhood N are respectively adjustedc(j) Weight coefficient omega between included node and input layer nodeijAnd a weight coefficient omega with the output layer nodejk；η1And η2Are respectively the weight value omegaijAnd ωjkThe learning rate of (d); r and eta1Decreases with increasing evolution time, eta2The evolution time increases along with the increase of the evolution time, wherein n is the iteration time, and the change rules of the three are respectively as follows:
step four: initializing parameters, wherein the parameters comprise: connection weight omega between input layer and competition layerijAnd the connection weight omega between the output layer and the competition layerjkNeighborhood radius r, ωijSpeed of learningRate eta1And ωjkLearning rate η of2，ωijInitialisation to a random number, omegajkInitialized to 0, η1max，η1min，η2maxAnd η2minTaking the value of 0-1, selecting the neighborhood radius r according to the scale of the competition layer, wherein rminTaking the value of 0-1;
step five: the method comprises the steps that load characteristic vectors of all electrical appliances are used as network input, the categories of the electrical appliances are used as network output, an S _ Kohonen network is trained through a training set, a network is used for testing a test set sample after training is finished to obtain a recognition result, the recognition accuracy and the total recognition accuracy of all the electrical appliances are calculated to test the network performance, the designed network is trained through the training set sample, and training is stopped when the variation of the total recognition accuracy is smaller than a threshold value or the iteration times are reached; and (3) carrying out identification test on the test set sample by using the trained network, and checking the identification accuracy and the overall identification accuracy of each electric appliance, wherein the accuracy is defined as follows:
the identification accuracy rate of a single electric appliance is equal to the correct identification number of the electric appliances/the number of the electric appliance samples,
the total identification accuracy rate is the correct identification number of all the electric appliances/the total sample number of all the electric appliances;
step six: adjusting the scale of a competition layer of the S _ Kohonen neural network, a termination threshold or the maximum iteration number, researching the relation between the network parameters and the network performance, and selecting the appropriate parameters to realize the optimal network performance, wherein the relation between the network parameters and the network performance is the training time, the single event recognition time and the overall recognition accuracy of the network under different parameters, so that the appropriate parameter setting is selected for improving the working efficiency and realizing the online load monitoring.
An embodiment of the present invention:
as shown in fig. 3-5, simulation tests were performed using a blue database published by university of tomilong in the card, which continuously collects data of voltage, current, power, etc. for one week at power ports of 1 U.S. home using a sampling frequency of 12kHz, and marks the switching time of each electrical appliance. And taking the difference of steady-state currents at the front and rear ports of each switching event as a switching load current sample, extracting 100 load current samples from each switching event, and correcting the current waveform phase according to the voltage. The training data adopts 500 load switching events, and 5 ten thousand load current samples in total are used as a training set; the test data adopts another 232 load switching events in the database, and 2.32 ten thousand load current samples in total are taken as a test set. The training set and the test set each contained 7 appliance categories. Among the 7 kinds of electric appliances, the refrigerator, the backyard lamp, the bathroom ceiling lamp and the bedroom lamp are low-power electric appliances; kitchen choppers, refrigerators, etc. have a variety of operating conditions; kitchen choppers and blowers and certain conditions of refrigerators and bathroom headlights are two sets of appliances with similar characteristics, with similar current waveforms and load characteristics.
According to the steps of the non-invasive resident load identification method based on the S _ Kohonen, various types of load switching events in BLUED are identified, and a parameter eta is given during the training process of the S _ Kohonen network1min＝0.01,η1max＝0.1；η2min＝0.5,η2max＝1；rmax＝5,rmin0.2; taking the scale of a competition layer as 50 multiplied by 50; iterating for 10 times, wherein each iteration trains 50000 samples; after training, carrying out recognition test on 23200 load samples in the test set, and calculating the recognition accuracy, wherein the test results are shown in table 1:
TABLE 1 recognition result of S _ Kohonen neural network for household appliances
As can be seen from Table 1, the method of the invention has very high identification accuracy for various types of loads, and can realize high-precision identification for small-power electric appliances, multi-state electric appliances and electric appliances with similar characteristics.
In addition, the number of sample iterations and the scale of competition layers during training of the S _ Kohonen network were adjusted, and the influence of the number of sample iterations and the scale of competition layers on the performance of the S _ Kohonen network was studied.
TABLE 2 Overall accuracy vs. iteration number relationship
| Number of iterations | 1 | 2 | 3 | 4 | 5 |
| Overall rate of accuracy | 16.20％ | 99.62％ | 99.78％ | 99.86％ | 99.92％ |
| Number of iterations | 6 | 7 | 8 | 9 | 10 |
| Overall rate of accuracy | 99.89％ | 99.96％ | 99.92％ | 99.89％ | 99.97％ |
As can be seen from table 2, the overall recognition accuracy is already high in the 2 nd iteration, and the overall accuracy is basically stabilized at about 99.9% after the 5 th iteration, so that the network stability is good.
TABLE 3 network Performance vs. Contention layer Scale
From a study of two-dimensional arrays of competition layer sizes from 10 × 10 to 80 × 80, it can be seen from table 3 that as competition layer sizes increase, network training time, individual event recognition time, and overall accuracy all increase. After the competitive layer size is 40 multiplied by 40, the overall recognition accuracy rate is basically stabilized to be about 99.97%, the load recognition capability of the network is basically stabilized, and the training time and the single event recognition time continue to increase. By researching the relation between the network performance and the iteration times and the scale of the competition layer and adjusting the optimal parameters, the comprehensive performance of the network can be optimal.
The invention can realize high-precision identification of the load of residents, and solves the problem that the current small-power electric appliances, multi-working-state electric appliances and electric appliances with similar characteristics can not be accurately identified. And properly adjusting network parameters to enable the network performance to be suitable for load online monitoring.
Finally, it should be pointed out that: the above examples are only for illustrating the technical solutions of the present invention, and are not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (3)
1. A non-invasive resident load identification method based on S _ Kohonen is characterized by comprising the following steps:
the method comprises the following steps: judging a switching event according to the change of active power at a home power inlet, collecting a current sample of an electric appliance with the switching event at the home power inlet when the switching event occurs, specifically, collecting voltage, current and active power values by using a collecting device arranged at the home power inlet, judging that the switching event occurs when the change of the active power value at the home power inlet exceeds 20W according to a rule judgment method, taking the difference between a steady-state current value before the switching event occurs and a steady-state current value after the switching event occurs as a current sample of a load with the switching event, and recording the type of the electric appliance with the switching event as an event label;
step two: carrying out frequency domain analysis on the collected electric appliance current samples, and extracting frequency domain harmonic amplitudes of the electric appliance current samples as load characteristics of each electric appliance to form a load characteristic library;
step three: designing an S _ Kohonen neural network suitable for identifying the load of residents, determining the number of neurons of an input layer and an output layer of the S _ Kohonen neural network and the scale of a competition layer, and determining a selection mode of winning neurons and a learning algorithm for weight adjustment; the S _ Kohonen neural network comprises an input layer, a competition layer and an output layer; the number of neurons in the input layer is the dimension of the load characteristic, the neurons in the competition layer are arranged in a two-dimensional array, the scale of the S _ Kohonen neural network is determined according to the number of load clusters, and the number of neurons in the output layer is the number of load categories to be identified in residents; the selection mode of the competitive layer winning neuron is an Euclidean distance method, namely, an input vector X is calculated to be (X)1，x2，...xm) Distance d to competition layer neuron jjM is the dimension of the input load feature vector, and the distance calculation formula is as follows:
selecting the competition layer neuron c with the minimum distance with the input vector X as a winning neuron, namely realizing the mapping between the input vector X and the competition layer neuron c; in addition, the weights of the winning neuron and its surrounding neighborhood neurons are adjusted, and the neighborhood is defined as:
Nc(j)＝(j|find(norm(posj，posc)＜r))
j＝1，2，...，l
posj、poscthe positions of neurons j and c, respectively; norm is the Euclidean distance between two neurons; r is the selected neighborhood radius, l is the number of competition layer neurons; best matching neuron c and its neighborhood Nc(j) Weight ω between inner neuron and input layerijAccording to XiAdjust to gradually approach XiAnd the weight ω of the output layerjkThen Y is output according to the load expectationkAdjusting:
ωij＝ωij+y(j)*η1(Xi-ωij)
ωjk＝ωjk+y(j)*η2(Yk-ωjk)
wherein y (j) is a learning algorithm for weight adjustment, and a chef cap learning function is adopted here:
that is, the weight of the neuron in the neighborhood of radius r and the winning neuron are adjusted in the same way, while the weight of the neuron outside the neighborhood is not adjusted, so that the neuron c and the neighborhood N are respectively adjustedc(j) Weight coefficient omega between included node and input layer nodeijAnd a weight coefficient omega with the output layer nodejk；η1And η2Are respectively the weight value omegaijAnd ωjkThe learning rate of (d); r and eta1Decreases with increasing evolution time, eta2The evolution time increases along with the increase of the evolution time, wherein n is the iteration time, and the change rules of the three are respectively as follows:
step four: initializing parameters, wherein the parameters comprise: connection weight omega between input layer and competition layerijAnd the connection weight omega between the output layer and the competition layerjkNeighborhood radius r, ωijLearning rate η of1And ωjkLearning rate η of2Wherein, ω isijInitialisation to a random number, omegajkInitialized to 0, η1max，η1min，η2maxAnd η2minTaking the value of 0-1, selecting the neighborhood radius r according to the scale of the competition layer, wherein rminTaking the value of 0-1;
step five: the method comprises the steps that load characteristic vectors of all electrical appliances are used as network input, the categories of the electrical appliances are used as network output, an S _ Kohonen network is trained through a training set, a network is used for testing a test set sample after training is finished to obtain a recognition result, the recognition accuracy and the overall recognition accuracy of all the electrical appliances are calculated to test the network performance, specifically, the training set sample is used for training a designed network, and the training is stopped when the variation of the overall recognition accuracy is smaller than a threshold value or the iteration times are reached; and (3) carrying out identification test on the test set sample by using the trained network, and checking the identification accuracy and the overall identification accuracy of each electric appliance, wherein the accuracy is defined as follows:
the identification accuracy rate of a single electric appliance is equal to the correct identification number of the electric appliances/the number of the electric appliance samples,
the total identification accuracy rate is the correct identification number of all the electric appliances/the total sample number of all the electric appliances;
step six: and adjusting the scale of a competition layer of the S _ Kohonen neural network, a termination threshold or the maximum iteration number, researching the relation between the network parameters and the network performance, and selecting proper parameters to realize the optimal network performance.
2. The non-intrusive resident load recognition method based on S _ Kohonen according to claim 1, wherein: and secondly, carrying out frequency domain analysis on the acquired steady-state current of each electric appliance, extracting frequency domain characteristics of each electric appliance through discrete Fourier series expansion, projecting each harmonic in a current period into a frequency domain, and taking the harmonic with larger amplitude as the load characteristics of each electric appliance.
3. The non-intrusive resident load recognition method based on S _ Kohonen according to claim 1, wherein: in the sixth step, the relationship between the network parameters and the network performance is the training time, the single event recognition time and the overall recognition accuracy of the network under different parameters, so as to improve the working efficiency and select proper parameter settings for online load monitoring.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710448867.7A CN107330517B (en) | 2017-06-14 | 2017-06-14 | Non-invasive resident load identification method based on S _ Kohonen |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710448867.7A CN107330517B (en) | 2017-06-14 | 2017-06-14 | Non-invasive resident load identification method based on S _ Kohonen |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107330517A CN107330517A (en) | 2017-11-07 |
| CN107330517B true CN107330517B (en) | 2021-08-06 |
Family
ID=60194732
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710448867.7A Active CN107330517B (en) | 2017-06-14 | 2017-06-14 | Non-invasive resident load identification method based on S _ Kohonen |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107330517B (en) |
Families Citing this family (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108333423A (en) * | 2017-12-27 | 2018-07-27 | 安徽机电职业技术学院 | Non-intrusion type residential power load testing method |
| CN108399221B (en) * | 2018-02-11 | 2021-07-30 | 山东建筑大学 | Indoor electrical equipment classification and identification method and system based on big data correlation analysis |
| CN108664876A (en) * | 2018-03-05 | 2018-10-16 | 佛山科学技术学院 | A kind of recognition methods of electrical appliance |
| CN108429254B (en) * | 2018-03-07 | 2020-01-31 | 国网四川省电力公司电力科学研究院 | non-invasive electric load identification method |
| CN108537385A (en) * | 2018-04-12 | 2018-09-14 | 广东电网有限责任公司 | A kind of non-intrusion type residential electricity consumption load recognition methods |
| CN109165604A (en) * | 2018-08-28 | 2019-01-08 | 四川大学 | The recognition methods of non-intrusion type load and its test macro based on coorinated training |
| CN109617067A (en) * | 2019-01-16 | 2019-04-12 | 江苏圣通电力新能源科技有限公司 | Non-intrusive electrical load recognition methods neural network based |
| CN109934303A (en) * | 2019-03-25 | 2019-06-25 | 宁夏隆基宁光仪表股份有限公司 | A kind of non-invasive household electrical appliance load recognition methods, device and storage medium |
| CN109993424B (en) * | 2019-03-26 | 2023-06-23 | 广东艾胜物联网科技有限公司 | Non-interference type load decomposition method based on width learning algorithm |
| CN110119816A (en) * | 2019-04-15 | 2019-08-13 | 东南大学 | A kind of load characteristic self-learning method suitable for non-intrusion type electric power monitoring |
| CN110569876A (en) * | 2019-08-07 | 2019-12-13 | 武汉中原电子信息有限公司 | Non-invasive load identification method and device and computing equipment |
| CN110751385B (en) * | 2019-10-08 | 2023-05-05 | 威胜集团有限公司 | Non-invasive load identification method, terminal device and storage medium |
| CN110991263B (en) * | 2019-11-12 | 2022-03-18 | 华中科技大学 | Non-invasive load identification method and system for resisting background load interference |
| CN111027408A (en) * | 2019-11-19 | 2020-04-17 | 广西电网有限责任公司电力科学研究院 | Load identification method based on support vector machine and V-I curve characteristics |
| CN110956220B (en) * | 2019-12-11 | 2023-10-31 | 深圳市活力天汇科技股份有限公司 | Non-invasive household appliance load identification method |
| CN110889465B (en) * | 2019-12-12 | 2020-11-03 | 武汉大学 | Power demand side equipment identification method and system based on self-adaptive resonant network |
| CN111158239B (en) * | 2020-01-09 | 2022-06-21 | 上海电力大学 | Association rule algorithm and neural network-based pulverizing system performance optimization method |
| CN111461189B (en) * | 2020-03-23 | 2022-11-15 | 东南大学 | Resident rhythm detection method based on non-invasive measurement technology |
| CN111476298B (en) * | 2020-04-07 | 2022-04-26 | 杭州国彪超声设备有限公司 | Power load state identification method in home and office environment |
| CN111563827B (en) * | 2020-04-10 | 2023-08-04 | 杭州电子科技大学 | Load decomposition method based on physical characteristics of electrical appliance and electricity utilization behaviors of residents |
| CN111614159A (en) * | 2020-05-12 | 2020-09-01 | 国网河北省电力有限公司电力科学研究院 | Non-invasive household load identification method for demand side management |
| CN112016587A (en) * | 2020-07-09 | 2020-12-01 | 冯燕钧 | Energy consumption monitoring cloud collaborative non-invasive identification method based on master station feature library technology |
| CN112180193B (en) * | 2020-09-28 | 2021-11-02 | 华中科技大学 | Non-invasive load identification system and method based on track image identification |
| CN112396098A (en) * | 2020-11-12 | 2021-02-23 | 天津天大求实电力新技术股份有限公司 | Non-embedded apartment electrical appliance load identification and analysis method, system and application |
| CN113033775B (en) * | 2021-03-10 | 2023-08-18 | 南方电网数字电网研究院有限公司 | Non-invasive load identification network architecture based on supervised learning |
| CN113554483A (en) * | 2021-07-19 | 2021-10-26 | 浙江工企信息技术股份有限公司 | Order full life cycle management method and system |
| 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 |
| CN114184870A (en) * | 2021-12-14 | 2022-03-15 | 河北科技大学 | Non-invasive load identification method and equipment |
| CN115456034A (en) * | 2022-11-09 | 2022-12-09 | 广东浩迪创新科技有限公司 | Automatic identification and monitoring method and system for electric bicycle charging |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016079229A1 (en) * | 2014-11-21 | 2016-05-26 | Universite Libre De Bruxelles | Improved non-intrusive appliance load monitoring method and device |
| CN106093630B (en) * | 2016-06-02 | 2019-01-15 | 华北电力大学 | A kind of non-intrusion type household electrical appliance discrimination method |
-
2017
- 2017-06-14 CN CN201710448867.7A patent/CN107330517B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN107330517A (en) | 2017-11-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107330517B (en) | Non-invasive resident load identification method based on S _ Kohonen | |
| Hassan et al. | An empirical investigation of VI trajectory based load signatures for non-intrusive load monitoring | |
| Mahela et al. | Recognition of power quality disturbances using S-transform based ruled decision tree and fuzzy C-means clustering classifiers | |
| CN110956220B (en) | Non-invasive household appliance load identification method | |
| CN109598451B (en) | Non-invasive load identification method based on PCA (principal component analysis) and LSTM (least Square TM) neural network | |
| Rigoni et al. | Representative residential LV feeders: A case study for the North West of England | |
| CN110504679B (en) | KM matching algorithm-based non-invasive load identification method | |
| Li et al. | Long-term system load forecasting based on data-driven linear clustering method | |
| He et al. | Invisible units detection and estimation based on random matrix theory | |
| Munshi et al. | Unsupervised nonintrusive extraction of electrical vehicle charging load patterns | |
| CN112580471A (en) | Non-invasive load identification method based on AdaBoost feature extraction and RNN model | |
| CN109359665B (en) | Household appliance load identification method and device based on support vector machine | |
| Jia et al. | Non-intrusive load identification using reconstructed voltage–current images | |
| CN108345906B (en) | Non-invasive electrical appliance identification method based on Boost model | |
| CN103439573A (en) | Method and system for identifying household loads based on transient characteristic close degree matching | |
| de Paiva Penha et al. | Convolutional neural network applied to the identification of residential equipment in non-intrusive load monitoring systems | |
| Dinesh et al. | Incorporating time-of-day usage patterns into non-intrusive load monitoring | |
| Kangping et al. | Analysis on residential electricity consumption behavior using improved k-means based on simulated annealing algorithm | |
| Hernandez et al. | Development of a non-intrusive load monitoring (nilm) with unknown loads using support vector machine | |
| Zarabie et al. | Load profile based electricity consumer clustering using affinity propagation | |
| Lingqing et al. | Detection method for power theft based on SOM neural network and K-means clustering algorithm | |
| Dinesh et al. | Individual power profile estimation of residential appliances using low frequency smart meter data | |
| Dinesh et al. | Non-intrusive load monitoring based on low frequency active power measurements | |
| CN113591960A (en) | Voltage sag event type identification method based on improved generation countermeasure network | |
| Dinesh et al. | Residential appliance monitoring based on low frequency smart meter measurements |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |