CN113901726B - Non-invasive load decomposition method based on seq2point model - Google Patents
Non-invasive load decomposition method based on seq2point model Download PDFInfo
- Publication number
- CN113901726B CN113901726B CN202111251467.XA CN202111251467A CN113901726B CN 113901726 B CN113901726 B CN 113901726B CN 202111251467 A CN202111251467 A CN 202111251467A CN 113901726 B CN113901726 B CN 113901726B
- Authority
- CN
- China
- Prior art keywords
- seq2point
- model
- data
- electrical equipment
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 18
- 238000012952 Resampling Methods 0.000 claims abstract description 14
- 238000005070 sampling Methods 0.000 claims abstract description 8
- 230000002159 abnormal effect Effects 0.000 claims abstract description 6
- 239000012634 fragment Substances 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 34
- 210000002569 neuron Anatomy 0.000 claims description 25
- 238000012360 testing method Methods 0.000 claims description 9
- 238000005406 washing Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 4
- 230000002457 bidirectional effect Effects 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000005315 distribution function Methods 0.000 claims description 2
- 238000011478 gradient descent method Methods 0.000 claims description 2
- 210000002364 input neuron Anatomy 0.000 claims description 2
- 230000007246 mechanism Effects 0.000 claims description 2
- 210000004205 output neuron Anatomy 0.000 claims description 2
- 230000006870 function Effects 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000003872 anastomosis Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/06—Power analysis or power optimisation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Supply And Distribution Of Alternating Current (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a non-invasive load decomposition method based on a seq2point model, which comprises the following steps of S1, acquiring the total power of an ammeter corresponding to target electrical equipment to be detected, and resampling the total power by adopting a preset sampling frequency to obtain resampled data; s2, comparing the resampled data with the total power standard data of the ammeter, and deleting abnormal fragments in the resampled data to obtain input data of the network; s3, calculating the optimal sequence length of the input data according to the initial length of the input data; and S4, identifying the input data adjusted to the optimal sequence length by adopting the trained seq2point model, and obtaining a power curve of the target electrical equipment to be tested.
Description
Technical Field
The invention relates to a power data decomposition technology, in particular to a non-invasive load decomposition method based on a seq2point model.
Background
In recent years, intelligent electric meters have been widely deployed around the world, collect, store and manage massive fine-grained power consumption data, provide large data conditions for deep learning, provide end-to-end processing schemes for data processing, and enable load decomposition to be directly performed without detecting electrical switching events for non-invasive load monitoring tasks.
The existing seq2point model (sequence-to-point model, refer to paper C.Zhang,M.Zhong,Z.Wang,N.Goddard,and C.Sutton,"Sequence-to-point learning with neural networks for non-intrusive load monitoring,"in Proc.32nd AAAI Conf.Artif.Intell.(AAAI),2018,pp.1–8.), is widely used and can obtain better decomposition performance on non-invasive load detection tasks, but has poorer identification and prediction performance on long-time multi-state electric appliances, and has important importance in improving the network structure and eliminating the overfitting because of the overfitting problem of the identification and prediction performance on other devices.
Disclosure of Invention
Aiming at the defects in the prior art, the non-invasive load decomposition method based on the seq2point model solves the problem that the existing seq2point model is poor in prediction performance.
In order to achieve the aim of the invention, the invention adopts the following technical scheme:
There is provided a non-invasive load decomposition method based on a seq2point model, comprising the steps of:
s1, acquiring the total power of an ammeter corresponding to target electrical equipment to be detected, and resampling the total power by adopting a preset sampling frequency to obtain resampled data;
S2, comparing the resampled data with the total power standard data of the ammeter, and deleting abnormal fragments in the resampled data to obtain input data of the network;
S3, calculating the optimal sequence length of the input data according to the initial length of the input data:
ESL＝SL×SI×RF
wherein SL is the initial length of the input data; SI is the employed interval; RF is the resampling rate; ESL is the optimal sequence length;
s4, identifying the input data adjusted to the optimal sequence length by adopting a trained seq2point model to obtain a power curve of the target electrical equipment to be tested:
xτ＝Fp(Yt：t+W-1)+ε
Wherein x τ is the midpoint of the current device output window; y t：t+W-1 is a sliding window segment of the input data; x t：t+W-1 is a sliding window sequence of the target electrical equipment to be detected;
the network structure of the seq2point model comprises five convolution layers which are sequentially connected, wherein the output of the last convolution layer is sequentially connected with the full connection layer and the output layer, and discontinue one's studies layers are inserted between adjacent convolution layers and between the last convolution layer and the full connection layer.
The beneficial effects of the invention are as follows:
1. According to the scheme, the optimal sequence can be selected according to the power characteristics of each target electrical equipment, the sampling rate and the resampling rate are combined, the optimal sequence length of each equipment is obtained, the input window sequence length of the network model is corrected, the self-adaptive adjustment of the network model input is completed, the network model can learn the complete and accurate characteristics of the target electrical equipment, and the full preparation is made for subsequent load decomposition.
2. Elimination of overfitting: the dropout layer is added after the convolution layer, so that the characteristic quantity of the middle layer can be reduced, redundancy is reduced, orthogonality among all layers is better fitted, identification accuracy of target electrical equipment is further improved, and accuracy of more than 85% is achieved in predicting power curve change of the target electrical equipment.
3. By adopting the scheme to carry out load decomposition, the identification and prediction performance of the long-time multi-state equipment are improved, the close anastomosis is carried out on the power curve, and the defect that the existing seq2point model cannot identify and predict the power continuously variable equipment is overcome.
Drawings
FIG. 1 is a flow chart of a non-intrusive load decomposition method based on the seq2point model.
Fig. 2 is a comparison chart of predicted power curves and real power curves of refrigerators in houses 1,2, and 3, (a) comparison chart of house 1, (b) comparison chart of house 2, and (c) comparison chart of house 3 in the examples.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and all the inventions which make use of the inventive concept are protected by the spirit and scope of the present invention as defined and defined in the appended claims to those skilled in the art.
Referring to FIG. 1, FIG. 1 shows a flow chart of a non-intrusive load decomposition method based on a seq2point model; as shown in fig. 1, the method includes steps S1 to S4.
In step S1, acquiring the total power of an ammeter corresponding to the target electrical equipment to be tested, and resampling the total power by adopting a preset sampling frequency to obtain resampled data; the scheme is preferable that the preset sampling frequency is 8 s/time, and granularity uniformity can be realized after resampling, so that the wave bands of data of different electric meters and the same target electrical equipment under different electric meters are consistent.
The target electrical equipment comprises a kettle, a microwave oven, a refrigerator, a dish washer and a washing machine, and is of the same type when training is carried out.
In step S2, comparing the resampling data with the total power standard data of the ammeter, and deleting abnormal fragments in the resampling data to obtain input data of the network;
in step S3, the optimal sequence length of the input data is calculated according to the initial length of the input data:
ESL＝SL×SI×RF
wherein SL is the initial length of the input data; SI is the employed interval; RF is the resampling rate; ESL is the optimal sequence length;
The load decomposition method of the scheme further comprises filling a preset number of zeros in the head and tail of input data before inputting the input data with the optimal sequence length into the seq2point model.
In step S4, the trained seq2point model is adopted to identify the input data adjusted to the optimal sequence length, so as to obtain the power curve of the target electrical equipment to be tested:
xτ＝Fp(Yt：t+W-1)+ε
Wherein x τ is the midpoint of the current device output window; y t：t+W-1 is a sliding window segment of the input data; x t：t+W-1 is a sliding window sequence of the target electrical equipment to be detected;
The network structure of the seq2point model comprises five convolution layers which are sequentially connected, wherein the output of the last convolution layer is sequentially connected with a full connection layer and an output layer, and discontinue one's studies layers are inserted between adjacent convolution layers and between the last convolution layer and the full connection layer.
According to the scheme, the input window size of each device is adjusted by utilizing the optimal sequence, so that the power section of each device is completely input, and the neural network is beneficial to completely learning the characteristics of the target electrical equipment.
Meanwhile, the dropout layer is utilized to change the network construction of the original seq2point model, so that the neurons in the network layer can be forced to give up the connection of specific neurons, and the connection with more other neurons can be searched, so that more common characteristics can be learned or the weight of certain characteristics can be reduced, and the robustness of the network can be improved; and the joint adaptability between the network layers is enhanced, so that the generalization capability is enhanced.
In addition, the input of the seq2point model of the scheme is a sliding window segment of total load (total power), the output is a single point of the target electrical equipment, and the representation capability of the network is concentrated at the middle point of the window, so that more accurate prediction is generated.
In implementation, the loss function of the seq2point model is preferably:
Wherein θ p is a network parameter of the seq2point model; t is the sequence length; w is the window of the sliding window; log p () is the log of the features.
In one embodiment of the present invention, the training method of the seq2point model includes:
S41, acquiring a data set formed by the total power of a plurality of electric meters connected with target electrical equipment and the power data of the corresponding target electrical equipment;
The data set used in the scheme is REDD, UK-Dale, and specific information is as follows:
UK-Dale dataset: the data set contains measurements of more than 10 types of appliances, the present solution is only of interest for kettles, microwave ovens, refrigerators, dishwashers and washing machines, which are popular appliances for evaluating non-invasive power load monitoring methods, implemented using room 1, 3, 4, 5 training neural networks, room 2 as test data, since only rooms 1 and 2 have these devices.
REDD dataset: the device and power readings were recorded every 3 seconds and 1 second, respectively. The data set contains measurements from six households, the present protocol uses house nos. 2 to 6 for training, house No. 1 for algorithmic testing, and only microwave ovens, refrigerators, dishwashers and washing machines are utilized as there is no kettle data.
When the scheme is used for realizing load decomposition, if the power data of the refrigerator in the house is required to be monitored, when the seq2point model is trained, the data set only needs the power data of the refrigerator and the total power of the corresponding ammeter, and if the power data of the refrigerator and the washing machine of two target electrical equipment are required to be monitored, the seq2point model for identifying the refrigerator and the seq2point model for identifying the washing machine are required to be trained respectively during training.
In this scheme, five kinds of typical devices are used for training and testing, and the device names and relevant important attribute values are listed as follows:
microwave: power threshold = 200, maximum power = 3969, power mean = 500, power variance = 800;
Fridge: power threshold=50, maximum power=3323, power average=200, power variance=400;
DISHWASHER: power threshold = 10, maximum power = 3964, power mean = 700, power variance = 1000;
WASHINGMACHINE: power threshold = 20, maximum power = 3999, power mean = 400, power variance = 700;
Kettle: power threshold=2000, maximum power=3998, power average=700, power variance=1000.
S42, resampling the data in the data set by adopting a preset sampling frequency, comparing the resampled data with standard data, and deleting abnormal fragments in the resampled data to obtain an updated data set;
S43, respectively calculating the optimal sequence length of each data according to the initial length of each data in the updated data set to obtain a training set with the optimal sequence length of each data;
S44, training the seq2point model by adopting a training set, and then testing the trained seq2point model by adopting a testing set formed by power data in an ammeter connected with target electrical equipment to obtain a power curve of each target electrical equipment;
In practice, the training of the seq2point model using the training set preferably further comprises:
s441, randomly deleting half of hidden neurons in the seq2point model, wherein the input and output neurons remain unchanged;
s442, inputting data of a preset proportion in the training set, deleting half of network forward propagation of hidden neurons, and carrying out backward propagation on the obtained loss result; updating network parameters on the reserved neurons by adopting a random gradient descent method;
S443, recovering the deleted hidden neurons, and judging whether the network parameter loss function is smaller than a preset error or whether the iteration number is larger than a preset iteration number; if any condition is satisfied, training of the seq2point model is completed, otherwise, the process returns to step S441.
The calculation formula of the network in the training process of the seq2point model is as follows:
wherein r j (l) is a Bernoulli distribution value formed according to the probability p; bernoulli (-) is a Bernoulli distribution function; p is probability; for neuron outputs via random discard; y (l) is the raw output of the neuron; /(I) The output at layer l+1 for the ith neuron; /(I)Weights at layer l+1 for the ith neuron in the convolutional layer; /(I)Bias at layer l+1 for the ith neuron; /(I)Is the output of the neuron; f () is the activation function of the neuron.
S45, comparing the power curve of each target electrical equipment with the original power curve of the target electrical equipment, and obtaining the identification accuracy of the seq2point model according to the comparison result;
S46, judging whether the recognition accuracy is greater than a preset accuracy, if so, finishing training of the seq2point model, otherwise, entering a step S47;
s47, correcting the trained seq2point model by adopting a cavity residual deep network of a bidirectional attention mechanism to obtain the trained seq2point model.
Examples
Taking REDD datasets as an example, the device fridge is identified, and the relevant attributes and datasets of fridge are divided as follows:
'fridge': { 'input window size': 599, 'power threshold': 50, 'maximum power': 3323, 'power average': 200, 'power variance': 400, 'sequence length': 512, 'training house number': [1,2,3], 'power channel': [5,9,7]}
Extracting data of three channels of channel 5 of house 1, channel 9 of house 2 and channel 7 of house 3 in Redd, unifying granularity according to the resampling rate of 8 s/time, and obtaining the optimal sequence length (ESL).
The extracted data is divided into a training set, a verification set and three test sets.
By adopting the scheme, a seq2point model is constructed, and parameters in the model are set as follows: training batch size=1000, training sample size=100000, training round number=5, input window size=599, verification frequency=1, output window size=1, loss function= 'mean squared error', index= 'mse, msle, mae', learning rate=0.001.
After the related data set and the model parameters are completed, the training set and the verification set are simultaneously input into the network model for starting training, and the related model and the parameter information of fridge are generated according to the training method of the scheme, for example, MSE (mean square error) is 0.0926, MAE (mean aggregate error) is 0.2754, the weight of a convolution layer is 37200, and the weight of a full connection layer is 30669824.
The mean squared error (MAE) is calculated as:
Wherein x t is a predicted value, Is a true value.
The Mean Square Error (MSE) is calculated as:
wherein r represents a true value, Representing inferred values,/>
In the test process, a trained seq2point model about fridge is called, the total power of the house 1, 2 and 3 is input, the total power is decomposed, and further a predicted power curve of fridge in the house is obtained, the predicted power curve is compared with a real power curve, and the comparison result is shown in fig. 2.
As can be seen from the comparison of FIG. 2, the type of the device can be better identified by using the seq2point model of the optimal sequence, the change of the power curve can be better predicted, the power section of the power jump of the device is removed, and the accuracy rate can reach more than 85%.
Claims (3)
1. A non-invasive load decomposition method based on a seq2point model, comprising the steps of:
s1, acquiring the total power of an ammeter corresponding to target electrical equipment to be detected, and resampling the total power by adopting a preset sampling frequency to obtain resampled data;
S2, comparing the resampled data with the total power standard data of the ammeter, and deleting abnormal fragments in the resampled data to obtain input data of the network;
S3, calculating the optimal sequence length of the input data according to the initial length of the input data:
wherein SL is the initial length of the input data; SI is the employed interval; RF is the resampling rate; ESL is the optimal sequence length;
s4, identifying the input data adjusted to the optimal sequence length by adopting a trained seq2point model to obtain a power curve of the target electrical equipment to be tested:
Wherein, Outputting a midpoint of a window for the current device; /(I)A sliding window segment for input data;
the network structure of the seq2point model comprises five convolution layers which are sequentially connected, wherein the output of the last convolution layer is sequentially connected with a full connection layer and an output layer, and Dropout layers are inserted between adjacent convolution layers and between the last convolution layer and the full connection layer;
The training method of the seq2point model comprises the following steps:
S41, acquiring a data set formed by the total power of a plurality of electric meters connected with target electrical equipment and the power data of the corresponding target electrical equipment;
S42, resampling the data in the data set by adopting a preset sampling frequency, comparing the resampled data with standard data, and deleting abnormal fragments in the resampled data to obtain an updated data set;
S43, respectively calculating the optimal sequence length of each data according to the initial length of each data in the updated data set to obtain a training set with the optimal sequence length of each data;
S44, training the seq2point model by adopting a training set, and then testing the trained seq2point model by adopting a testing set formed by power data in an ammeter connected with target electrical equipment to obtain a power curve of each target electrical equipment;
S45, comparing the power curve of each target electrical equipment with the original power curve of the target electrical equipment, and obtaining the identification accuracy of the seq2point model according to the comparison result;
s46, judging whether the recognition accuracy is greater than a preset accuracy, if so, completing training of the seq2point model, otherwise, entering a step S47;
S47, correcting the trained seq2point model by adopting a cavity residual deep network of a bidirectional attention mechanism to obtain a trained seq2point model;
the training of the seq2point model using the training set further comprises:
s441, randomly deleting half of hidden neurons in the seq2point model, wherein the input and output neurons remain unchanged;
S442, inputting data of a preset proportion in a training set, deleting half of network forward propagation of hidden neurons, carrying out reverse propagation on the obtained loss result, and then updating network parameters on the reserved neurons by adopting a random gradient descent method;
S443, recovering the deleted hidden neurons, and judging whether the network parameter loss function is smaller than a preset error or whether the iteration number is larger than a preset iteration number; if any condition is satisfied, training of the seq2point model is completed, otherwise, returning to step S441;
The calculation formula of the network in the seq2point model training process is as follows:
Wherein, To be according to probability/>A structured bernoulli distribution value; /(I)Is a Bernoulli distribution function; p is probability; /(I)For neuron outputs via random discard; /(I)Is the raw output of the neuron; /(I)The output at layer l+1 for the ith neuron; /(I)Weights at layer l+1 for the ith neuron in the convolutional layer; /(I)Bias at layer l+1 for the ith neuron; /(I)Is the output of the current neuron; f () is the output function of the neuron;
Inputting the input data of the optimal sequence length into the seq2point model further comprises filling a preset number of zeros at the head and tail of the input data.
2. The non-invasive load decomposition method according to claim 1, wherein the loss function of the seq2point model is:
Wherein, Network parameters that are the seq2point model; t is the sequence length; w is the window of the sliding window; log p () is the log of the features.
3. The non-invasive load decomposition method based on the seq2point model according to claim 1 or 2, wherein the target electrical equipment comprises kettles, microwave ovens, refrigerators, dish washers and washing machines, and the target electrical equipment is of the same type when training is performed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111251467.XA CN113901726B (en) | 2021-10-27 | 2021-10-27 | Non-invasive load decomposition method based on seq2point model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111251467.XA CN113901726B (en) | 2021-10-27 | 2021-10-27 | Non-invasive load decomposition method based on seq2point model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113901726A CN113901726A (en) | 2022-01-07 |
| CN113901726B true CN113901726B (en) | 2024-06-21 |
Family
ID=79026886
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111251467.XA Active CN113901726B (en) | 2021-10-27 | 2021-10-27 | Non-invasive load decomposition method based on seq2point model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113901726B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115330553B (en) * | 2022-06-22 | 2023-04-11 | 四川大学 | Non-invasive load decomposition method based on equipment characteristic multi-layer optimization |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110852655A (en) * | 2019-11-29 | 2020-02-28 | 湖南科技大学 | Household power load decomposition system with solar power supply system and decomposition method |
| CN111199014A (en) * | 2019-12-27 | 2020-05-26 | 国家电网有限公司 | Time sequence based seq2point NILM method and device |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111753968B (en) * | 2020-07-07 | 2023-08-08 | 苏州联电能源发展有限公司 | Non-invasive load monitoring intelligent ammeter and electric quantity decomposition method |
| CN112215406B (en) * | 2020-09-23 | 2024-04-16 | 国网甘肃省电力公司电力科学研究院 | Non-invasive resident electricity load decomposition method based on time convolution neural network |
| US10983496B1 (en) * | 2020-10-13 | 2021-04-20 | King Abdulaziz University | Apparatus and method for non-intrusive load monitoring aided hybrid high and low frequency approaches |
-
2021
- 2021-10-27 CN CN202111251467.XA patent/CN113901726B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110852655A (en) * | 2019-11-29 | 2020-02-28 | 湖南科技大学 | Household power load decomposition system with solar power supply system and decomposition method |
| CN111199014A (en) * | 2019-12-27 | 2020-05-26 | 国家电网有限公司 | Time sequence based seq2point NILM method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113901726A (en) | 2022-01-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112215406B (en) | Non-invasive resident electricity load decomposition method based on time convolution neural network | |
| CN110087207B (en) | Method for reconstructing missing data of wireless sensor network | |
| CN112787736B (en) | Long-short term memory cooperative spectrum sensing method based on covariance matrix | |
| CN113901726B (en) | Non-invasive load decomposition method based on seq2point model | |
| CN108924847B (en) | Cognitive radio frequency spectrum prediction method and device based on ANN | |
| CN114723040A (en) | LSTM and sequence-to-point translation model-based non-invasive load monitoring method | |
| CN111639586A (en) | Non-invasive load identification model construction method, load identification method and system | |
| CN112307927A (en) | BP network-based identification research for MPSK signals in non-cooperative communication | |
| CN117076955A (en) | Fault detection method and system for high-voltage frequency converter | |
| CN118068195A (en) | Method and system for establishing battery health state prediction model | |
| Steinert et al. | Long-term adaptation and distributed detection of local network changes | |
| CN114004258A (en) | Semi-supervised electrocardio abnormality detection method | |
| CN106100870A (en) | A kind of community network event detecting method based on link prediction | |
| CN116361709B (en) | Self-adaptive power load identification method and device | |
| Herath et al. | Comprehensive analysis of convolutional neural network models for non-instructive load monitoring | |
| CN116821619A (en) | Time sequence anomaly detection method based on multi-element time sequence relation learning | |
| CN116742795A (en) | Non-invasive load monitoring method based on transfer learning and self-attention feature fusion | |
| CN114510992A (en) | Equipment switch state detection method based on deep learning | |
| CN115859190A (en) | Non-invasive household electrical classification method based on causal relationship | |
| CN116008747A (en) | Yogi-mLSTM cable partial discharge identification method and diagnosis system based on wavelet threshold denoising | |
| CN115422603A (en) | Block cipher hardware security assessment method based on template principal component regression | |
| CN115330553A (en) | Equipment feature based multi-layer optimal selection non-intrusive load decomposition method | |
| CN113920362A (en) | Non-invasive load decomposition method based on attention | |
| CN113687182A (en) | Household load identification method, program and system based on noise reduction automatic encoder | |
| Joseph et al. | A scalable algorithm for anomaly detection via learning-based controlled sensing |
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 |