CN112434783A - Non-invasive load decomposition method and system - Google Patents
Non-invasive load decomposition method and system Download PDFInfo
- Publication number
- CN112434783A CN112434783A CN202011081904.3A CN202011081904A CN112434783A CN 112434783 A CN112434783 A CN 112434783A CN 202011081904 A CN202011081904 A CN 202011081904A CN 112434783 A CN112434783 A CN 112434783A
- Authority
- CN
- China
- Prior art keywords
- total power
- decomposed
- power data
- data
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 77
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000010606 normalization Methods 0.000 claims abstract description 65
- 238000012545 processing Methods 0.000 claims abstract description 36
- 230000008569 process Effects 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims description 25
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 17
- 238000010586 diagram Methods 0.000 description 13
- 238000005516 engineering process Methods 0.000 description 10
- 238000000605 extraction Methods 0.000 description 10
- 238000004590 computer program Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 7
- 238000005070 sampling Methods 0.000 description 5
- 238000012544 monitoring process Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000005265 energy consumption Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000015654 memory Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000005406 washing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a non-invasive load decomposition method and a system, wherein the method comprises the following steps: carrying out normalization processing on total power data to be decomposed; decomposing the normalized total power data to be decomposed based on a pre-trained decomposition network to obtain power normalization values of various electrical appliances in the total power data to be decomposed; carrying out reverse normalization processing on the power normalization values of various electrical appliances in the total power data to be decomposed to obtain the power values of various electrical appliances in the total power data to be decomposed; the technical scheme provided by the invention can improve the efficiency of non-invasive load decomposition and reduce the operation difficulty of the non-invasive load decomposition and errors generated in the operation process.
Description
Technical Field
The invention relates to the technical field of automatic analysis of power systems, in particular to a non-invasive load decomposition method and system.
Background
The load decomposition technology and the load monitoring technology are key technologies for realizing the smart grid, the traditional load power utilization monitoring generally adopts an intrusive method, namely, a sensor is installed on an electric appliance of a user, and then the use condition of the user on each electric appliance is recorded.
The advent of non-intrusive load splitting, a process for estimating the energy consumed by individual appliances by analyzing meter readings throughout a home, remedies the deficiencies in intrusive load splitting. In other words, non-intrusive load shedding is the ability to separate (estimate) the energy consumption of each appliance from a single household meter without the need to install a large number of monitoring devices in the household appliance. The non-invasive load decomposition can monitor energy sources, detect various electrical appliance faults and analyze the faults by means of an intelligent technology, so that the technology has great application value in optimal configuration and analysis of electric power.
There are many related studies directed to non-intrusive load splitting techniques. For example: the particle swarm optimization algorithm carries out non-invasive load decomposition experiments on a small number of electric appliances, the algorithm can decompose total power data to all electric appliances at the same time, but the obtained decomposition result error is still large. Aiming at the problem of large decomposition error, the sparse optimization algorithm carries out a non-invasive load decomposition algorithm, and the decomposition error is reduced to a certain extent. The multi-factor hidden Markov algorithm carries out a non-invasive load decomposition experiment, a single electric appliance hidden Markov chain is utilized, then a kronecker product is utilized to obtain a combined parameter of the combined hidden Markov chain, in a decomposition stage, the multi-factor hidden Markov can search the optimal combination according to given total power data and determined parameters, the method is similar to a combined optimization algorithm, firstly, electric appliance state power is obtained through clustering, the coding and decoding process of the algorithm is a process of optimizing power values obtained through clustering, the result obtained through decomposition is also the combination of the power values obtained through clustering, and a more accurate electric appliance power consumption value cannot be obtained. Other researches, such as algorithms based on Adaboost algorithm, K-nearest neighbor algorithm, support vector machine algorithm, fuzzy algorithm, neural network algorithm and the like, all obtain certain achievement in the non-invasive load decomposition task.
However, the above algorithms mainly have several problems, the above decomposition errors are relatively large, the efficiency is relatively low, while the accuracy is improved, a large amount of load characteristic information needs to be added, the operation difficulty of obtaining the characteristic information in the actual non-invasive load decomposition task is large, a large amount of manual extraction is needed, the manual extraction of the characteristics needs to spend a large amount of time, the manually designed characteristic extraction method is poor in noise robustness, the nilm method based on the optimization theory is a discretization algorithm, and is only suitable for analyzing the switching states of the electrical appliances, the total power data cannot be accurately decomposed into the continuous power values of the electrical appliances, and the above methods are difficult to accurately decompose the low-frequency total power data into the power consumption of the electrical appliances.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a non-invasive load decomposition method which can achieve the aim of quickly and accurately acquiring the power values of various electrical appliances.
The purpose of the invention is realized by adopting the following technical scheme:
the invention provides a non-invasive load decomposition method, which is improved in that the method comprises the following steps:
carrying out normalization processing on total power data to be decomposed;
decomposing the normalized total power data to be decomposed based on a pre-trained decomposition network to obtain power normalization values of various electrical appliances in the total power data to be decomposed;
and performing inverse normalization processing on the power normalization values of various electrical appliances in the total power data to be decomposed to obtain the power values of various electrical appliances in the total power data to be decomposed.
Preferably, the normalizing process performed on the total power data to be decomposed includes:
determining the normalized value x of the ith total power value in the total power data to be decomposed according to the following formulai *：
In the above formula, xmaxIs the maximum value of power, x, in the total power data to be decomposedminIs the minimum value of power, x, in the total power data to be decomposediAnd the number is the ith total power value in the total power data to be decomposed, I belongs to I, and I is the total number of the total power values in the total power data to be decomposed.
Preferably, the obtaining process of the pre-trained decomposition network includes:
normalizing the historical total power data and the power values of various electrical appliances in the historical total power data;
taking the normalized value of the historical total power data as input layer training data of an initial residual error gated circulation unit network, taking the normalized value of the power value of each type of electrical appliances in the historical total power data as output layer training data of the initial residual error gated circulation unit network, and training the initial residual error gated circulation unit network by adopting a gradient descent method;
and the data length of the historical total power data is the same as that of the total power data to be decomposed.
Further, the decomposing network based on pre-training decomposes the total power data to be decomposed after the normalization processing to obtain power normalization values of various electrical appliances in the total power data to be decomposed, including:
and inputting the normalized total power data to be decomposed into a pre-trained decomposition network, and acquiring power normalization values of various electrical appliances in the total power data to be decomposed output by the pre-trained decomposition network.
Preferably, the inverse normalization processing is performed on the power normalization values of various electrical appliances in the total power data to be decomposed, and includes:
determining the power value of the kth type electrical appliance in the total power data to be decomposed according to the following formula
In the above formula, xmaxFor the first in the total power data to be decomposedMaximum power of class k appliances, xminIs the power minimum value, x, of the kth electrical appliance in the total power data to be decomposedpredAnd obtaining a power normalization value of the kth electrical appliance in the total power data to be decomposed, wherein K belongs to K, and K is the total number of the electrical appliances in the total power data to be decomposed.
Based on the same inventive concept, the invention also provides a non-intrusive load decomposition system, and the improvement is that the system comprises:
a processing module: the device is used for carrying out normalization processing on total power data to be decomposed;
a decomposition module: the system comprises a decomposition network, a power normalization value acquisition unit, a power normalization value generation unit and a power normalization value generation unit, wherein the decomposition network is used for decomposing the total power data to be decomposed after normalization processing based on a pre-trained decomposition network to obtain the power normalization value of each electric appliance in the total power data to be decomposed;
an acquisition module: and the device is used for performing inverse normalization processing on the power normalization values of various electrical appliances in the total power data to be decomposed to obtain the power values of various electrical appliances in the total power data to be decomposed.
Preferably, the processing module is specifically configured to:
determining the normalized value x of the ith total power value in the total power data to be decomposed according to the following formulai *：
In the above formula, xmaxIs the maximum value of power, x, in the total power data to be decomposedminIs the minimum value of power, x, in the total power data to be decomposediAnd the number is the ith total power value in the total power data to be decomposed, I belongs to I, and I is the total number of the total power values in the total power data to be decomposed.
Preferably, the obtaining process of the pre-trained decomposition network includes:
normalizing the historical total power data and the power values of various electrical appliances in the historical total power data;
taking the normalized value of the historical total power data as input layer training data of an initial residual error gated circulation unit network, taking the normalized value of the power value of each type of electrical appliances in the historical total power data as output layer training data of the initial residual error gated circulation unit network, and training the initial residual error gated circulation unit network by adopting a gradient descent method;
and the data length of the historical total power data is the same as that of the total power data to be decomposed.
Preferably, the decomposition module is specifically configured to:
and inputting the normalized total power data to be decomposed into a pre-trained decomposition network, and acquiring power normalization values of various electrical appliances in the total power data to be decomposed output by the pre-trained decomposition network.
Preferably, the obtaining module is specifically configured to:
determining the power value of the kth type electrical appliance in the total power data to be decomposed according to the following formula
In the above formula, xmaxIs the maximum value of the power, x, of the kth electrical appliance in the total power data to be decomposedminIs the power minimum value, x, of the kth electrical appliance in the total power data to be decomposedpredAnd obtaining a power normalization value of the kth electrical appliance in the total power data to be decomposed, wherein K belongs to K, and K is the total number of the electrical appliances in the total power data to be decomposed.
Compared with the closest prior art, the invention has the following beneficial effects:
the technical scheme provided by the invention provides a non-invasive load decomposition method and a non-invasive load decomposition system, and the method comprises the following steps of firstly, carrying out normalization processing on total power data to be decomposed; secondly, decomposing the normalized total power data to be decomposed based on a pre-trained decomposition network to obtain power normalization values of various electrical appliances in the total power data to be decomposed; and finally, performing inverse normalization processing on the power normalization values of various electrical appliances in the total power data to be decomposed to obtain the power values of various electrical appliances in the total power data to be decomposed. The scheme can obtain the power values of various electrical appliances through the pre-trained decomposition network model, improve the efficiency of non-invasive load decomposition, reduce the operation difficulty of the non-invasive load decomposition and the errors generated in the operation process, and greatly save the time of the non-invasive load decomposition.
Drawings
FIG. 1 is a flow chart of a non-intrusive load splitting method provided by the present invention;
FIG. 2 is a schematic diagram of a sliding sampling principle in an embodiment of the invention;
FIG. 3 is a diagram of a network of initial residual gated loop units in an embodiment of the invention;
FIG. 4 is a diagram of a residual unit structure in an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a non-intrusive load splitting system provided by the present invention.
Detailed Description
The following describes embodiments of the present invention in further detail with reference to the accompanying drawings.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to analyze the power consumption of each electrical appliance to obtain the power consumption behavior of a user, the traditional load power consumption monitoring generally adopts an intrusive method, namely, a sensor is installed on the electrical appliance of the user, and then the use condition of the user on each electrical appliance is recorded.
The non-invasive load decomposition technology overcomes the defects in the invasive load decomposition, and the non-invasive load decomposition can analyze the energy consumption of each electric appliance from a single household electric meter without installing a large amount of monitoring equipment in the household electric appliance. The non-invasive load decomposition can monitor energy sources, detect various electrical appliance faults and analyze the faults by means of an intelligent technology, so that the technology has great application value in optimal configuration and analysis of electric power.
At present, many related researches are carried out on a non-invasive load decomposition technology, but the decomposition error is relatively large, the efficiency is relatively low, the operation difficulty is high, a large amount of manual extraction is needed, a large amount of time is needed for manually extracting features, and the low-frequency total power data are difficult to be accurately decomposed into the power consumption of each electric appliance.
Aiming at the defects of large error, low efficiency, high operation difficulty and the like of the existing non-invasive load decomposition technology, the invention provides a non-invasive load decomposition method, as shown in figure 1, the method comprises the following steps:
step (1) normalizing total power data to be decomposed;
decomposing the normalized total power data to be decomposed based on a pre-trained decomposition network to obtain power normalization values of various electrical appliances in the total power data to be decomposed;
and (3) carrying out reverse normalization processing on the power normalization values of various electrical appliances in the total power data to be decomposed, and obtaining the power values of various electrical appliances in the total power data to be decomposed.
Specifically, in step (1), the normalized value x of the ith total power value in the total power data to be decomposed may be determined according to the following formulai *：
In the above formula, xmaxTo be decomposed intoMaximum value of power, x, in total power dataminIs the minimum value of power, x, in the total power data to be decomposediAnd the number is the ith total power value in the total power data to be decomposed, I belongs to I, and I is the total number of the total power values in the total power data to be decomposed.
In an embodiment provided by the present invention, the obtaining process of the pre-trained decomposition network includes:
normalizing the historical total power data and the power values of various electrical appliances in the historical total power data;
taking the normalized value of the historical total power data as input layer training data of an initial residual error gated circulation unit network, taking the normalized value of the power value of each type of electrical appliances in the historical total power data as output layer training data of the initial residual error gated circulation unit network, and training the initial residual error gated circulation unit network by adopting a gradient descent method;
in the aspect of training data selection, in the embodiment provided by the invention, a sliding window method can be adopted to perform sliding processing on the whole total power time sequence data and the power consumption time sequence data of each electric appliance, the sliding window method is similar to convolution operation, and the difference from the convolution operation is that the sliding window only performs sliding processing on the time sequence data, and each sliding can obtain time sequence data with a time sequence length equal to the size of the sliding window, unlike the convolution operation that the data is subjected to filtering processing. The purpose of sliding sampling is to select a certain amount of training data to be sent to the network model.
In the above scheme, the data length of the historical total power data is the same as the data length of the total power data to be decomposed.
In the preferred embodiment of the present invention, the sliding sampling process may be:
and taking the original total power data as input data, and taking the real power of each electrical appliance as a training label. The design of the invention adopts sliding input, namely the data size input by the network each time is a matrix with the size of [ N, input _ size ] on the right side as shown in figure 2. In fig. 2, M represents the total number of times of sampling data for one family, and N represents the size of the sliding window. The input data of this time is the original total power and differential data, so the value of input _ size is 3. And intercepting a section of input data on the starting position of the input data X by the window with the length of N, and then intercepting the input data at a moment when the window does not slide downwards until the sequence is finished. Therefore, M-N +1 pieces of data are intercepted, and a matrix with the dimension of [ M-N +1, N, input _ size ] is formed.
The original total power data uses public data, the used experimental data is from public data sets UK _ DALE and WikiEnergy, and the WikiEnergy data set is power data issued by a unit. The data contains power consumption data of more than six hundred households, wherein the data comprises a household main power supply and data of each independent household power supply, and the sampling frequency of the data set is 1/60 HZ. The device starts in month 1 of 2011 and is still collecting data for most buildings. The WiKiEnergy dataset contains 600 long-term power consumption information of various appliances of a plurality of households and the total amount of power consumption of the whole household. Among them, the historical power consumption data of air conditioners, refrigerators, washing machines, dishwashers, microwave ovens in No. 18 homes was used for the load breaking task.
Further, in the preferred embodiment of the present invention, the initial residual gated loop cell network comprises three parts: multi-scale convolution kernel processing, a residual error network and a gated cyclic unit network. The multi-scale convolution adopts convolution cores with three different sizes to perform convolution operation on the total power data. The characteristic information, namely characteristic graphs, is obtained after the characteristic extraction of the convolutions of three different scales, and the characteristic graphs obtained by each convolution are spliced, so that the characteristic information of the shallow layers of different scales can be obtained, and then the characteristic information of the shallow layers of different scales is sent into a residual error network. The high-order semantic feature information is further obtained by processing the shallow feature information through the residual error network, and the residual error network can further deepen the network to solve the problem of gradient disappearance. And finally, updating and maintaining the high-order semantic feature information by using a gating mechanism of a gating cycle unit network so as to obtain an output power value of the standard electrical appliance, wherein as shown in fig. 3, the multi-scale convolution kernel is divided into three sizes: 1x1, 1x3, 1x 5. The Residual network includes Residual block 1(Residual block1), Residual block 2(Residual block2), and Residual block 3(Residual block 3). The Residual block1 includes 3 Residual units (Residual units), the Residual block2 includes 4 Residual units (Residual units), and the Residual block3 includes 4 Residual units (Residual units). GRU denotes a gated round-robin unit network. The first GRU is followed by n branches, one GRU after the other on each branch. dense denotes the fully connected layer, which has a length N. The sequence length of the output of each dense is N. the number of targetsequence is n.
In the embodiment provided by the invention, the number of the target electrical appliances can be assumed to be n. The inputsequence represents the input total power data, has a length of N, represents the total power data of N different time instants, and is the input data of the network model. And the target sequence has N groups, each group represents one target electrical appliance, and the length of each group is also N and represents the target electrical appliance power value of N different time points output by the network model. In the model training phase, the total power data is required to be input into the network model as input data, and the power data of the n target electrical appliances are also input into the network model as tag data. And the total power data and the power data of the n target electrical appliances are used as training data to train and learn the network model.
After training is finished, the network model can predict the power of n target electrical appliances according to the input total power data, based on the model, the invention inputs the normalized total power data to be decomposed into the decomposition network trained in advance, and obtains the power normalization value of various electrical appliances in the total power data to be decomposed output by the decomposition network trained in advance, namely the power of the n target electrical appliances.
Specifically, in the embodiment of the invention, a group of multi-scale convolutions is arranged in an initial residual gated cyclic unit network to perform shallow multi-scale time sequence feature extraction, three convolution scales of 1x1, 3x1 and 5x1 are used in the first layer of multi-scale convolution to perform multi-scale feature extraction, the number of 1x1 convolution kernels is 6, the number of 3x1 convolution kernels is 10, and the number of 5x1 convolution kernels is 14, after feature extraction of convolution of three different scales, feature maps obtained by each convolution are subjected to splicing operation, so that shallow feature information of different scales can be obtained, and then the shallow feature information is sent to a residual network part.
In the embodiment of the invention, each residual unit in the residual network comprises three convolution operations, the sizes of convolution layer convolution kernels in each residual unit are 1x1, 3x1 and 1x1 respectively, and before each convolution operation, BatchNorm (batch normalization), ReLU function activation and the like are used. And constructing a residual block by taking the residual unit as a basic module, wherein the three residual blocks form a residual network. The residual unit is shown in fig. 4.
The residual error unit structure is simpler. Assuming that in a single residual unit structure, the output of the residual block is h (x), according to the thought source of the residual unit structure, we can fit an arbitrary function by a neural network stacked by a plurality of hidden layers, in the residual unit structure, the function that we need to fit is h (x) -x, and the input of the neural network is added to the output of the stacked hidden layers through cross-layer connection, so as to obtain h (x). Thus, h (x) -x can be defined as f (x), the residual unit structure is shown in fig. 4, each residual unit structure has three layers, and x represents the input.
FIG. 4 is a diagram of the structure of a residual unit, where x is the input to the residual unit, F (x) is the input to the residual unit, and H (x) is the sum of x and F (x).
In the first residual block, deep level feature extraction is performed using a residual block composed of 3 residual units. By analogy, the second residual block in the multi-scale residual network is composed of 4 residual units, the 3 rd residual block is composed of 6 residual units, and the structure of each residual unit is the same as that of the residual unit in the first residual block. The number of convolution kernels of a first residual block is 30, the number of convolution kernels of a second residual block is 40, the number of convolution kernels of a third residual block is 50, a feature map processed by the third residual block actually contains rich multi-scale load features, in a last residual unit, a convolution with a larger scale is used for further extracting deep-level features, then the number of the convolution kernels is 50 for final characterization, and sequence information after feature extraction of a plurality of residual blocks is sent to a gating cycle unit network.
The gated circulation unit network is provided with two gates including a reset gate and an update gate, and the state information is maintained and controlled through the update gate and the reset gate, so that long-term memory is realized. And the sequence information after the network characteristics of the gating circulation unit are extracted is sent into the network of the gating circulation unit, and then the output of the network model, namely the power value of each target electrical appliance, is obtained through a dense layer.
And after power normalization values of various electrical appliances in the total power data to be decomposed output by the pre-trained decomposition network are obtained, performing inverse normalization processing to obtain power values of various electrical appliances in the total power data to be decomposed.
In the embodiment provided by the invention, the power value of the kth type electrical appliance in the total power data to be decomposed is determined by using the following formula
In the above formula, xmaxIs the maximum value of the power, x, of the kth electrical appliance in the total power data to be decomposedminIs the power minimum value, x, of the kth electrical appliance in the total power data to be decomposedpredAnd obtaining a power normalization value of the kth electrical appliance in the total power data to be decomposed, wherein K belongs to K, and K is the total number of the electrical appliances in the total power data to be decomposed.
In summary, the non-invasive load decomposition method provided by the invention can achieve the purpose of accurately and rapidly obtaining the power values of various electrical appliances according to the pre-trained decomposition network.
Based on the same inventive concept, the present invention further provides a non-intrusive load splitting system, as shown in fig. 5, the system includes:
a processing module: the device is used for carrying out normalization processing on total power data to be decomposed;
a decomposition module: the system comprises a decomposition network, a power normalization value acquisition unit, a power normalization value generation unit and a power normalization value generation unit, wherein the decomposition network is used for decomposing the total power data to be decomposed after normalization processing based on a pre-trained decomposition network to obtain the power normalization value of each electric appliance in the total power data to be decomposed;
an acquisition module: and the device is used for performing inverse normalization processing on the power normalization values of various electrical appliances in the total power data to be decomposed to obtain the power values of various electrical appliances in the total power data to be decomposed.
Preferably, the processing module is specifically configured to:
determining the normalized value x of the ith total power value in the total power data to be decomposed according to the following formulai *：
In the above formula, xmaxIs the maximum value of power, x, in the total power data to be decomposedminIs the minimum value of power, x, in the total power data to be decomposediAnd the number is the ith total power value in the total power data to be decomposed, I belongs to I, and I is the total number of the total power values in the total power data to be decomposed.
Preferably, the obtaining process of the pre-trained decomposition network includes:
normalizing the historical total power data and the power values of various electrical appliances in the historical total power data;
taking the normalized value of the historical total power data as input layer training data of an initial residual error gated circulation unit network, taking the normalized value of the power value of each type of electrical appliances in the historical total power data as output layer training data of the initial residual error gated circulation unit network, and training the initial residual error gated circulation unit network by adopting a gradient descent method;
and the data length of the historical total power data is the same as that of the total power data to be decomposed.
Preferably, the decomposition module is specifically configured to:
and inputting the normalized total power data to be decomposed into a pre-trained decomposition network, and acquiring power normalization values of various electrical appliances in the total power data to be decomposed output by the pre-trained decomposition network.
Preferably, the obtaining module is specifically configured to:
determining the power value of the kth type electrical appliance in the total power data to be decomposed according to the following formula
In the above formula, xmaxIs the maximum value of the power, x, of the kth electrical appliance in the total power data to be decomposedminIs the power minimum value, x, of the kth electrical appliance in the total power data to be decomposedpredAnd obtaining a power normalization value of the kth electrical appliance in the total power data to be decomposed, wherein K belongs to K, and K is the total number of the electrical appliances in the total power data to be decomposed.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.
Claims (10)
1. A non-intrusive load splitting method, the method comprising:
carrying out normalization processing on total power data to be decomposed;
decomposing the normalized total power data to be decomposed based on a pre-trained decomposition network to obtain power normalization values of various electrical appliances in the total power data to be decomposed;
and performing inverse normalization processing on the power normalization values of various electrical appliances in the total power data to be decomposed to obtain the power values of various electrical appliances in the total power data to be decomposed.
2. The method of claim 1, wherein normalizing the total power data to be decomposed comprises:
determining the normalized value x of the ith total power value in the total power data to be decomposed according to the following formulai *
In the above formula, xmaxIs the maximum value of power, x, in the total power data to be decomposedminIs the minimum value of power, x, in the total power data to be decomposediAnd the number is the ith total power value in the total power data to be decomposed, I belongs to I, and I is the total number of the total power values in the total power data to be decomposed.
3. The method of claim 1, wherein the acquisition process of the pre-trained decomposition network comprises:
normalizing the historical total power data and the power values of various electrical appliances in the historical total power data;
taking the normalized value of the historical total power data as input layer training data of an initial residual error gated circulation unit network, taking the normalized value of the power value of each type of electrical appliances in the historical total power data as output layer training data of the initial residual error gated circulation unit network, and training the initial residual error gated circulation unit network by adopting a gradient descent method;
and the data length of the historical total power data is the same as that of the total power data to be decomposed.
4. The method according to claim 1, wherein the decomposing the normalized total power data to be decomposed based on the pre-trained decomposing network to obtain power normalization values of various electrical appliances in the total power data to be decomposed comprises:
and inputting the normalized total power data to be decomposed into a pre-trained decomposition network, and acquiring power normalization values of various electrical appliances in the total power data to be decomposed output by the pre-trained decomposition network.
5. The method as claimed in claim 1, wherein the inverse normalization of the power normalization values of the various types of electrical appliances in the total power data to be decomposed includes:
determining the power value of the kth type electrical appliance in the total power data to be decomposed according to the following formula
In the above formula, xmaxIs the maximum value of the power, x, of the kth electrical appliance in the total power data to be decomposedminIs the power minimum value, x, of the kth electrical appliance in the total power data to be decomposedpredAnd obtaining a power normalization value of the kth electrical appliance in the total power data to be decomposed, wherein K belongs to K, and K is the total number of the electrical appliances in the total power data to be decomposed.
6. A non-intrusive load splitting system, the system comprising:
a processing module: the device is used for carrying out normalization processing on total power data to be decomposed;
a decomposition module: the system comprises a decomposition network, a power normalization value acquisition unit, a power normalization value generation unit and a power normalization value generation unit, wherein the decomposition network is used for decomposing the total power data to be decomposed after normalization processing based on a pre-trained decomposition network to obtain the power normalization value of each electric appliance in the total power data to be decomposed;
an acquisition module: and the device is used for performing inverse normalization processing on the power normalization values of various electrical appliances in the total power data to be decomposed to obtain the power values of various electrical appliances in the total power data to be decomposed.
7. The system of claim 6, wherein the processing module is specifically configured to:
determining the normalized value x of the ith total power value in the total power data to be decomposed according to the following formulai *：
In the above formula, xmaxIs the maximum value of power, x, in the total power data to be decomposedminIs the minimum value of power, x, in the total power data to be decomposediAnd the number is the ith total power value in the total power data to be decomposed, I belongs to I, and I is the total number of the total power values in the total power data to be decomposed.
8. The system of claim 6, wherein the acquisition process of the pre-trained decomposition network comprises:
normalizing the historical total power data and the power values of various electrical appliances in the historical total power data;
taking the normalized value of the historical total power data as input layer training data of an initial residual error gated circulation unit network, taking the normalized value of the power value of each type of electrical appliances in the historical total power data as output layer training data of the initial residual error gated circulation unit network, and training the initial residual error gated circulation unit network by adopting a gradient descent method;
and the data length of the historical total power data is the same as that of the total power data to be decomposed.
9. The system of claim 6, wherein the decomposition module is specifically configured to:
and inputting the normalized total power data to be decomposed into a pre-trained decomposition network, and acquiring power normalization values of various electrical appliances in the total power data to be decomposed output by the pre-trained decomposition network.
10. The system of claim 6, wherein the acquisition module is specifically configured to:
determining the power value of the kth type electrical appliance in the total power data to be decomposed according to the following formula
In the above formula, xmaxIs the maximum value of the power, x, of the kth electrical appliance in the total power data to be decomposedminIs the power minimum value, x, of the kth electrical appliance in the total power data to be decomposedpredAnd obtaining a power normalization value of the kth electrical appliance in the total power data to be decomposed, wherein K belongs to K, and K is the total number of the electrical appliances in the total power data to be decomposed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011081904.3A CN112434783A (en) | 2020-10-12 | 2020-10-12 | Non-invasive load decomposition method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011081904.3A CN112434783A (en) | 2020-10-12 | 2020-10-12 | Non-invasive load decomposition method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112434783A true CN112434783A (en) | 2021-03-02 |
Family
ID=74690555
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011081904.3A Pending CN112434783A (en) | 2020-10-12 | 2020-10-12 | Non-invasive load decomposition method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112434783A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113408210A (en) * | 2021-07-12 | 2021-09-17 | 内蒙古电力（集团）有限责任公司乌兰察布电业局 | Deep learning based non-intrusive load splitting method, system, medium, and apparatus |
| CN113837894A (en) * | 2021-08-06 | 2021-12-24 | 国网江苏省电力有限公司南京供电分公司 | Non-invasive resident user load decomposition method based on residual convolution module |
-
2020
- 2020-10-12 CN CN202011081904.3A patent/CN112434783A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113408210A (en) * | 2021-07-12 | 2021-09-17 | 内蒙古电力（集团）有限责任公司乌兰察布电业局 | Deep learning based non-intrusive load splitting method, system, medium, and apparatus |
| CN113837894A (en) * | 2021-08-06 | 2021-12-24 | 国网江苏省电力有限公司南京供电分公司 | Non-invasive resident user load decomposition method based on residual convolution module |
| CN113837894B (en) * | 2021-08-06 | 2023-12-19 | 国网江苏省电力有限公司南京供电分公司 | Non-invasive resident user load decomposition method based on residual convolution module |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109685314B (en) | Non-intrusive load decomposition method and system based on long-term and short-term memory network | |
| CN108616120B (en) | Non-invasive power load decomposition method based on RBF neural network | |
| CN109598451B (en) | Non-invasive load identification method based on PCA (principal component analysis) and LSTM (least Square TM) neural network | |
| CN109978079A (en) | A kind of data cleaning method of improved storehouse noise reduction self-encoding encoder | |
| CN110991263B (en) | Non-invasive load identification method and system for resisting background load interference | |
| CN112434783A (en) | Non-invasive load decomposition method and system | |
| Mao et al. | Anomaly detection for power consumption data based on isolated forest | |
| CN113408341B (en) | Load identification method and device, computer equipment and storage medium | |
| CN113902104A (en) | Non-invasive load monitoring method combining unsupervised domain self-adaptive strategy and attention mechanism | |
| CN111563827A (en) | Load decomposition method based on electrical appliance physical characteristics and residential electricity consumption behaviors | |
| CN116401532B (en) | Method and system for recognizing frequency instability of power system after disturbance | |
| CN108197425A (en) | A kind of intelligent grid data resolving method based on Non-negative Matrix Factorization | |
| CN111275069B (en) | Non-invasive load monitoring method | |
| CN114444539A (en) | Power load identification method, apparatus, device, medium, and program product | |
| CN112529053A (en) | Short-term prediction method and system for time sequence data in server | |
| CN116245019A (en) | Load prediction method, system, device and storage medium based on Bagging sampling and improved random forest algorithm | |
| CN116597635B (en) | Wireless communication intelligent gas meter controller and control method thereof | |
| Schirmer et al. | Improving energy disaggregation performance using appliance-driven sampling rates | |
| CN111090679A (en) | Time sequence data representation learning method based on time sequence influence and graph embedding | |
| CN116467631A (en) | Power fingerprint identification model training method, power fingerprint identification method and device | |
| Huang et al. | An online non-intrusive load monitoring method based on Hidden Markov model | |
| Li et al. | Non-intrusive Load Monitoring Method Based on Transfer Learning and Sequence-to-point Model | |
| Yang et al. | Smart grid data analysis and prediction modeling | |
| CN117559450B (en) | Method for improving non-intrusive load monitoring precision based on feedback model | |
| CN109521296A (en) | A kind of non-intrusion type electrical load under steady state condition identifies optimization algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |