[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN111476298B - Power load state identification method in home and office environment - Google Patents

Power load state identification method in home and office environment Download PDF

Info

Publication number
CN111476298B
CN111476298B CN202010265417.6A CN202010265417A CN111476298B CN 111476298 B CN111476298 B CN 111476298B CN 202010265417 A CN202010265417 A CN 202010265417A CN 111476298 B CN111476298 B CN 111476298B
Authority
CN
China
Prior art keywords
load
class
tpw
samples
sample
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
Application number
CN202010265417.6A
Other languages
Chinese (zh)
Other versions
CN111476298A (en
Inventor
郑松
孔亚广
陈张平
徐平
张帆
赵晓东
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Guobiao Ultrasonic Equipment Co ltd
Hangzhou Dianzi University
Original Assignee
Hangzhou Guobiao Ultrasonic Equipment Co ltd
Hangzhou Dianzi University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hangzhou Guobiao Ultrasonic Equipment Co ltd, Hangzhou Dianzi University filed Critical Hangzhou Guobiao Ultrasonic Equipment Co ltd
Priority to CN202010265417.6A priority Critical patent/CN111476298B/en
Publication of CN111476298A publication Critical patent/CN111476298A/en
Application granted granted Critical
Publication of CN111476298B publication Critical patent/CN111476298B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Educational Administration (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Tourism & Hospitality (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Operations Research (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Game Theory and Decision Science (AREA)
  • Molecular Biology (AREA)
  • Quality & Reliability (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Supply And Distribution Of Alternating Current (AREA)
  • Testing Electric Properties And Detecting Electric Faults (AREA)

Abstract

The invention discloses a method for identifying the state of an electrical load in a home and office environment. In order to more accurately and quickly identify the working state of the electrical load, the transient power waveform TPW during switching of the electrical load (such as a refrigerator, an air conditioner, a water dispenser and the like) in the household and office environment and an accurate label during load switching are firstly acquired so as to be conveniently used for load identification and algorithm performance evaluation. Then, aiming at the difficulties existing in the identification of low power and similar loads, the invention provides the steps of carrying out system clustering on similar loads TPW, selecting a typical TPW template and improving the identification rate of the loads. Finally, in order to solve the difficulty of increasing the load types, the invention provides a method for optimizing parameters by using a PSO algorithm and determining the optimal combination of load typical TPW cluster number so as to ensure that the comprehensive performance of the load state identification algorithm is optimal. The method provided by the invention makes important contributions to load state identification and intelligent power utilization in the home and office environment.

Description

Power load state identification method in home and office environment
Technical Field
The invention belongs to the technical field of non-invasive load identification, and particularly relates to a method for identifying a power load state in a home and office environment.
Background
With the development of social economy, electric power becomes one of the most widely used energy sources in the current society. Intelligent power usage is becoming increasingly important in order to provide more reliable, green power to users and to mitigate the effects of global warming and climate change. In order to better utilize electric energy, firstly, load monitoring is needed to be carried out on the power utilization load, the load monitoring is to acquire power utilization data of users and grasp detailed running states and use frequencies of the user load, and sampling and mining of power utilization behaviors are key links for constructing flexible interactive intelligent power utilization.
The power load monitoring method can be roughly divided into an invasive type and a non-invasive type, and invasive load monitoring usually requires a corresponding data acquisition and sensing device to be installed on each electrical appliance load, which brings a large amount of consumption of manpower and material resources. The non-invasive load monitoring only needs to install non-invasive monitoring equipment at an inlet of power supply, and the type and the operation condition of a single load in a load cluster can be analyzed and obtained by monitoring signals such as voltage, current and the like at the position, so that the online monitoring of all the electric loads in the whole system range is realized. Compared with the traditional intrusive load monitoring method, the non-intrusive load monitoring (NLIM) does not interrupt the power supply of equipment, is easy to be accepted by users, does not need to install a large amount of detection equipment, saves the investment and time required for purchasing, installing and maintaining the hardware equipment, and is an important development direction of future power load monitoring.
Non-intrusive load monitoring techniques can be generally classified as data measurement, event detection, load identification. For the power grid, load monitoring can help power grid enterprises to carefully master load composition and user power utilization behaviors, guidance is provided for planning and power generation scheduling of the power grid, and service is provided for bidirectional interaction and intelligent power utilization. For the user, the load monitoring can give specific power utilization information of each power utilization device, so that the user is guided to change power utilization habits, power utilization behaviors are optimized, and power utilization cost is saved. The combination of load monitoring and intelligent power utilization is perfectly embodied in the intelligent power grid, and the non-invasive load monitoring technology has wide application prospect and can bring benefits for various aspects such as users, power companies and the like.
The traditional load monitoring technology mostly utilizes the steady-state characteristics of the load to identify, such as information of steady-state current, steady-state voltage and the like during the working of the load, and then finds out the optimal combination of each electric load under the current total load based on an optimization problem solution method. And a part of load identification algorithms are carried out based on load transient characteristics, but due to the uniqueness of load characteristic extraction, the most typical characteristics of deep digging electric loads cannot be subdivided, so that the part of algorithms have poor identification effect on similar electric equipment and low-power electric appliances.
Disclosure of Invention
Aiming at the problems that the existing method has complex steady-state characteristic modeling and low load identification degree; the transient characteristics provide a load state identification algorithm based on PSOSC-MDTW (Particle swarm optimization-System Clustering-Multi-dimensional DTW) for identifying the state of the power load in the home office environment, aiming at the problems of low identification precision of similar equipment and low power load.
The invention specifically comprises the following steps:
step1, constructing a load characteristic template library
1-1, acquiring an active power time sequence and a reactive power time sequence when the electrical load is switched in a household or office environment, and recording the two power time sequences as transient power waveforms TPW of the electrical load.
And acquiring an event label corresponding to the switching event, and forming a complete data set dataset by the transient power waveforms TPW of all the electric loads. And randomly taking 50% of data from the transient power waveform TPW of each electric load as a training set, recording the data as TR, and using the data for selecting a load characteristic template library. And taking the other 50% of data in the dataset as a test set, recording as TE, and using the TE for the performance test of the electrical load state identification method.
1-2, calculating element omega in training setiMaximum cluster class number mimaxWherein Ω isiA set of transient power waveforms TPW representing the ith electrical load.
1-3, initially, set ΩiEach of whichThe samples are all a single class, and the inter-class distance is the same as the inter-sample distance. Compute set ΩiThe DTW distance between each class in the image and obtain the distance matrix D (omega) between the classesi)。
1-4, pair set omegaiThe two types of samples with the middle distance being the closest are merged, and the rest types of samples are kept unchanged. The set Ω is represented by C (j, k)iThe j-th type sample C (j) and the k-th type sample C (k) are combined.
Figure GDA0002489215640000021
Wherein D (j, k) represents the set ΩiDTW distance between class j and class k samples in (1).
1-5, comparison TC-1 and a preset cluster number miIn relation to each other, if TC-1＝miIf yes, jumping to 1-6, and outputting the current classification condition of the sample; if TC-1>miThen return to 1-3 to recalculate the inter-class distance matrix.
1-6, dividing the data into m after systematic clusteringiSample class, set denoted C1,C2,…,Cr,…,Cmi， r∈1,2,…,miBy mrRepresenting the number of samples selected from the r set, then
Figure GDA0002489215640000022
Wherein N isrIndicating the number of samples in the r-th set.
1-7, will set omegaiThe selected sample set is expressed as kui
Figure GDA0002489215640000023
Wherein
Figure GDA0002489215640000024
Represents a slave set CrA set of samples is selected.
1-8, for training set TR ═ Ω1,Ω1,…,ΩnAll the sets in the method are subjected to clustering screening to obtain a complete load characteristic template library ku
ku＝{ku1,ku2,…,kui}
Step2, identifying the state of the electrical load
2-1, selecting TPW to be tested from the test set TE and recording the TPW as TeAnd recording the typical template waveform in the load characteristic template library ku as Tz。
2-2, adopting the following electric load state identification rule: will TeAnd TzAnd performing DTW calculation, finding out a typical template waveform with the minimum DTW distance with the TPW to be detected, and giving a state label to the TPW to be detected, thereby obtaining the load state of the electric load.
Figure GDA0002489215640000031
De,z(Te,Tz) Represents the calculation of TeAnd TzThe distance between DTW and argmin () represents the value De,z(Te,Tz) The value of the variable at the minimum time,
Figure GDA0002489215640000032
representing a typical template waveform selected from the load characteristic template library that is most similar to the sample to be tested.
Step3, carrying out parameter optimization on the load state identification method
3-1: setting evaluation index of load state identification method
The evaluation indexes of the electric load state identification comprise precision, sensitivity and F measurement. Precision refers to the probability of a class C sample; sensitivity refers to the probability that a class C sample is correctly identified. The F-measure is a comprehensive indicator of the combination of accuracy and sensitivity.
3-2: parameter optimization
1) Obtaining a comprehensive evaluation index F metric
According to the complete load characteristic template library ku and the test data set TE, nearest neighbor template matching is carried out to obtain F measurement
F＝lab(dtw(g(m1,m2,…,mn,TR),TE))
Wherein n represents the number of electrical load types, m1,m2,…,mnRespectively representing the clustering class number of each type of electric load, g () representing the load characteristic template library constructed in the step1, dtw () representing the electric load state identification rule in the step 2-2, and lab () representing the mapping relation of the F measurement.
2) Optimization process
M is to be1,m2,…,mnAnd as a variable parameter, the comprehensive evaluation index F metric is used as an optimization target, and the PSO optimization is carried out on the load identification method. The optimization objective function is shown below:
maxF(H)＝lab(dtw(g(H,TR),TE))
Figure GDA0002489215640000033
wherein Z+Representing a positive integer.
The invention has the beneficial effects that:
the method selects the transient power waveform during load switching in the home and office environment as the characteristic and is used for load identification; aiming at the difficulties existing in the identification of small power and similar loads, the invention provides the method for carrying out system clustering on similar loads TPW, selecting a typical TPW template, creating a load template characteristic library and improving the identification rate of the loads. Because the negative template is finer, the template library obtained by screening has more typicality and higher accuracy in load state matching identification.
Aiming at the difficulty existing in the increase of the load types, the invention provides the method for optimizing the parameters by utilizing the PSO to determine the optimal combination of the load typical TPW cluster number so as to ensure that the comprehensive performance of the load state identification algorithm is optimal. The invention can still ensure the best overall performance of load state identification when the load types are more.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
In order to more accurately and quickly identify the working state of the power load, firstly, a transient power waveform TPW (transient power waveform) during switching of the power load (such as a refrigerator, an air conditioner, a water dispenser and the like) in a home environment and an office environment and an accurate label during load switching are acquired so as to be conveniently used for load identification and algorithm performance evaluation. Then, aiming at the difficulties existing in the identification of low power and similar loads, the invention provides the steps of carrying out system clustering on similar loads TPW, selecting a typical TPW template and improving the identification rate of the loads. Finally, in order to solve the difficulty of increasing the load types, the invention provides a method for optimizing parameters by using a PSO algorithm and determining the optimal combination of load typical TPW cluster number so as to ensure that the comprehensive performance of the load state identification algorithm is optimal. The method provided by the invention makes important contributions to load state identification and intelligent power utilization in the home and office environment.
The method comprises the following steps:
1. load characteristic template library constructed based on MDTW-SC method
Aiming at the problems that a general load identification algorithm is complex in modeling based on steady-state characteristics and low in load identification degree; based on the problems of low identification precision of transient characteristics to similar equipment and low power load and the like, a load characteristic template library screening method based on MDTW-SC (Multi dimensional DTW-System Clustering) is firstly designed.
The basic idea is as follows: in home and office environments, there are many electrical loads (such as refrigerators, air conditioners, water dispensers, and the like), and for a specific electrical load, there are many typical load states, and for example, an air conditioner has many typical load states such as start-up, shut-down, cooling, heating, and the like. Therefore, when the template library of the electrical load is selected, the typical working waveform of the load needs to be accurately selected into the template library, so that the matching state can be identified more quickly and accurately.
The method comprises the following specific steps:
STEP1, obtaining the active power time sequence P ═ P when the electrical load is switched in the home or office environment1,p2,…,pE]TAnd a reactive power time series Q ═ Q1,q2,…,qE]TAnd E represents the length of the two power time series, and the two power time series are recorded as the transient power waveform TPW of the electric load.
And acquiring an event label corresponding to the switching event, and forming a complete data set dataset by the transient power waveforms TPW of all the electric loads. And randomly taking out 50% of TPW of each electric load as a training set, recording the training set as TR, and using the training set for selecting a load characteristic template library.
Let TR be { omega ═ Ω1,Ω1,…,ΩnWhere n denotes the kind of electrical load in the training set TR, ΩiAnd i ═ {1,2, …, n } represents the set of the i-th electrical loads TPW. Set omegaiThe specific representation form of (A) is as follows:
Figure GDA0002489215640000051
Niindicates the TPW number of the i-th electric load, [ P, Q ]]f， f＝{1,2,…,NiThe concrete expression of TPW.
And taking the other 50% of data in the dataset as a test set, called TE, and using the test set for the performance test of the power load working state identification algorithm.
STEP2, pair set omegaiMaximum cluster class number mimaxAnd (6) performing calculation. m isimaxAnd the cluster type represents the most aggregatable class of the ith electric load after clustering, namely the maximum value of the cluster number of the ith electric load. The invention intends to combine the set omegaiPoly is miClass, miCalled the preset cluster class number, easy to know mi≤mimax。
Figure GDA0002489215640000052
STEP3, initial caseSet omegaiEach sample in the group is a separate class, and the inter-class distance is the same as the inter-sample distance. Compute set ΩiThe DTW distance between each class (the shortest distance method is used for the distance between the classes), and an inter-class distance matrix D (omega) is obtainedi) The concrete expression is shown in formula (2).
Figure GDA0002489215640000053
Wherein D (j, k), j, k ═ {1,2, …, TCDenoted is the set omegaiDTW distance between class j and class k samples in (1). m isi≤TC≤Ni，TC∈Z+Represents the set omegaiThe current cluster class number.
STEP4, pair set omegaiThe two types of samples with the middle distance being the closest are merged, and the rest types of samples are kept unchanged. Where C (j, k) represents the set ΩiThe j-th type sample C (j) and the k-th type sample C (k) are combined.
Figure GDA0002489215640000054
STEP5, COMPARATIVE TC-1 and a preset cluster number miIn relation to each other, if TC-1＝miIf yes, jumping to STEP6, and outputting the current classification condition of the sample; if TC-1>miThen, returning to STEP3, the inter-class distance matrix is recalculated.
STEP6, subdividing data into m after system clusteringiA sample class, set represented as
Figure GDA0002489215640000055
M is calculated according to equation (4)r，mrRepresenting the number of samples taken from the r-th set, NrIndicating the number of samples in the r-th set. And selecting sufficient samples in a random selection mode so as to construct a load characteristic template library.
Figure GDA0002489215640000061
STEP7, set ΩiThe selected sample set is expressed as kuiThe expression form is shown in formula (5). Wherein
Figure GDA0002489215640000062
Represented by a set CrA set of samples is selected.
Figure GDA0002489215640000063
STEP8, training set TR ═ Ω1,Ω1,…,ΩnAnd (4) all the sets in the database are subjected to cluster screening to obtain a complete load characteristic template library ku, wherein the form of the complete load characteristic template library ku is shown in a formula (6).
ku＝{ku1,ku2,…,kui} (6)
2. Identifying electrical load conditions
The basic idea is as follows: first, a test set sample is obtained, which is derived from the test set T, as a training set structureeAnd then based on the nearest neighbor matching principle, giving the label identified by the algorithm to the load to be measured.
STEP1, test set TE and training set TR together form data set Dataset, so the transient power waveform to be tested, called T, is first selected from TEeIts specific expression form is Te＝[P,Q]^The template waveform in the template library ku is called Tz。
STEP2, identifying the electric load state according to the electric load state identification rule shown in formula (7), and converting T into TeAnd respectively carrying out DTW calculation on the typical template waveforms in the load characteristic template library ku to find out the typical template waveform with the minimum DTW distance with the TPW to be detected, and giving a state label to the TPW to be detected, so as to obtain the load state of the electric load.
Figure GDA0002489215640000064
D in formula (7)e,z(Te,Tz) Represents the calculation of TeAnd TzThe distance between DTW and argmin () represents the value De,z(Te,Tz) The value of the variable at the minimum time,
Figure GDA0002489215640000065
representing a typical template waveform selected from the load characteristic template library that is most similar to the sample to be tested.
3. And optimizing parameters based on the load state identification method of PSOSC-MDTW.
Because of a plurality of load types, in order to improve the overall recognition effect of the algorithm on all loads, the invention provides a multi-dimensional DTW (PSO System Clustering-Multi dimensional DTW, PSOSC-MDTW) load state recognition algorithm based on PSO optimization and System Clustering, sets a comprehensive evaluation index of a load state recognition method, performs parameter optimization by using the PSO algorithm, and determines the optimal combination among the typical working waveform cluster numbers of the load switching state, so that the comprehensive evaluation index of the load state recognition is optimal.
Step 1: setting evaluation index of load state identification method
The evaluation indexes for the electrical load state identification generally include accuracy and sensitivity and F-measure. Precision refers to the probability of a class C sample, denoted as PC(ii) a Sensitivity is the probability that the class C sample is correctly identified, denoted as SC. F measurement is a comprehensive index combining precision and sensitivity, and F measurement of class C is marked as FC. And defining a comprehensive evaluation index F measurement, and marking as F. The specific index is defined as follows.
1) Accuracy of measurement
Figure GDA0002489215640000071
Wherein C represents the total class of electrical loads in the test set TE, TPCIndicates the number of correctly identified class C samples, FPCIndicating a fault asNumber of class C samples.
2) Sensitivity of the probe
Figure GDA0002489215640000072
Wherein, FNCIndicating a class C sample that was incorrectly identified as another class sample.
3) F measurement
Figure GDA0002489215640000073
Figure GDA0002489215640000074
Wherein N is the number of the types of the electric loads, NiThe total number of events for the i-th class of electrical loads. The index F is a comprehensive reflection of the identification precision and sensitivity of all the electric loads.
Step 2: and optimizing parameters of the load state identification method based on PSOSC-MDTW.
1) Obtaining a comprehensive evaluation index F metric
And (3) performing nearest neighbor template matching according to the complete load characteristic template library ku and the test data set TE to obtain a comprehensive F measurement index shown in a formula (12).
F＝lab(dtw(g(m1,m2,…,mn,TR),TE)) (12)
Wherein n represents the number of electrical loads, m1,m2,…,mnRespectively representing the clustering number of each type of electric load, TR and TE respectively representing a training set and a test set, g () representing the construction of a load characteristic template base based on the MDTW-SC method, dtw () representing the electric load state identification rule shown in the formula (7), and lab () representing the calculation of the electric load F measurement based on the formula (11).
2) And (4) optimizing a load state identification algorithm.
The number m of the typical working waveform clusters during each load switching1,m2,…,mnAnd as a variable parameter, the comprehensive evaluation index F measurement is used as an optimization target, and the load recognition algorithm is subjected to PSO optimization. The optimization objective function is shown in equation (13).
maxF(H)＝lab(dtw(g(H,TR),TE))
Figure GDA0002489215640000081
The optimization process is shown in fig. 1, and the specific steps of the load identification algorithm are as follows.
STEP1, acquiring original power data, constructing a training set TR and a test set TE, and a set of transient power data { omega } of various electric devices with clear event labels in the training set1,Ω1,…,Ωn}。
STEP2, initializing parameters including population size M of particle swarm, and setting constraint condition Vmax， Vmin，Xmax，Xmin(ii) a Maximum number of iterations MgenDimension n of the search space, boundary button and corresponding iteration end conditions.
STEP3, initializing velocity information and position information V of each particlei 0＝[vi1vi2…vin]T，
Figure GDA0002489215640000082
Wherein xik∈Z+,xik≤mkmax，k＝{1,2,…,n}，mkmaxRepresents the maximum number of clusters for the kth load determined by equation (1).
STEP4, the initialized location, where each location information is actually a combination of load cluster class numbers, then calculates the F metric, i.e., fitness value, according to equation (13).
STEP5, calculating the fitness value of all the particles in the particle group according to the set objective function, and then evaluating the historical optimal solution P of the particlesbestAnd global optimal solution Gbest。
STEP6, judging whether the fitness value corresponding to the global optimal solution meets the requirement set by the algorithm; or whether the number of iterations has reached a maximum. When any one of the requirements is met, the algorithm finishes iteration and outputs an optimal solution; otherwise, jump to STEP 7.
STEP7 updates the particle velocity information according to equation (14). The first part in the formula is called a memory term and represents the influence of the last speed magnitude and direction; the second part is called self-knowledge item and represents the part of the particle whose action comes from self experience; the third part is called group recognition item, which reflects cooperative cooperation and knowledge sharing among particles, and the particles determine the next step of movement through own experience and the best experience of partners.
Figure GDA0002489215640000083
Wherein,
Figure GDA0002489215640000084
representing the velocity of particle i after the kth iteration;
Figure GDA0002489215640000085
representing the individual extreme value of the particle i after the kth iteration and the current global optimal solution;
Figure GDA0002489215640000091
representing the velocity of the particle after the (k + 1) th iteration; rand1()，rand2 () Is a random number between (0, 1), c1、c2The learning factor is expressed, omega is called as an inertia factor, and the global and local optimizing capability of the optimization algorithm is influenced by the value of omega.
STEP8, and new particle position information according to formula (15)
Figure GDA0002489215640000092
Wherein
Figure GDA0002489215640000093
The positions of the particles after the k < th > and k +1 < th > iterations are respectively expressed, and the positions of the particles are related to the positions and the speeds at the new moment and the last moment through a formula.
STEP9 adjusts the velocity and position of the particle according to the constraints shown in equation (16).
Figure GDA0002489215640000094
And after the speed and the position of the particles are adjusted, jumping to STEP5, and then repeatedly executing STEP6 to STEP9 until the iteration of the algorithm is finished, and outputting the global optimal solution and the fitness value corresponding to the optimal solution.
The load state identification method is optimized based on PSOSC-MDTW parameters, the optimized optimal parameters are the optimal combination of the electricity load clustering numbers, and the screened load characteristic template library ku can enable the comprehensive index F of load identification to be very high and is superior to most existing methods.

Claims (2)

1. A method for identifying the state of a power load in a home office environment is characterized by comprising the following steps:
step1, constructing a load characteristic template library
1-1, acquiring an active power time sequence and a reactive power time sequence when the electrical load is switched in a home or office environment, and recording the two power time sequences as transient power waveforms TPW of the electrical load;
acquiring an event label corresponding to a switching event, and forming a complete data set dataset by transient power waveforms TPW of all electrical loads; randomly taking 50% of data of the transient power waveform TPW of each electric load as a training set, recording the data as TR, and selecting a load characteristic template library; taking the other 50% of data in the dataset as a test set, marking as TE, and using the TE for the performance test of the electrical load state identification method;
1-2, calculating element omega in training setiMaximum cluster class number mimaxWherein Ω isiInstantaneous representation of the i-th electrical loadA set of state power waveforms TPW;
1-3, initially, set ΩiEach sample in the system is an independent class, and the distance between the classes is the same as the distance between the samples; compute set ΩiThe DTW distance between each class in the image and obtain the distance matrix D (omega) between the classesi)；
1-4, pair set omegaiMerging the two types of samples with the middle distance being the closest, and keeping the other types of samples unchanged; the set Ω is represented by C (j, k)iThe j-th sample C (j) and the k-th sample C (k) are combined;
Figure FDA0002489215630000011
wherein D (j, k) represents the set ΩiDTW distance between jth and kth class samples in (1)
1-5, comparison TC-1 and a preset cluster number miIn relation to each other, if TC-1＝miIf yes, jumping to 1-6, and outputting the current classification condition of the sample; if TC-1>miReturning to 1-3, and recalculating the inter-class distance matrix;
1-6, dividing the data into m after systematic clusteringiA sample class, set represented as
Figure FDA0002489215630000012
By mrRepresenting the number of samples selected from the r set, then
Figure FDA0002489215630000021
Wherein N isrRepresenting the number of samples in the r set;
1-7, will set omegaiThe selected sample set is expressed as kui
Figure FDA0002489215630000022
Wherein
Figure FDA0002489215630000023
Represents a slave set CrSelecting a set of samples;
1-8, for training set TR ═ Ω1,Ω1,…,ΩnAll the sets in the method are subjected to clustering screening to obtain a complete load characteristic template library ku
ku＝{ku1,ku2,…,kui}
Step2, identifying the state of the electrical load
2-1, selecting TPW to be tested from the test set TE and recording the TPW as TeAnd recording the typical template waveform in the load characteristic template library ku as Tz；
2-2, adopting the following electric load state identification rule: will TeAnd TzPerforming DTW calculation, finding out a typical template waveform with the minimum DTW distance with the TPW to be detected, and giving a state label to the TPW to be detected so as to obtain the load state of the electric load;
Figure FDA0002489215630000024
De,z(Te,Tz) Represents the calculation of TeAnd TzThe distance between DTW and argmin () represents the value De,z(Te,Tz) The value of the variable at the minimum time,
Figure FDA0002489215630000025
representing a typical template waveform which is selected from a load characteristic template library and is most similar to a sample to be detected;
step3, carrying out parameter optimization on the load state identification method
3-1: setting evaluation index of load state identification method
The evaluation indexes of the electric load state identification comprise precision, sensitivity and F measurement; precision refers to the probability of a class C sample; sensitivity refers to the probability that a class C sample is correctly identified; f measurement is a comprehensive index combining precision and sensitivity;
3-2: optimizing parameters;
1) obtaining a comprehensive evaluation index F metric
According to the complete load characteristic template library ku and the test data set TE, nearest neighbor template matching is carried out to obtain F measurement
F＝lab(dtw(g(m1,m2,…,mn,TR),TE))
Wherein n represents the number of electrical load types, m1,m2,…,mnRespectively representing the clustering class number of each type of electric load, g () representing the load characteristic template library constructed in the step1, dtw () representing the electric load state identification rule in the step 2-2, and lab () representing the mapping relation of F measurement;
2) optimization process
M is to be1,m2,…,mnAs variable parameters, the comprehensive evaluation index F measurement is used as an optimization target, and PSO optimization is carried out on the load identification method; optimizing the target function;
maxF(H)＝lab(dtw(g(H,TR),TE))
Figure FDA0002489215630000031
wherein Z+Representing a positive integer.
2. The method according to claim 1, wherein the method comprises the following steps:
precision PCThe calculation is as follows:
Figure FDA0002489215630000032
wherein C represents the total class of electrical loads in the test set TE, TPCIndicates the number of correctly identified class C samples, FPCRepresenting the number of class C samples that were erroneously identified;
sensitivity SCThe calculation is as follows:
Figure FDA0002489215630000033
wherein, FNCIndicating a class C sample that is incorrectly identified as other class samples;
the comprehensive evaluation index metric F is calculated as follows:
Figure FDA0002489215630000034
Figure FDA0002489215630000035
wherein N isiRepresents the total number of samples of the i-th class electrical load.
CN202010265417.6A 2020-04-07 2020-04-07 Power load state identification method in home and office environment Active CN111476298B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010265417.6A CN111476298B (en) 2020-04-07 2020-04-07 Power load state identification method in home and office environment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010265417.6A CN111476298B (en) 2020-04-07 2020-04-07 Power load state identification method in home and office environment

Publications (2)

Publication Number Publication Date
CN111476298A CN111476298A (en) 2020-07-31
CN111476298B true CN111476298B (en) 2022-04-26

Family

ID=71750130

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010265417.6A Active CN111476298B (en) 2020-04-07 2020-04-07 Power load state identification method in home and office environment

Country Status (1)

Country Link
CN (1) CN111476298B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111553281A (en) * 2020-04-29 2020-08-18 国网重庆市电力公司营销服务中心 Electrical appliance state identification method and device
CN112085111B (en) * 2020-09-14 2024-01-23 南方电网科学研究院有限责任公司 Load identification method and device

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9639642B2 (en) * 2013-10-09 2017-05-02 Fujitsu Limited Time series forecasting ensemble
CN106226572A (en) * 2016-07-13 2016-12-14 国家电网公司 Household loads recognition methods based on transient characteristic cluster
CN106786534B (en) * 2016-12-28 2019-07-26 天津求实智源科技有限公司 A kind of non-intrusive electrical load transient process discrimination method and system
CN107330517B (en) * 2017-06-14 2021-08-06 华北电力大学 Non-invasive resident load identification method based on S _ Kohonen
CN109754113B (en) * 2018-11-29 2022-08-30 南京邮电大学 Load prediction method based on dynamic time warping and long-and-short time memory

Also Published As

Publication number Publication date
CN111476298A (en) 2020-07-31

Similar Documents

Publication Publication Date Title
CN106786534B (en) A kind of non-intrusive electrical load transient process discrimination method and system
Najafi et al. Data analytics for energy disaggregation: Methods and applications
Liu et al. Energy disaggregation of appliances consumptions using ham approach
US9898442B2 (en) Data processing device, data processing method, and program
Berges et al. User-centered nonintrusive electricity load monitoring for residential buildings
Yu et al. Nonintrusive appliance load monitoring for smart homes: Recent advances and future issues
CN106296465B (en) Method for detecting abnormal electricity utilization behavior of smart power grid
CN109633301B (en) Non-invasive electrical appliance load identification method based on quantum genetic optimization
CN111476298B (en) Power load state identification method in home and office environment
Batra et al. Neighbourhood nilm: A big-data approach to household energy disaggregation
Zhai et al. Analysis of dynamic appliance flexibility considering user behavior via non-intrusive load monitoring and deep user modeling
CN110569876A (en) Non-invasive load identification method and device and computing equipment
CN113484813B (en) Intelligent ammeter fault rate prediction method and system under multi-environment stress
Rehman et al. Comparative evaluation of machine learning models and input feature space for non-intrusive load monitoring
CN111046913A (en) Load abnormal value identification method
Parson et al. A scalable non-intrusive load monitoring system for fridge-freezer energy efficiency estimation
Ghosh et al. Cross-correlation based classification of electrical appliances for non-intrusive load monitoring
Yang et al. Mining the big data of residential appliances in the smart grid environment
Giri et al. Novel techniques for the detection of on and off states of appliances for power estimation in non-intrusive load monitoring
Liu et al. Automatic appliance labeling for unsupervised NILM based on hierarchical decision-making
Karijadi et al. Electricity Load Prediction using Fuzzy c-means Clustering EMD based Support Vector Regression for University Building
Chen et al. Technique of Feature Extraction Based on Interpretation Analysis for Multilabel Learning in Nonintrusive Load Monitoring With Multiappliance Circumstances
Pavlou et al. Non-intrusive water usage classification considering limited training data
Murti et al. Comparison of Decision Tree and K-Nearest Neighbors Methods on Classifying Household Electrical Appliances Based on Electricity Usage Profiles
CN117220417B (en) Dynamic monitoring method and system for consumer-side electrical load

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