CN109359665A - A kind of family's electric load recognition methods and device based on support vector machines - Google Patents
A kind of family's electric load recognition methods and device based on support vector machines Download PDFInfo
- Publication number
- CN109359665A CN109359665A CN201810990146.3A CN201810990146A CN109359665A CN 109359665 A CN109359665 A CN 109359665A CN 201810990146 A CN201810990146 A CN 201810990146A CN 109359665 A CN109359665 A CN 109359665A
- Authority
- CN
- China
- Prior art keywords
- load
- data
- training
- parameter
- identification model
- 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.)
- Granted
Links
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
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 embodiment of the present invention provides a kind of family's electric load recognition methods based on support vector machines and device, this method comprises: obtaining the total load data for having turned on household appliance；The total load data are inputted into trained load identification model, obtain identification data.A kind of family's electric load recognition methods and device based on support vector machines provided in an embodiment of the present invention, it is identified using non-intrusion type load, without monitoring the operating status and power consumption of each electric appliance, installment work is not necessarily to enter inside electric appliance, investment is smaller, does not influence normal production work, is suitable for popularizing in an all-round way, algorithm of support vector machine is applied in household electrical appliances identification, an accuracy rate for electric load identification is improved.
Description
Technical field
The present invention relates to non-intrusion type cutting load testing field more particularly to a kind of family's electric load knowledges based on support vector machines
Other method and device.
Background technique
For electric system, electric load monitoring is significant, it is not only advantageous to improve load composition, guides user
Reasonable consumption reduces electric cost, while also helping distributing rationally for State Grid's resource.Electric load decomposition data can
So that power consumer is understood the power consumption of all kinds of electrical equipments of its different periods in more detail, it is helped to formulate reasonable energy conservation
Plan adjusts the use of electrical equipment, reduces power consumption, reduces electricity charge spending, while it is more true also to help Utilities Electric Co.
Understanding electric system load composition, specification load electricity consumption, each type load of reasonable arrangement use the time, improve power grid utilize
Efficiency reduces electric system investment, reduces the operation network loss of system, therefore, building efficient load monitor system is very must
It wants, with investment expense few as far as possible, improves China's electric power monitoring level, realize that system optimized operation, malfunction monitoring are accurate, be
The purpose of system loss reduces, has important practical significance and economic value power industry.
The research for transient-wave identification household electrical appliances gradually increases in recent years, as the raising people of sample rate know household electrical appliances
Other required precision is also being continuously improved, and large batch of load data and complex situations will affect the accurate of load identification model
Degree.
Existing residential power load monitoring technology is to be equipped with a sensor for each electric appliance to obtain its power information,
Belong to intrusive load monitoring.Sensor with digital communication functions is mounted on connecing for each electric appliance and power grid by intrusive mood
Mouthful, to monitor the operating status and power consumption of each electric appliance.This method accurate measurement, but be that investment is larger, installation
Need of work enters inside electric appliance, influences normal production work, is not suitable for popularizing in an all-round way.
Summary of the invention
In order to overcome the above technical defects, the embodiment of the present invention provides a kind of family's electric load identification based on support vector machines
Method and device.
In a first aspect, the embodiment of the present invention provides a kind of family's electric load recognition methods based on support vector machines, comprising: obtain
Take the total load data for having turned on household appliance；The total load data are inputted into trained load identification model, are known
Other data.
Second aspect, the embodiment of the present invention provide a kind of family's electric load identification device based on support vector machines, comprising: obtain
Modulus block, for obtaining the total load data for having turned on household appliance；Identification module is instructed for inputting the total load data
The load identification model perfected obtains identification data.
The third aspect, the embodiment of the present invention provide a kind of electronic equipment, including memory and processor, the processor and
The memory completes mutual communication by bus；The memory, which is stored with, to be referred to by the program that the processor executes
It enables, the processor calls described program to instruct a kind of household electrical appliances based on support vector machines being able to carry out as described in relation to the first aspect
Load recognition methods.
Fourth aspect, the embodiment of the present invention provide a kind of non-transient computer readable storage medium, are stored thereon with calculating
Machine program, which is characterized in that realize that one kind as described in relation to the first aspect is based on supporting when the computer program is executed by processor
Family's electric load recognition methods of vector machine.
A kind of family's electric load recognition methods and device based on support vector machines provided in an embodiment of the present invention, is invaded using non-
Enter the identification of formula load, without monitoring the operating status and power consumption of each electric appliance, installment work is not necessarily to enter inside electric appliance, throw
Enter smaller, do not influence normal production work, be suitable for popularizing in an all-round way, algorithm of support vector machine is applied in household electrical appliances identification, is mentioned
The high accuracy rate of family's electric load identification.
Detailed description of the invention
Fig. 1 is a kind of flow diagram of family's electric load recognition methods based on support vector machines of the embodiment of the present invention；
Fig. 2 is algorithm flow schematic diagram of the embodiment of the present invention using PSO to SVM parameter optimization；
Fig. 3 is that a kind of another process of family's electric load recognition methods based on support vector machines of the embodiment of the present invention is illustrated
Figure；
Fig. 4 is to be identified electric-opening period waveform diagram of the embodiment of the present invention；
Fig. 5 is that the SVM parameter selection of 1 training pattern of household electrical appliances of the embodiment of the present invention and accuracy rate are distributed 3D view；
Fig. 6 is particle swarm algorithm of embodiment of the present invention parameter optimization fitness curve graph；
Fig. 7 is 1 event detection error of household electrical appliances of the embodiment of the present invention when being in 500 data points SVM parameter selection and accuracy rate
3D view respectively；
Fig. 8 is 1 event detection error of household electrical appliances of the embodiment of the present invention when being in 1000 data points SVM parameter selection and accuracy rate
It is distributed 3D view；
Fig. 9 is a kind of structural schematic diagram of family's electric load identification device based on support vector machines of the embodiment of the present invention；
Figure 10 is the entity structure schematic diagram of a kind of electronic equipment of the embodiment of the present invention.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached in the embodiment of the present invention
Figure, is clearly and completely described the technical solution in the present invention, it is clear that described embodiment is one of the invention
Divide embodiment, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making
Every other embodiment obtained, shall fall within the protection scope of the present invention under the premise of creative work.
Fig. 1 is a kind of flow diagram of family's electric load recognition methods based on support vector machines of the embodiment of the present invention, such as
Shown in Fig. 1, comprising:
Step 11, the total load data for having turned on household appliance are obtained；
Step 12, the total load data are inputted into trained load identification model, obtains identification data.
In the embodiment of the present invention, the household electrical appliances recognition methods of use belongs to non-intrusion type load identification field, and non-intrusion type is negative
Lotus monitoring (non-intrusive load monitoring, NILM) device can measure voltage, the electric current etc. for obtaining total load
The signal of power information is carried, these information contain the information of different characteristics load ingredient.By extracting these electrical quantity
Characteristic information, NILM system can be achieved with load decomposition.
That is, can detect user by electric power signals such as voltage, electric current or the power of detection total load and use
What electric appliance.
By obtaining the data of non-intrusion type cutting load testing, load inspection from the main circuit outside building or user's residence
The type of the data of survey can be chosen for power signal.It is had turned on using the instrument inlet detection on the outside of building and enlivens household electrical appliances
Total power signal after device aggregation.
The embodiment of the present invention is based on load identification model and is identified, total power signal is inputted trained load and is identified
Model obtains identification data, i.e., the power signal of each special installation, to realize the decomposition and identification to each household appliance.
A kind of family's electric load recognition methods based on support vector machines provided in an embodiment of the present invention, it is negative using non-intrusion type
Lotus identification, without monitoring the operating status and power consumption of each electric appliance, installment work is not necessarily to enter inside electric appliance, investment compared with
It is small, normal production work is not influenced, is suitable for popularizing in an all-round way, and algorithm of support vector machine is applied in household electrical appliances identification, is improved
The accuracy rate of family's electric load identification.
On the basis of the above embodiments, the trained load identification model is obtained by following steps:
Training sample set is constructed using load identification types；
The data that the training sample is concentrated are input to training in load identification model, based on the K-CV in cross validation
The load identification model of method acquisition pre-training；
Classification accuracy is calculated using the load identification model of the pre-training, is based on the classification accuracy and population
Optimization algorithm adjusts the load identification model of the pre-training, obtains trained load identification model.
Training sample set is constructed according to load identification types first, load identification types can reflect that an electrical equipment is being transported
Information of unique reflection electricity condition, such as voltage, active waveform, starting current etc. embodied in row.These features
It is determined by the operating condition of electrical equipment, load identification types can be divided into stable state, transient state, 3 class of operational mode accordingly, wherein
Stable state and transient state depend on the component feature inside equipment；Operational mode is determined by the operation control strategy of equipment.In equipment
In operational process, these load identification types can repeat, and be based on this, we can identify each electric appliance.This
In inventive embodiments, using transient state type.
After obtaining training sample set, according to algorithm of support vector machine, the data that training sample is concentrated is input to load and are known
Training in other model.Penalty parameter c and kernel functional parameter g are the major parameters for needing to adjust in SVM, and c represents load identification mould
To load waveform and the asynchronous tolerance of learning process in type, g is the parameter of the kernel function of mapping.Using training sample set as
Raw data set simultaneously utilizes the K-fold Cross Validation (K- in cross validation (Cross Validation, CV)
CV the negative of the classification accuracy of training sample set under this group of penalty parameter c and kernel functional parameter g and corresponding pre-training) is obtained
Lotus identification model.
The parameter optimization range of c and g is set, assists carrying out parameter optimization using particle swarm optimization algorithm.The fitness of PSO
For the classification accuracy acquired after determining parameter c, g using K-CV method, to find optimized parameter c, g it is obtained at CV
The load identification model of pre-training is adjusted, trained load is obtained according to optimal parameter c and g to highest classification accuracy
Identification model.
A kind of family's electric load recognition methods based on support vector machines provided in an embodiment of the present invention, it is negative using non-intrusion type
Lotus identification, without monitoring the operating status and power consumption of each electric appliance, installment work is not necessarily to enter inside electric appliance, investment compared with
It is small, normal production work is not influenced, is suitable for popularizing in an all-round way, and algorithm of support vector machine is applied in household electrical appliances identification, is improved
The accuracy rate of family's electric load identification.Meanwhile using K-CV method training load identification model, assisted using particle swarm optimization algorithm
Parameter optimization is carried out, to obtain the higher load identification model of classification accuracy, to improve to adjust the major parameter in SVM
The precision of model.
On the basis of the above embodiments, described to construct training sample set using load identification types, it specifically includes:
Pretreated load data is generated into corresponding simulated data sets according to load identification types；
The simulated data sets are divided into training set and test set by preset ratio；
The training set and the test set are normalized, the training sample set is obtained.
It is described that pretreated load data is generated into corresponding simulated data sets according to load identification types, it is specific to wrap
It includes:
Load data when a certain number of household appliances are individually opened within the unit time is obtained, is sieved according to preset rules
The load data is selected, obtains generating analogue data；
The simulated data sets are synthesized using the generation analogue data.
The load identification types chosen in the embodiment of the present invention are the power signals of transient state, and power is equal to voltage and current
Product.With indicating, mathematic(al) representation can be approximated to be total power signal:
P (t)=p1(t)+p2(t)+...+pn(t)
Wherein n is that the sum for enlivening household appliance, p are had turned in measurement periodi(t) i-th of household appliance is represented in t
Power consumption of the moment to the individual equipment of entire polymerization measurement result contribution.The purpose of NILM system is by the total of instrument measurement
Power signal P (t) is decomposed into the power signal of each special installation, to realize point to each household appliance in building construction
Solution and identification.
Voltage and current data when a certain number of household appliances are individually opened within the unit time are obtained first, then
It is screened, apparent data are fluctuated in selection.The preset rules of screening are according to event detection.
Event detection is exactly according to certain rules according to the variation of signal to determine whether there is new events in simple terms
It generates.Simplest method is exactly the variation by calculating adjacent moment or load in some time data, and by itself and setting
Threshold value is compared, and when variation is more than threshold value, event occurs for judgement.This method is simple to operation, and problem is threshold value
Setting has craftsmenship, the too big or too small event detection outcome that can all lead to mistake.For this reason, it may be necessary to a large amount of sample pair
Parameter is trained.One threshold value is rule of thumb set in the embodiment of the present invention, and when fluctuation is more than the threshold value, representative has event
Occur, it is on the contrary then do not have that event occurs.After screening, obtain generating analogue data.
Then according to analogue data synthesis simulated data sets are generated, synthetic method is as follows:
The load data of each household electrical appliance may be expressed as:
q1(n)={ q1(1),q1(2),...,q1(N1)}
q2(n)={ q2(1),q2(2),...,q2(N2)}
......
qm(n)={ qm(1),qm(2),...,qm(Nm)}
In formula, m is data set sum, NiFor the number of sampling points of equipment, q (i) is the apparent energy sampled value of household electrical appliances.It will
Household electrical appliances load data synthesis, it is assumed that the corresponding household electrical appliances of the appliance data of synthesis are m and n, and synthesis mode is as follows:
Sr(n)=s (1), s (2) ..., s (br),s(br+1),...,s(br+ar),s(br+ar+1),
...s(Nm-cr+br+1)}
=0,0 ..., qm(cr),qm(cr+1),...,qm(cr+ar)+qn(1),
qm(cr+ar+1)+qn(2),...,qm(Nm-cr+1)+qn(Nn-ar)}
In formula, SrIt (n) is the simulated data sets after synthesis, s (i) is the apparent energy sampled value of simulated data sets, brFor family
Position at the time of electric m is opened, arFor the distance away from household electrical appliances m start-up time position, crAfter extracting activation for household electrical appliances n
Reading position.C most of the timerIt is 1, when household electrical appliances m is opened prior to household electrical appliances n and opening feature is imperfect, then will appear cr> 1
Situation.
After obtaining simulated data sets, it is classified as simulation collection and test set, and simulation collection and test set are uniformly returned
One change processing, obtains training sample set, the normalized mapping of use is as follows:
Xmin=min (x)
Xmax=max (x)
Wherein, x represents training set and test set data, and y is the data acquisition system after normalized.The mesh of normalized
Be by data normalization.The standardisation process of data, which refers to, zooms in and out data according to ratio, falls into it one smaller
The characteristics of section in, by removal data unit limit, so that it is converted into nondimensional pure values, this data handling procedure
Be conducive to the data not between commensurate or magnitude to be compared and weight.And in general, if the data of input have mean value,
The efficiency and precision of machine learning algorithm are higher, therefore common data normalization means are normalized.Normalized knot
Fruit is the number be converted to former data in [0,1] section, i.e. y ∈ [0,1], i=1,2 ..., n.
A kind of family's electric load recognition methods based on support vector machines provided in an embodiment of the present invention, it is negative using non-intrusion type
Lotus identification, without monitoring the operating status and power consumption of each electric appliance, installment work is not necessarily to enter inside electric appliance, investment compared with
It is small, normal production work is not influenced, is suitable for popularizing in an all-round way, and algorithm of support vector machine is applied in household electrical appliances identification, is improved
The accuracy rate of family's electric load identification.
On the basis of the above embodiments, the K-CV method based in cross validation obtains the load identification of pre-training
Model specifically includes:
The training sample set is divided into K group training subset；
Each training subset is made into one-time authentication collection, remaining K-1 group training subset is trained tests respectively as training set
Card obtains the load identification model of the K pre-training.
Penalty parameter c and kernel functional parameter g are the major parameters for needing to adjust in SVM, to load waveform in c representative model
With the asynchronous tolerance of learning process, g is the parameter of the kernel function of mapping.Using training sample set as raw data set and benefit
This group of c and g is obtained with the K-fold Cross Validation (K-CV) in cross validation (Cross Validation, CV)
The classification accuracy of lower training set.The realization process of K-CV method is as follows, and initial data is divided into K group, by each number of subsets
According to one-time authentication collection is respectively made, remaining K-1 group subset data will obtain K after being trained verifying respectively as training set
A training pattern.Take its average as the performance of classifier under K-CV the classification accuracy that this K model finally verifies collection
Index.Wherein, the generally ascending beginning value of K, minimum 3, can take 2 when data are less.
On the basis of the above embodiments, described described pre- based on the classification accuracy and particle swarm optimization algorithm adjustment
Trained load identification model obtains trained load identification model, specifically includes:
Parameter optimization range is set, and the parameter optimization range includes punishment parameter range and kernel functional parameter range；
Optimizing iteration is carried out to punishment parameter and kernel functional parameter, is calculated each time based on the K-CV method in cross validation
Classification accuracy after iteration；
After reaching the number of iterations, highest one group corresponding first punishment parameter of output category accuracy rate and the first kernel function
Parameter obtains first punishment parameter and the corresponding trained load identification model of first kernel functional parameter.
After highest one group corresponding first punishment parameter of the output category accuracy rate and the first kernel functional parameter,
Further include:
If the classification accuracy highest is corresponding with multiple groups, export the smallest one group of punishment parameter in the multiple groups it is corresponding
First punishment parameter and the first kernel functional parameter；
If the punishment parameter is the smallest to be corresponding with multiple groups, one group of corresponding first punishment parameter searched at first is exported
With the first kernel functional parameter.
It assists carrying out parameter optimization using particle swarm optimization algorithm (Particle Swarm Optimization, PSO),
The fitness of PSO is the classification accuracy acquired after determining parameter c, g using K-CV method, so that finding optimized parameter c, g makes
It obtains it and obtains highest classification accuracy at CV.
Fig. 2 is algorithm flow schematic diagram of the embodiment of the present invention using PSO to SVM parameter optimization, as shown in Figure 2, comprising:
Step 201, fitness function is determined according to optimization problem；
Step 202, it sets Search Range and relevant parameter, relevant parameter includes penalty parameter c and kernel functional parameter g；
Step 203, suitable population position and initial velocity are generated；
Step 204, initial fitness is calculated；
Step 205, new individual extreme value, global extremum are recorded, determines iteration direction；
Step 206, speed updates, population recruitment；
Step 207, adaptive particle variations；
Step 208, optimal solution, recording individual extreme value, global extremum are determined；
Step 209, judge whether to reach termination the number of iterations, if not up to, executing step 206 if having reached and executing step
Rapid 210；
Step 210, multiple groups optimal solution is judged whether there is, if so, step 211 is executed, if it is not, executing step 212；
Step 211, choose wherein the smallest group of penalty parameter c be optimal solution；
Step 212, optimal solution (optimized parameter c and g) is exported.
The fitness for defining PSO is the classification accuracy acquired after determining parameter c, g using K-CV method, is rule of thumb set
After setting parameter optimization range and relevant parameter, population position (c, g are generated according to the population quantity of setting and Search Range at random
Value), and define initial velocity.
In next step its corresponding fitness (classification accuracy) is calculated according to c, g value in each population and record population and
Optimal value between individual, and it is based on this Population Regeneration and speed, the position of population is optimal corresponding with global optimum to individual
C, g variation.Final result is influenced in order to avoid falling into local optimum during population " migration ", in population position by speed
On the basis of control, the random variation for giving some individuals small probability may.
When reaching final the number of iterations, searching process is terminated, output is optimal fitness (highest classification accuracy)
Population position (optimized parameter c, g).If there are the c of multiple groups and g to correspond to highest classification accuracy in result, it is contemplated that c
As punishment parameter, if c value is excessively high, easily lead to the generation of over-fitting state, i.e., it is very high training sample set accuracy rate easily occur,
But the too low problem of the obtained model generalization ability of training, thus select wherein parameter c the smallest that group of c, g as optimal ginseng
Number, and if the corresponding parameter of the smallest c also have multiple groups, choose that group searched at first as optimized parameter.
A kind of family's electric load recognition methods based on support vector machines provided in an embodiment of the present invention, it is negative using non-intrusion type
Lotus identification, without monitoring the operating status and power consumption of each electric appliance, installment work is not necessarily to enter inside electric appliance, investment compared with
It is small, normal production work is not influenced, is suitable for popularizing in an all-round way, and algorithm of support vector machine is applied in household electrical appliances identification, is improved
The accuracy rate of family's electric load identification.Meanwhile by particle swarm optimization algorithm auxiliary parameter optimizing, know to adjust the load of pre-training
Other model enables to load identification model to have higher classification accuracy, improves the performance of model.
Fig. 3 is that a kind of another process of family's electric load recognition methods based on support vector machines of the embodiment of the present invention is illustrated
Figure, as shown in figure 3, including three phases, wherein the stage, one data preparation included:
Step 301, load data pre-processes；
Step 302, simulated data sets are generated according to load identification types；
Step 303, simulated data sets normalize；
Step 304, activation is extracted；
Step 305, label is generated according to load data content；
Step 306, simulated data sets are divided into training set and test set；
Stage two data training includes:
Step 307, parameter setting, including punishment parameter and kernel functional parameter；
Step 308, Selection of kernel function；
Step 309, PSO parameter and parameter optimization range are set；
Step 310, SVM parameter optimization iteration；
Step 311, according to cross validation to the training of SVM load identification model；
Step 312, judge whether searching process terminates, if being not finished, execute step 310, if terminating, execute step 313；
Step 313, optimized parameter c, g and trained load identification model are obtained according to highest accuracy rate；
The identification of stage three, which is verified, includes:
Step 314, event detection；
Step 315, total load data are input to progress load identification in trained load identification model；
Step 316, recognition result and test set label result are compared；
Step 317, test set accuracy rate and recognition result are exported.
The method provided below by an example the present invention is described in detail embodiment.
In example, load data sample rate is 30000Hz, it is assumed that the unlatching temporal characteristics of all household electrical appliances in former data are dynamic
A length of 1.5s when making, when sequence of household electrical appliances time window a length of 2s, two household appliance opening times are spaced in 0.5s, each
The data length of the corresponding time series of family's electric-opening event is 60000 units.Therefore the generation mould after extracting activation
The specific rules of the simulated data sets of quasi- Data Synthesis are as shown in table 1:
The specific rules of the synthesis simulated data sets of table 1
Wherein, ai、bi, for the random number of 1-15000, and ai、biMeet ai+bi≤ 15001 have the data of activation equipment
Type is the appliance data to be detected chosen, not with activation equipment data class as interference in addition to identifying household electrical appliances its
The data of his household electrical appliances, the appliance data of selection are to select at random from the data set after screening.
In the analogue data that the analogue data concentration addition of synthesis contains only single activation equipment is for bootstrap algorithm
Habit process, the data containing activation equipment and the data bulk for being free of activation equipment in the Wave data of the same type generated in table
Ratio is identical, and the purpose is to reduce in training process the case where over-fitting and poor fitting occur.No matter which kind of situation, require to make
The opening feature of deactivated device containing activation equipment and as " incorrect " is concentrated in the analogue data of generation and keeps complete
In favor of the training learning process of model, meet a in tablei+bi≤ 15001 and open position be 1-15000 random number purpose
It is to guarantee that the time between the deactivated device of activation equipment and interference is no more than 0.5s.
Emulation selects the corresponding household electrical appliances of 4 data as activation equipment at random from original data set, and experiment is respectively for each
The analogue data of household electrical appliances is learnt and is deduced, and Fig. 4 is to be identified electric-opening period waveform diagram of the embodiment of the present invention, in Fig. 4
(a) corresponding household electrical appliances 1, (b) corresponding household electrical appliances 2, (c) corresponding household electrical appliances 3, (d) corresponding household electrical appliances 4, the unlatching period of each household electrical appliances regards in function
The waveform of rate as shown in figure 4, the simulation collection containing to be identified electrical waveform is marked after generate simulated data sets, wherein uniformly
80% data are randomly selected as training set, the data of residue 20% are used to verify the accurate of identification model as test set
Rate.Fig. 5 is the SVM parameter selection of 1 training pattern of household electrical appliances of the embodiment of the present invention and accuracy rate is distributed 3D view, and Fig. 6 is the present invention
Embodiment particle swarm algorithm parameter optimization fitness curve graph, as shown in Figure 5 and Figure 6.Recognition result is as shown in table 2.
2 load recognition result of table
Above-mentioned experimentation is built upon that event detection is accurate, i.e., model training and load ideally identifies feelings
Condition, from table 2 it can be seen that the better performances of load recognition methods in the ideal case, but in view of in practical applications due to
The interference meeting that its feature is superimposed and is likely to occur when a large amount of load operations is so that the activation and detection of load unlatching event become
It is more difficult.In order to cope with load open event detection there are errors the case where, using data progress model training process
In, need to be added certain error in event detection, while to retain effective household electrical appliances opening feature, event detection being obtained
Household electrical appliances open position be set as carrying out model instruction away from the random site between 500 or 1000 data points before practical open position
The available load identification model for possessing event detection error " tolerance " after white silk process, Fig. 7 are household electrical appliances of the embodiment of the present invention 1
SVM parameter selection and accuracy rate distinguish 3D view when event detection error is in 500 data points, and Fig. 8 is man of the embodiment of the present invention
SVM parameter selection and accuracy rate are distributed 3D view when electric 1 event detection error is in 1000 data points, instruct as shown in Figure 7, Figure 8
The SVM parameter selection and accuracy of 1 model of household electrical appliances got are distributed, and recognition result is as shown in table 3.
Table 3 has the load recognition result of event detection error
From the recognition result of table 3 can be seen that comparison event detection it is accurate when the case where, considering event detection error
In the case of the load recognition accuracy of training pattern declined, and for its identification knot of family's electric load waveform of different characteristic
Fruit is also different.Recognition accuracy by comparing event detection error size can be seen that error is bigger, and accuracy rate is got over
It is low.It can be seen that by comparison diagram 7, Fig. 8 and Fig. 5, Fig. 6 since training environment changes, the variation model of SVM parameter distribution
It is wider to enclose span, changing rule tends towards stability, and training set entirety accuracy rate decreases, and compares ideally in training process
Learning efficiency is poor.But by the increase of training data, can effective training for promotion collection accuracy rate, to a certain extent to survey
The accuracy rate of examination collection also has a certain upgrade effect.
Fig. 9 is a kind of structural schematic diagram of family's electric load identification device based on support vector machines of the embodiment of the present invention, such as
Shown in Fig. 9, including obtain module 91 and identification module 92, wherein obtain module 91 for obtaining and have turned on the total of household appliance
Load data；Identification module 92 is used to the total load data inputting trained load identification model, obtains identification data.
In the embodiment of the present invention, the household electrical appliances recognition methods of use belongs to non-intrusion type load identification field, and non-intrusion type is negative
Lotus monitoring (non-intrusive load monitoring, NILM) device can measure voltage, the electric current etc. for obtaining total load
The signal of power information is carried, these information contain the information of different characteristics load ingredient.By extracting these electrical quantity
Characteristic information, NILM system can be achieved with load decomposition.
That is, can detect user by electric power signals such as voltage, electric current or the power of detection total load and use
What electric appliance.
By obtaining the data of non-intrusion type cutting load testing, load inspection from the main circuit outside building or user's residence
The type of the data of survey can be chosen for power signal.Module 91 is obtained to have opened using the instrument inlet detection on the outside of building
Open the total power signal after enlivening household appliance polymerization.
The embodiment of the present invention is based on load identification model and is identified, identification module 92 trains total power signal input
Load identification model, obtain identification data, i.e., the power signal of each special installation, thus realize each household appliance is divided
Solution and identification.Device provided in an embodiment of the present invention is the specific process and in detail for executing above-mentioned each method embodiment
Introduction refers to above-mentioned each method embodiment, and details are not described herein again.
A kind of family's electric load identification device based on support vector machines provided in an embodiment of the present invention, it is negative using non-intrusion type
Lotus identification, without monitoring the operating status and power consumption of each electric appliance, installment work is not necessarily to enter inside electric appliance, investment compared with
It is small, normal production work is not influenced, is suitable for popularizing in an all-round way, and algorithm of support vector machine is applied in household electrical appliances identification, is improved
The accuracy rate of family's electric load identification.
The embodiment of the present invention provides a kind of family's electric load identifying system based on support vector machines comprising following several moulds
Block:
Data preparation module
In data preparation module, corresponding simulation is generated according to load identification types first with pretreated load data
Data set then occurs position according to event and extracts activated positon and according to load identification content to simulated data sets label, presses
Simulated data sets are divided into training set and test set by ratio.
Data training module
It determines algorithm parameter and training method, and parameter optimization range is set, be based on cross validation method, by the negative of generation
Lotus identification model carries out optimizing by training set, and searching can make optimal parameter c, g of training set accuracy rate and its corresponding negative
Lotus identification model.
Test authentication module
It determines that position occurs for the event of test set data using event detection, and is carried out using the load identification model generated
Identification.
A kind of family's electric load identifying system based on support vector machines provided in an embodiment of the present invention utilizes analysis non-intruding
Implementation method of the formula cutting load testing in residential loads identification proposes one kind and is based on for the defects of load recognition methods
The household electrical appliances recognition methods of support vector machines, and meet by house the particularity of identification problem, it devises a kind of according to limited number
The data training process of simulated data sets is generated according to collection, and utilizes particle swarm algorithm to supporting vector on the basis of grid data service
Machine parameter optimization determines optimized parameter according to the resulting training set accuracy of parameter.The invention is at low cost, easy for installation, identification
Effect is good.
Figure 10 illustrates the entity structure schematic diagram of a kind of electronic equipment, and as shown in Figure 10, which may include:
Processor (processor) 101, communication interface (Communications Interface) 102, memory (memory) 103
With bus 104, wherein processor 101, communication interface 102, memory 103 complete mutual communication by bus 104.Always
Line 104 can be used for the transmission of the information between electronic equipment and sensor.Processor 101 can call patrolling in memory 103
Instruction is collected, to execute following method: obtaining the total load data for having turned on household appliance；The total load data are inputted and are trained
Good load identification model obtains identification data.
In addition, the logical order in above-mentioned memory 103 can be realized by way of SFU software functional unit and conduct
Independent product when selling or using, can store in a computer readable storage medium.Based on this understanding, originally
Substantially the part of the part that contributes to existing technology or the technical solution can be in other words for the technical solution of invention
The form of software product embodies, which is stored in a storage medium, including some instructions to
So that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation of the present invention
The all or part of the steps of example the method.And storage medium above-mentioned include: USB flash disk, mobile hard disk, read-only memory (ROM,
Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. it is various
It can store the medium of program code.
The embodiment of the present invention provides a kind of non-transient computer readable storage medium, the non-transient computer readable storage medium
Matter stores computer instruction, which makes computer execute pseudo-base station localization method provided by above-described embodiment, example
It such as include: to obtain the total load data for having turned on household appliance；The total load data are inputted into trained load and identify mould
Type obtains identification data.
The above description is only a preferred embodiment of the present invention, is not intended to restrict the invention.The technical field of the invention
Technical staff can make various modifications or additions to the described embodiments, but without departing from of the invention
Spirit surmounts the range that the appended claims define.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, and those skilled in the art is it is understood that it still can be right
Technical solution documented by foregoing embodiments is modified or equivalent replacement of some of the technical features；And this
It modifies or replaces, the spirit and model of technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution
It encloses.
Claims (10)
1. a kind of family's electric load recognition methods based on support vector machines characterized by comprising
Obtain the total load data for having turned on household appliance；
The total load data are inputted into trained load identification model, obtain identification data.
2. the method according to claim 1, wherein the trained load identification model passes through following steps
It obtains:
Training sample set is constructed using load identification types；
The data that the training sample is concentrated are input to training in load identification model, based on the K-CV method in cross validation
Obtain the load identification model of pre-training；
Classification accuracy is calculated using the load identification model of the pre-training, is based on the classification accuracy and particle group optimizing
Algorithm adjusts the load identification model of the pre-training, obtains trained load identification model.
3. according to the method described in claim 2, it is characterized in that, it is described using load identification types construct training sample set,
It specifically includes:
Pretreated load data is generated into corresponding simulated data sets according to load identification types；
The simulated data sets are divided into training set and test set by preset ratio；
The training set and the test set are normalized, the training sample set is obtained.
4. according to the method described in claim 3, it is characterized in that, it is described according to load identification types by pretreated load
Data generate corresponding simulated data sets, specifically include:
Load data when a certain number of household appliances are individually opened within the unit time is obtained, screens institute according to preset rules
Load data is stated, obtains generating analogue data；
The simulated data sets are synthesized using the generation analogue data.
5. according to the method described in claim 2, it is characterized in that, the K-CV method based in cross validation obtains pre- instruction
Experienced load identification model, specifically includes:
The training sample set is divided into K group training subset；
Each training subset is made into one-time authentication collection, remaining K-1 group training subset is trained verifying as training set respectively,
Obtain the load identification model of the K pre-training.
6. according to the method described in claim 2, it is characterized in that, described be based on the classification accuracy and Particle Swarm Optimization
Method adjusts the load identification model of the pre-training, obtains trained load identification model, specifically includes:
Parameter optimization range is set, and the parameter optimization range includes punishment parameter range and kernel functional parameter range；
Optimizing iteration is carried out to punishment parameter and kernel functional parameter, iteration each time is calculated based on the K-CV method in cross validation
Classification accuracy afterwards；
After reaching the number of iterations, highest one group corresponding first punishment parameter of output category accuracy rate and the first kernel function ginseng
Number, obtains first punishment parameter and the corresponding trained load identification model of first kernel functional parameter.
7. according to the method described in claim 6, it is characterized in that, corresponding at highest one group of the output category accuracy rate
After first punishment parameter and the first kernel functional parameter, further includes:
If the classification accuracy highest is corresponding with multiple groups, it is one group corresponding first the smallest to export punishment parameter in the multiple groups
Punishment parameter and the first kernel functional parameter；
If the punishment parameter is the smallest to be corresponding with multiple groups, one group of corresponding first punishment parameter searched at first and the are exported
One kernel functional parameter.
8. a kind of family's electric load identification device based on support vector machines characterized by comprising
Module is obtained, for obtaining the total load data for having turned on household appliance；
Identification module obtains identification data for the total load data to be inputted trained load identification model.
9. a kind of electronic equipment, which is characterized in that including memory and processor, the processor and the memory pass through always
Line completes mutual communication；The memory is stored with the program instruction that can be executed by the processor, the processor tune
A kind of family's electric load based on support vector machines as described in claim 1 to 7 is any is able to carry out with described program instruction to know
Other method.
10. a kind of non-transient computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer
A kind of family's electric load knowledge based on support vector machines as described in any one of claim 1 to 7 is realized when program is executed by processor
Other method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810990146.3A CN109359665B (en) | 2018-08-28 | 2018-08-28 | Household appliance load identification method and device based on support vector machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810990146.3A CN109359665B (en) | 2018-08-28 | 2018-08-28 | Household appliance load identification method and device based on support vector machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109359665A true CN109359665A (en) | 2019-02-19 |
| CN109359665B CN109359665B (en) | 2020-11-06 |
Family
ID=65350049
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810990146.3A Active CN109359665B (en) | 2018-08-28 | 2018-08-28 | Household appliance load identification method and device based on support vector machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109359665B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110135452A (en) * | 2019-03-27 | 2019-08-16 | 广东浩迪创新科技有限公司 | Violation electrical appliance recognition and system based on intelligent electric meter |
| CN110414839A (en) * | 2019-07-29 | 2019-11-05 | 四川长虹电器股份有限公司 | Load recognition methods and system based on quantum genetic algorithm and SVM model |
| CN110504679A (en) * | 2019-07-25 | 2019-11-26 | 深圳供电局有限公司 | KM matching algorithm-based non-invasive load identification method |
| CN111242391A (en) * | 2020-03-06 | 2020-06-05 | 云南电网有限责任公司电力科学研究院 | Machine learning model training method and system for power load identification |
| CN112379159A (en) * | 2020-11-09 | 2021-02-19 | 北华航天工业学院 | Non-invasive household load decomposition method based on electric appliance operation mode |
| CN113076985A (en) * | 2021-03-29 | 2021-07-06 | 上海电力大学 | Power consumer load identification method and device and computer readable medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105956525A (en) * | 2016-04-22 | 2016-09-21 | 中国人民解放军总装备部军械技术研究所 | Static gesture recognizing method and system for AR auxiliary maintenance |
| CN106952183A (en) * | 2017-04-24 | 2017-07-14 | 武汉大学 | A kind of short-term load forecasting method based on particle group optimizing least square method supporting vector machine |
| CN107832705A (en) * | 2017-11-08 | 2018-03-23 | 天津大学 | A kind of non-intrusion type load recognition methods based on canonical correlation analysis |
-
2018
- 2018-08-28 CN CN201810990146.3A patent/CN109359665B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105956525A (en) * | 2016-04-22 | 2016-09-21 | 中国人民解放军总装备部军械技术研究所 | Static gesture recognizing method and system for AR auxiliary maintenance |
| CN106952183A (en) * | 2017-04-24 | 2017-07-14 | 武汉大学 | A kind of short-term load forecasting method based on particle group optimizing least square method supporting vector machine |
| CN107832705A (en) * | 2017-11-08 | 2018-03-23 | 天津大学 | A kind of non-intrusion type load recognition methods based on canonical correlation analysis |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110135452A (en) * | 2019-03-27 | 2019-08-16 | 广东浩迪创新科技有限公司 | Violation electrical appliance recognition and system based on intelligent electric meter |
| CN110135452B (en) * | 2019-03-27 | 2023-08-08 | 广东浩迪创新科技有限公司 | Illegal electric appliance identification method and system based on intelligent electric meter |
| CN110504679A (en) * | 2019-07-25 | 2019-11-26 | 深圳供电局有限公司 | KM matching algorithm-based non-invasive load identification method |
| CN110414839A (en) * | 2019-07-29 | 2019-11-05 | 四川长虹电器股份有限公司 | Load recognition methods and system based on quantum genetic algorithm and SVM model |
| CN111242391A (en) * | 2020-03-06 | 2020-06-05 | 云南电网有限责任公司电力科学研究院 | Machine learning model training method and system for power load identification |
| CN112379159A (en) * | 2020-11-09 | 2021-02-19 | 北华航天工业学院 | Non-invasive household load decomposition method based on electric appliance operation mode |
| CN113076985A (en) * | 2021-03-29 | 2021-07-06 | 上海电力大学 | Power consumer load identification method and device and computer readable medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109359665B (en) | 2020-11-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109359665A (en) | A kind of family's electric load recognition methods and device based on support vector machines | |
| Ciancetta et al. | A new convolutional neural network-based system for NILM applications | |
| CN113792825B (en) | Fault classification model training method and device for electricity information acquisition equipment | |
| CN109145949A (en) | Non-intrusive electrical load monitoring and decomposition method and system based on integrated study | |
| Ma et al. | Topology identification of distribution networks using a split-EM based data-driven approach | |
| CN108051660A (en) | A kind of transformer fault combined diagnosis method for establishing model and diagnostic method | |
| CN109165604A (en) | The recognition methods of non-intrusion type load and its test macro based on coorinated training | |
| CN110596506A (en) | Converter fault diagnosis method based on time convolution network | |
| CN111639586B (en) | Non-invasive load identification model construction method, load identification method and system | |
| CN116842459B (en) | Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning | |
| CN114066236B (en) | Wind-solar active power output scene generation method and device, electronic equipment and storage medium | |
| CN109613440A (en) | Stage division, device, equipment and the storage medium of battery | |
| CN116681186B (en) | Power quality analysis method and device based on intelligent terminal | |
| CN111242161A (en) | Non-invasive non-resident user load identification method based on intelligent learning | |
| CN109376944A (en) | The construction method and device of intelligent electric meter prediction model | |
| CN114970633B (en) | LSTM-based non-invasive electrical appliance identification method, system and equipment | |
| CN105954695A (en) | Synchronization-based homogeneous-sensor mutation parameter recognizing method and device | |
| CN111582315B (en) | Sample data processing method and device and electronic equipment | |
| Ma et al. | An intelligent classification framework for complex PQDs using optimized KS-transform and multiple fusion CNN | |
| CN115712871A (en) | Power electronic system fault diagnosis method combining resampling and integrated learning | |
| CN115758151A (en) | Combined diagnosis model establishing method and photovoltaic module fault diagnosis method | |
| Li et al. | Stacking ensemble learning-based load identification considering feature fusion by cyber-physical approach | |
| CN117272184A (en) | ISSA-RF algorithm-based photovoltaic power generation system fault classification method | |
| Jazizadeh et al. | Unsupervised clustering of residential electricity consumption measurements for facilitated user-centric non-intrusive load monitoring | |
| CN116502149A (en) | Low-voltage power distribution network user-transformation relation identification method and system based on current characteristic conduction |
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 |