CN111738521A - Non-invasive power load monitoring sequence generation method, system, equipment and medium - Google Patents
Non-invasive power load monitoring sequence generation method, system, equipment and medium Download PDFInfo
- Publication number
- CN111738521A CN111738521A CN202010588708.9A CN202010588708A CN111738521A CN 111738521 A CN111738521 A CN 111738521A CN 202010588708 A CN202010588708 A CN 202010588708A CN 111738521 A CN111738521 A CN 111738521A
- Authority
- CN
- China
- Prior art keywords
- data sequence
- sequence
- data
- network
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000012544 monitoring process Methods 0.000 title claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 76
- 238000005070 sampling Methods 0.000 claims description 52
- 230000005611 electricity Effects 0.000 claims description 32
- 230000008569 process Effects 0.000 claims description 23
- 238000012545 processing Methods 0.000 claims description 18
- 230000002159 abnormal effect Effects 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 10
- 238000003860 storage Methods 0.000 claims description 7
- 238000013461 design Methods 0.000 description 8
- 238000012217 deletion Methods 0.000 description 7
- 230000037430 deletion Effects 0.000 description 7
- 230000006870 function Effects 0.000 description 6
- 238000007781 pre-processing Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 5
- 238000009826 distribution Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 5
- 238000004891 communication Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000002864 sequence alignment Methods 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000000819 phase cycle Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 239000004984 smart glass Substances 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
- G06N3/08—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)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- General Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Tourism & Hospitality (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Primary Health Care (AREA)
- Development Economics (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Remote Monitoring And Control Of Power-Distribution Networks (AREA)
Abstract
The present disclosure discloses a method, a system, a device and a medium for generating a non-intrusive power load monitoring sequence, which includes: acquiring a total data sequence of the electric meter; inputting the total data sequence of the electric meter into a pre-trained conditional countermeasure generating network, and outputting the data sequence of the target electric appliance by the pre-trained conditional countermeasure generating network; wherein the conditional countermeasure generating network comprises a generating network and a discriminating network; the generating network is used for generating a synthetic data sequence similar to the real data sequence of the target electrical appliance, and the judging network is used for judging whether the synthetic data sequence generated by the generating network is the real data sequence of the target electrical appliance in the training phase of the conditional countermeasure generating network.
Description
Technical Field
The disclosure belongs to the field of power load prediction, and relates to a non-intrusive power load monitoring sequence generation method, system, device and medium.
Background
The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.
In the course of implementing the present disclosure, the inventors found that the following technical problems exist in the prior art:
in the field of power load prediction, non-intrusive power load monitoring is an important issue. Non-invasive power load monitoring, also known as power data decomposition, is a technique for estimating the power consumption of individual consumers from the total power consumption monitored by an electric meter, and is limited to non-invasive methods because sensors do not need to be installed on the electrical devices. In the household power consumption, there is a general ammeter usually, if can derive the power consumption condition of independent electrical apparatus from the monitoring data of general ammeter, then can help the user to know the behavior of electrical apparatus, in time discover the unusual electrical apparatus of operating condition, help its saving electric energy. If the comprehensive control method can be integrated on a larger regional dimension, a dispatcher can be helped to better carry out power grid dispatching and regulation. Fig. 1 shows total data collected by an electricity meter and electricity consumption data of a certain electrical appliance collected at the same time, wherein the abscissa represents a time stamp, and the ordinate represents electric power.
The research of NILM started in the 80 th 20 th century, and the main method is to analyze and extract the electricity utilization characteristics in an artificial manner according to the electricity utilization condition of an electric appliance so as to construct a data decomposition algorithm. The manual feature extraction is used, and the problems that the manual cost is high, the accuracy cannot be guaranteed and the like are faced. Due to the deep learning technology and the improvement of the computing power of the computer, the automatic feature extraction and data decomposition by using the computer become possible. Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Auto encoders (Auto encoders), and other technologies are used in NILM, and although certain results are achieved, they are still unsatisfactory in terms of quality of sequence generation.
Disclosure of Invention
In order to solve the defects of the prior art, the present disclosure provides a method, a system, a device and a medium for generating a Non-invasive power Load Monitoring sequence, which design a Non-invasive power Load Monitoring algorithm based on a CGAN by using Non-invasive Load Monitoring (NILM) as an entry point and combining a Conditional countermeasure generation network (CGAN) technology with information generation capability, and complete a power Load Monitoring task quickly, efficiently and accurately.
In a first aspect, the present disclosure provides a method for generating a non-intrusive power load monitoring sequence;
a non-intrusive power load monitoring sequence generation method comprises the following steps:
acquiring a total data sequence of the electric meter;
inputting the total data sequence of the electric meter into a pre-trained conditional countermeasure generating network, and outputting the data sequence of the target electric appliance by the pre-trained conditional countermeasure generating network; wherein the conditional countermeasure generating network comprises a generating network and a discriminating network; the generating network is used for generating a synthetic data sequence similar to the real data sequence of the target electrical appliance, and the judging network is used for judging whether the synthetic data sequence generated by the generating network is the real data sequence of the target electrical appliance in the training phase of the conditional countermeasure generating network.
In a second aspect, the present disclosure also provides a non-intrusive power load monitoring sequence generation system;
a non-intrusive power load monitoring sequence generation system, comprising:
an acquisition module configured to acquire a total data sequence of the electricity meter;
the output module is configured to input the total data sequence of the electric meter into a pre-trained conditional countermeasure generation network, and the pre-trained conditional countermeasure generation network outputs the data sequence of the target electric appliance; wherein the conditional countermeasure generating network comprises a generating network and a discriminating network; the generating network is used for generating a synthetic data sequence similar to the real data sequence of the target electrical appliance, and the judging network is used for judging whether the synthetic data sequence generated by the generating network is the real data sequence of the target electrical appliance in the training phase of the conditional countermeasure generating network.
In a third aspect, the present disclosure also provides an electronic device comprising a memory and a processor, and computer instructions stored on the memory and executed on the processor, wherein the computer instructions, when executed by the processor, perform the steps of the method of the first aspect.
In a fourth aspect, the present disclosure also provides a computer-readable storage medium for storing computer instructions which, when executed by a processor, perform the steps of the method of the first aspect.
Compared with the prior art, the beneficial effect of this disclosure is:
the invention takes a conditional countermeasure generation network as a means to tightly combine a deep learning technology into a non-intrusive power load monitoring subject. The method adopted by the invention does not need to manually extract features in the training process, and can ensure that the trained network has strong and accurate generation capacity by only utilizing a smaller data set, thereby having better performance compared with the prior method.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application.
FIG. 1 is a flow chart of a method of the first embodiment;
FIG. 2(a) is a data sequence of a simultaneous segment appliance;
FIG. 2(b) is a data sequence of the electricity meter;
fig. 3(a) and fig. 3(b) are images with the same start time stamp and the same number of sampling points;
FIG. 4 is a generator basic structure;
FIG. 5 is a detailed structure of the generator;
FIG. 6 shows the basic structure of an encoder;
FIG. 7 is a decoder basic structure;
FIGS. 8(a) and 8(b) are basic structures of the discriminator;
FIG. 9 shows the detailed structure of the discriminator;
FIG. 10 is a discriminator training process;
FIG. 11 is a generator training process.
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
In a first embodiment, the present embodiment provides a method for generating a non-intrusive power load monitoring sequence;
a non-intrusive power load monitoring sequence generation method comprises the following steps:
acquiring a total data sequence of the electric meter;
inputting the total data sequence of the electric meter into a pre-trained conditional countermeasure generating network, and outputting the data sequence of the target electric appliance by the pre-trained conditional countermeasure generating network; wherein the conditional countermeasure generating network comprises a generating network and a discriminating network; the generating network is used for generating a synthetic data sequence similar to the real data sequence of the target electrical appliance, and the judging network is used for judging whether the synthetic data sequence generated by the generating network is the real data sequence of the target electrical appliance in the training phase of the conditional countermeasure generating network.
As one or more embodiments, a corresponding conditional countermeasure generation network is trained for each target appliance. And preparing a training set corresponding to each target electrical appliance for the training of each target electrical appliance.
As one or more embodiments, before the step of inputting the total data sequence of the electricity meters into the pre-trained conditional countermeasure generating network, the pre-trained conditional countermeasure generating network outputs the data sequence of the target electrical appliance, the method further includes:
selecting a training data set and a testing data set from a historical database of total data of the electric meters; training data sets including total electric meter data and a real data sequence of a certain target electric appliance;
inputting the total data of the electric meters in the training data set into a generation network of the conditional countermeasure generation network for training; judging whether the synthetic data sequence generated by the generating network is the real data sequence of the current target electrical appliance or not through the judging network;
if the judgment network judges that the synthetic data sequence generated by the generation network is not the real data sequence of the current target electrical appliance, the parameters of the generation network are adjusted and then training is continued;
if the judgment network judges that the synthetic data sequence generated by the generation network is the real data sequence of the current target electrical appliance, the training is finished;
testing the generated model through the test data set to obtain a corresponding test result; and if the test result meets the set requirement, obtaining a trained conditional countermeasure generation network.
As one or more embodiments, the preparing of the training data set comprises:
selecting a data set, wherein the data set comprises a total data sequence of electric meters and a plurality of electric appliance electricity utilization data sequences in the same time period; the electric meter total data sequence and the electricity utilization sequence both comprise: the method comprises the steps of numbering electrical appliances, collecting timestamps of power of the electrical appliances and power values of the electrical appliances corresponding to the timestamps;
performing missing value processing and abnormal value processing on the data in the selected data set;
eliminating noise data from the data sequence after processing the missing value and the abnormal value;
sequences with noise data eliminated are divided into two categories: a data sequence in an operating state and a data sequence in a non-operating state;
selecting a certain electric appliance, and selecting the power utilization data sequence of the electric appliance according to the name of the electric appliance, wherein the power utilization data sequence comprises a data sequence of the electric appliance in a working state and a data sequence of the electric appliance in a non-working state;
selecting a total data sequence of electric meters in the same power utilization time range according to the power utilization time range of the electric appliance; the total data sequence of the electric meters in the same electricity utilization time range comprises a data sequence in an operating state and a data sequence in a non-operating state of all the electric appliances in the electricity utilization range;
and aligning the time stamps of the electricity utilization data sequence of the electric appliance to the total electricity utilization data sequence of the same electricity utilization time range.
As one or more embodiments, the missing value processing step includes:
step (11): scanning the electric meter total data sequence or the electric power utilization sequence from front to back, calculating the difference value of every two adjacent time stamps, and judging whether each difference value is in a set range;
step (12): if the current time is within the set range, continuing to execute the step (11); if the difference is not in the set range, comparing the difference with the set threshold, and if the difference is greater than the set threshold, continuing to execute the step (11); if the value is smaller than the set threshold value, entering the step (13);
step (13): calculating the mean value of the values corresponding to the two timestamps which are currently processed, and using the changed mean value as the reconstruction value of the missing data;
reconstructing the timestamp corresponding to the missing value: the starting time stamp of the missing interval is marked as s, the ending time stamp is marked as e, the number of the reconstructed time stamps is
The number of the reconstructed time stamps is recorded as c, the reconstructed time stamps are added with q each time from the starting time stamp s, and the process is repeated for c times;
step (15) of inserting the reconstruction value generated in step (13) and the reconstruction timestamp generated in step (14) into the original sequence in order;
step (16) repeats step (11) until the entire sequence of scans is complete.
As one or more embodiments, the specific steps of outlier processing include:
21) scanning the total data sequence or the power utilization sequence of the electric meter according to the sequence from front to back, and executing the step 22) if the value of the current sampling point is higher than the threshold value t for the data points except the beginning and the ending, otherwise, continuing to scan the value of the next sampling point;
22) subtracting the numerical values of two adjacent sampling points from the numerical value of the current sampling point, and respectively recording the obtained difference values as b and f;
23) if both b and f are larger than the set threshold, indicating that the value of the current sampling point is an abnormal value, and performing step 24); otherwise, step 21) is executed;
24) and using the average value of two points before and after the current sampling point as a reconstruction value, replacing the value of the current sampling point with the reconstruction value, and performing the step 21) until the sequence traversal is finished.
In the field of information generation, the generation-resistant neural Networks (GANs) is a novel and effective model, and is widely applied in the fields of image generation, sequence generation, machine translation and the like, and obtains a satisfactory effect, and the attention is paid more and more. The model contains two competing networks: generating a network-G (Generator), and learning the distribution condition of training set data; discriminant network-d (discriminator), the discriminant data being from the training set or probabilities generated by the generating network. Assuming the training set data is x, to learn the distribution p of xgG establishes a distribution p of well-defined noisez(z) mapping to production data, denoted G (z; θ)g) (ii) a As a discriminator, willUsing D (x; theta)d) To decide that a given data belongs to a training set x and a probability pgThe size of (2). P and G are trained simultaneously, according to the following formula:
if the training of the generator G and the discriminator D is restricted using certain conditions, the direction of the training and the effectiveness and interpretability of the generation can be controlled. If the given condition is denoted as y, the given condition needs to be added to G and D simultaneously during the training process, and the training process of G and D can be represented by the following formula:
for NILM, the data sequence corresponding to the electrical appliance obtained from the total data sequence of the electricity meter can be mapped as a sequence generation problem, and CGAN has obvious advantages in the field of sequence generation, and can obtain a high-quality generated sequence controlled by conditions.
The invention integrates strong generation capability of a conditional countermeasure generation network and provides a complete solution from data preprocessing to model design, training and application deployment. The invention takes a conditional countermeasure generation network as a means to tightly combine a deep learning technology into a non-intrusive power load monitoring subject. The method adopted by the invention does not need to manually extract features in the training process, and can ensure that the trained network has strong and accurate generation capacity by only utilizing a smaller data set, thereby having better performance compared with the prior method.
The innovation of the invention is realized by the following aspects:
(1) the method has the advantages that the NILM is innovatively mapped into the sequence generation problem, and the CGAN is used in a training model to greatly improve the quality of sequence generation;
(2) innovatively introducing a one-to-one training mode into a model training process;
(3) an overall solution from data set preprocessing to training and final deployment is proposed, making the solution more systematic and operational.
The solution proposed by the invention consists of three stages, namely a data preprocessing stage, a model training stage and an application deployment stage. The three stages will be described in detail below.
First, data preprocessing stage
1. Training data acquisition
In order to carry out academic research, relevant researchers collect and record total electricity utilization data and electricity utilization data of various types of electric appliances in a real electricity utilization environment in a mode of installing relevant sensors, arrange the data after long-time collection, and release the data in a public data set form for public study and research. To validate the feasibility of the present invention, we used the public data set widely used in academia for model training and validation.
The data set selected by the invention is UK-DALE (social application-Level electric and floor-house electric demand from UK homes) [1], the data set comprises the total electricity utilization condition of 5 houses and the electricity utilization condition of a plurality of electric appliances in the same time period, the sampling frequency is 1/6Hz, and the data is from 2012 to 2015. Fig. 2(a) and 2(b) are plotted against the data set.
[1]Kelly J,Knottenbelt W.The UK-DALE dataset,domestic appliance-levelelectricity demand and whole-house demand from five UK homes[J].Scientificdata,2015,2:150007.
2. Data pre-processing
Data in the UK-DALE dataset exist in a one-to-one correspondence mode of the time stamp and the power value at the corresponding moment, and a plurality of missing values and abnormal values exist, so that the data cannot be directly used for model training. Therefore, preprocessing work is required first before the model is trained using these data.
(1) Missing value handling
The sequences in the dataset are not always contiguous and may be missing. The missing data is relatively small in quantity and relatively short in duration, and is marked as type one missing in the invention; secondly, the electrical appliance is in a closed state, which results in data loss for a long time, and the data loss is marked as type II loss in the invention. For type one misses, the handling of data miss values is necessary. For the second, there is no need for recovery and no efficient method to recover because of the longer data loss, and therefore no processing is done.
Processing the miss value first requires defining the miss. In the present invention, a deletion is defined as two types, and defining the deletion requires determining a reasonable threshold t: if the difference value of the front timestamp and the rear timestamp in the sequence is greater than t, the two types of deletion is considered to occur; if the value is less than t, the type one deletion is considered to be generated, and if the type one deletion is generated, the deletion value needs to be recovered.
The process of determining the threshold t can be divided into the following steps:
1) scanning the sequence, calculating the difference value of every two adjacent time stamps in the sequence, and dividing the difference value by 6 to obtain a sampling point difference value sequence so as to correspond to the sampling point;
2) analyzing the difference sequence of the sampling points, observing the distribution condition of the difference, and selecting a boundary value to distinguish the first type loss from the second type loss;
3) in the part of the difference value smaller than the boundary value, the smallest upper difference value boundary is selected and the range of the difference value containing the most difference values is multiplied by 6 to obtain t.
The specific steps for recovering the missing value are as follows:
11) scanning the sequence from front to back, calculating the difference value of every two adjacent time stamps in the sequence, and checking whether the calculated difference value of each step is between 4 and 8 (considering the precision of data recording time and allowing slight fluctuation of sampling frequency);
12) if the difference is within the interval, proceeding to step 11), if the difference is not within the interval, comparing with t, if greater than t, it is a type II missing, proceeding to step 11); if the value is less than t, the type I deletion is performed, and the step 13) is performed;
13) calculating the mean value of the data corresponding to the two currently processed timestamps, and using the mean value as the reconstruction value of the missing data;
14) reconstructing the timestamp corresponding to the missing value: the starting time stamp of the missing interval is marked as s, the ending time stamp is marked as e, the number of the reconstructed time stamps is
Recording as c, adding 6 times from the starting time stamp s to the reconstructed time stamp, and repeating the steps for c times;
15) inserting the reconstruction value generated in the step 13) and the reconstruction timestamp generated in the step 14) into the original sequence in sequence;
16) repeat step 11) until the entire sequence scan is finished.
(2) Outlier processing
In addition to missing values, there are some outliers in the original sequence. As shown in fig. 2(a) and 2(b), in fig. 2(a) representing the electric appliance data, a value of one data point is relatively highlighted at a position of about 200 on the abscissa, whereas in fig. 2(b) corresponding to the total electric meter data, a value at the position in the electric appliance data series can be roughly judged as an abnormal value without such a high value appearing at the corresponding position. The abnormal value is usually caused by sensor failure or surge in the surrounding power environment, which causes the value of a certain sampling point to be abnormally high. For such outliers with distinct characteristics, they need to be handled.
The abnormal value processing method comprises the following specific steps:
21) scanning the sequence according to the sequence from the front to the back, and executing the step 22) if the value of the current sampling point is higher than the threshold value t for the data points except the beginning and the end, otherwise, continuing to scan the value of the next sampling point;
22) subtracting the numerical values of two adjacent sampling points from the numerical value of the current sampling point, and respectively recording the obtained difference values as b and f;
23) if b and f are both larger, it is indicated that the value of the current sampling point may be an abnormal value, and step 24) is performed; otherwise, step 21) is executed;
24) and using the average value of two points before and after the current sampling point as a reconstruction value, replacing the value of the current sampling point with the reconstruction value, and performing the step 21) until the sequence traversal is finished.
In the above abnormal value processing flow, the threshold t needs to be defined first, and considering the common expression form and the occurrence reason of the abnormal value, t is generally defined as a larger value, and the power of the electrical appliance in normal operation is generally used as a reference; in the present invention, this value is taken as 2000. In addition, the condition that "b and f are both large" is fuzzy, and is intuitively understood as the condition that the data of the current sampling point is higher than the data of the adjacent data points abnormally, and because the data of the current sampling point fluctuates to a certain degree in the actual use of the electric appliance, the condition needs to be carefully defined in order to avoid the algorithm interfering with the normal values of the sampling points. In the invention, if the values of b and f both exceed 1/2 corresponding to the values of the adjacent sampling points, the current sampling point value is considered to be an abnormal value and needs to be reconstructed.
3. Training set generation
After the original sequence is preprocessed, the problems of missing values and abnormal values are solved, and a training set needs to be generated in the next step. In order to enable the model to learn the relationship between the electric appliance and the total data of the electric meter in the working state and the non-working state, the relevant information of the two states needs to be added into a training set. It should be noted that, since the characteristics of each appliance are different, a model is trained for each appliance separately, and thus, when training the set configuration, it is also constructed for each appliance separately.
(1. extraction of data of electric appliance working phase
And generating training set data, firstly positioning the range of sampling points of the electric appliance in a working state, analyzing the range information to obtain a basic working model and working behavior of the electric appliance, and forming a training set with more complete information so as to better guide the training of the model.
For the sampling sequence of the electric appliance, the data of a part of the sampling points is 0, which indicates that the electric appliance is in a non-working state at the time node. The electric appliance is always maintained for a period of time in the working state, the numerical values corresponding to the period of time are not 0, whether the electric appliance is in the working state or not can be judged by utilizing the characteristics, and the corresponding timestamp and the corresponding numerical values in the working state are recorded.
The specific implementation steps are as follows:
31) for a certain electrical appliance sequence, scanning from front to back, if no non-0 numerical value appears, continuing scanning, and if non-0 data appears, performing step 32);
32) continuing to scan, if more than N0 data points continue to appear after more than M non-0 data points, regarding the data points as noise segments, and returning to the step 31) to continue to scan the residual sequence backwards;
if more non-0 data points continuously appear, the electric appliance is regarded as an effective working state, the sequence is continuously scanned backwards until a plurality of continuous 0 data points appear, the working state of the electric appliance is regarded as the end, and a segment between the initial non-0 data point and the last non-0 data point is recorded;
33) and (3) visualizing the recorded segments, observing the extraction effect, and if the extraction effect is not ideal, finely adjusting the specific parameters in the step 32), and then performing the step 31) to obtain a relatively optimal appliance working phase data sequence.
(2) Determining data window width
For model training, the input data must have the same size, and therefore, the appliance operating phase data extracted in 31) needs to be analyzed to determine a reasonable data window width. Analyzing the length of the sequence extracted in the step 31), and determining the number of sampling points to be included in the sequence according to the distribution condition of the sequence length, so that the sequence can include as many single and complete electric appliance working state sequences as possible, and the length of the sequence is made to be shorter as possible.
(3) Training set data construction
The appliance data sequence includes features of the appliance in the working state and the non-working state, and in order to enable the model to learn the features, the features are included in the constructed training set data. The training set data comprises two parts, namely electrical appliance data and ammeter data corresponding to time intervals. Firstly, determining an electric appliance data sequence to be acquired, and then intercepting electric meter data according to corresponding time periods, wherein the electric meter data and the electric meter data are in one-to-one correspondence.
The specific operation steps are as follows:
41) sequentially processing the obtained electric appliance working phase sequence, recording the current processing sequence as s, and recording the sequence length as | s |;
42) taking the central point of s as a reference, and taking 1.5| s | sampling points in front of the central point and 1.5| s | sampling points behind the central point, and recording as s';
43) taking the width of the data window determined in 42) as a reference, sampling s' from front to back, and adjusting the step length of each sampling, namely adjusting the number of generated sequences;
44) taking the starting time stamp of each sequence generated in the step 43) as a starting point, taking the width of the data window determined in the step 42) as a sampling width, intercepting the corresponding data sequence from the electric meter data sequence, naming and storing (for each pair of corresponding electric appliance and electric meter data sequence, the corresponding relation needs to be indicated and determined in terms of naming);
45) continuing to step 41) until the entire sequence has been processed.
(4) Timestamp alignment
According to the relevant document display of UK-DALE, the data sampling frequency is 1/6Hz, but in the actual data sampling process, the data sampling frequency obtained by the final sampling is not completely 1/6 under the influence of the numerical value rounding of the time stamp, the transmission delay of the sensor, the sampling precision of the sensor and the like, and the sampling frequency of each electric appliance is different from that of the total electric meter. The effect of this problem is best illustrated in fig. 3(a) and 3(b), where the timestamps of the two sequences start to be substantially the same, but there is a mismatch in the basic shapes of the sequences, indicating that the sampling intervals are not the same, and the two sequences need to be aligned according to the timestamps. The statistics shows that the sampling frequency of the electrical appliance sequence is slightly high, and in order to ensure the sequence alignment effect, the sequence with slightly high sampling frequency needs to be aligned to the sequence with slightly low sampling frequency.
The specific sequence alignment procedure is as follows:
51) scanning the timestamp data from front to back by taking the ammeter data sequence as a reference, and recording the currently processed timestamp as t;
52) searching two timestamps closest to t in the electric appliance data sequence, and calculating the average value of the two timestamps as a;
53) adding the pair (t, a) into an appliance data sequence;
54) and continuing to execute the step 51) until the electric meter data sequence is processed.
Second, model training phase
1. Model construction
The invention maps the non-intrusive power load monitoring problem to a sequence generation problem, namely, a data sequence of a certain electric appliance is generated from a data sequence of an electric meter. In consideration of the one-to-one correspondence relationship between the electric meter data and the electric appliance data, the invention introduces a one-to-one training mode into the training process of the model, simultaneously trains the model by using the CGAN Loss and the sequence L1 Loss as a Loss function, and designs the special model for solving the non-invasive power load monitoring problem by combining the strong information generation capability of the CGAN.
The model is introduced and divided into four parts, namely generator design, discriminator training process and generator training process. The AG and AP appearing hereinafter represent the data sequence of the electricity meter and the data sequence of the appliance at the same time interval, respectively, the pair of sequences being randomly selected from the training set, and AP being the generated sequence from the AG through the generator. For convenience of description, in the present invention, the sequence lengths of AG, AP, and AP are all 1024, and the sequence lengths and parameters in the model may be reasonably adjusted according to a specific application scenario.
(1) Generator design
The generator is used for obtaining an output AP from the input AG, namely, a data sequence of a corresponding electric appliance is predicted from a data sequence of an electric meter, and the basic structure of the generator is described in figure 4. Fig. 5 illustrates a detailed structure of the generator, in which AG is used as an input, and sequentially passes through 8 encoders to compress the sequence thereof, and then passes through 8 decoders to reconstruct the information, thereby generating a sequence AP, where the input and output dimensions of each calculation stage are labeled.
The encoder is used for compressing input information, and the basic structure of the encoder is as shown in FIG. 6. And the input sequence is subjected to convolution operation, regularization operation and a ReLU activation function to generate a final output sequence. The decoder acts in the opposite way to the encoder, its purpose is to reconstruct and recover the information, and the basic structure is as in fig. 7. And the input sequence is subjected to deconvolution operation, regularization operation and a ReLU activation function in sequence to obtain an output sequence. After the information compression of the encoder and the information reconstruction of the decoder, the input sequence AG has the feature transformation capability, and can generate the corresponding AP. By continuously training and updating the weights, the capabilities of the generator will be continuously enhanced, making the AP generated by the AG more accurate and realistic.
(2) Design of discriminator
The arbiter has a function of discriminating whether the input sequence is the sequence AP generated by the generator or the true sequence AP, and the basic structure is as shown in fig. 8(a) and 8 (b). Fig. 9 depicts the detailed design of the arbiter. The input is AG and AP or AP, firstly, the two sequences are spliced, then the two sequences sequentially pass through 5 encoders, the dimension of the final output sequence is 32 multiplied by 1, each value in the sequence is 0 or 1, and the credibility of each partial sequence in the input sequence is represented.
(3) Arbiter training process
FIG. 10 depicts the training process of the arbiter. Inputting AG into a generator to obtain AP, then using a discriminator to analyze AG and AP and AG and AP respectively, wherein the output result of the discriminator should be correct because AP is a real sequence, the expected output result is wrong because AP is a generated sequence, combining errors generated in two discrimination processes, and using an optimizer to guide the discriminator to update the weight. It is noted that the discriminators used in fig. 10 are essentially the same and therefore use shared weights as labels. The training process may enable the discriminator to have the ability to recognize true sequences and generate sequences, thereby facilitating the sequence generated by the generator to be closer to a true sequence.
(4) Generator training process
FIG. 11 depicts the training process of the generator. The weight update of the generator is driven by the result of the arbiter together with the difference between AP and AP, which is defined using L1 penalties, representing the sum of the sequence corresponding point differences, and which is trained in cooperation with the generator. It should be noted that the weight update of the generator also depends on the result of the discriminator, so even if the discriminator becomes more powerful through continuous training, the generator will continuously enhance the generating capability through training learning.
2. Model implementation and training
The model is realized flexibly, a Python programming language is used in the method, the model is modeled by combining a TensorFlow deep learning framework, and other voice and deep learning frameworks can also be used. In the invention, the batch processing size is selected to be 1, the learning rate is properly adjusted according to the training condition, and the GPU is used for completing the whole training process.
Third, application deployment phase
In the application stage of the model, the data to be processed is different from the training stage, and the data is usually a long sequence collected at an electricity meter and cannot be directly applied to the trained model. To this end, the invention proposes a sliding window based solution that can handle input sequences of arbitrary length. The specific implementation steps are as follows:
61) taking the s position of the initial ammeter sequence as a starting point, and taking a sequence with a certain length (the length is the same as the length of the sequence used in the training stage);
62) inputting the sequence obtained in the step 61) into a generator of a trained model to obtain a corresponding output sequence, and marking according to a time stamp of an input subsequence;
63) moving the position of the starting point back by i units, continuing to perform the step 61) until the original sequence is processed to the end, and performing the step 65), and performing the step 64) if the remaining sequence is less than the length of a complete sequence segment;
64) the insufficient portion is complemented by 0, and step 62) is executed;
65) after the steps are completed, all the subsequences obtained in the step 62) are overlapped according to the time stamps, the average value of each time stamp is calculated to serve as a final output sequence, and the length of the sequence is basically equal to that of the initial electric meter sequence.
In a second embodiment, the present embodiment provides a non-intrusive power load monitoring sequence generation system;
a non-intrusive power load monitoring sequence generation system, comprising:
an acquisition module configured to acquire a total data sequence of the electricity meter;
the output module is configured to input the total data sequence of the electric meter into a pre-trained conditional countermeasure generation network, and the pre-trained conditional countermeasure generation network outputs the data sequence of the target electric appliance; wherein the conditional countermeasure generating network comprises a generating network and a discriminating network; the generating network is used for generating a synthetic data sequence similar to the real data sequence of the target electrical appliance, and the judging network is used for judging whether the synthetic data sequence generated by the generating network is the real data sequence of the target electrical appliance in the training phase of the conditional countermeasure generating network.
In a third embodiment, the present disclosure further provides an electronic device, which includes a memory, a processor, and a computer instruction stored in the memory and executed on the processor, where when the computer instruction is executed by the processor, each operation in the method is completed, and for brevity, details are not described here again.
The electronic device may be a mobile terminal and a non-mobile terminal, the non-mobile terminal includes a desktop computer, and the mobile terminal includes a Smart Phone (such as an Android Phone and an IOS Phone), Smart glasses, a Smart watch, a Smart bracelet, a tablet computer, a notebook computer, a personal digital assistant, and other mobile internet devices capable of performing wireless communication.
It should be understood that in the present disclosure, the processor may be a central processing unit CPU, but may also be other general purpose processors, digital signal processors DSP, application specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory may include both read-only memory and random access memory, and may provide instructions and data to the processor, and a portion of the memory may also include non-volatile random access memory. For example, the memory may also store device type information.
In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in a processor or instructions in the form of software. The steps of a method disclosed in connection with the present disclosure may be embodied directly in a hardware processor, or in a combination of the hardware and software modules within the processor. The software modules may be located in ram, flash, rom, prom, or eprom, registers, among other storage media as is well known in the art. The storage medium is located in a memory, and a processor reads information in the memory and completes the steps of the method in combination with hardware of the processor. To avoid repetition, it is not described in detail here. Those of ordinary skill in the art will appreciate that the various illustrative elements, i.e., algorithm steps, described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is merely a division of one logic function, and there may be other divisions when actually implemented, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or units through some interfaces, and may be in an electrical, mechanical or other form.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (10)
1. The non-intrusive power load monitoring sequence generation method is characterized by comprising the following steps:
acquiring a total data sequence of the electric meter;
inputting the total data sequence of the electric meter into a pre-trained conditional countermeasure generating network, and outputting the data sequence of the target electric appliance by the pre-trained conditional countermeasure generating network; wherein the conditional countermeasure generating network comprises a generating network and a discriminating network; the generating network is used for generating a synthetic data sequence similar to the real data sequence of the target electrical appliance, and the judging network is used for judging whether the synthetic data sequence generated by the generating network is the real data sequence of the target electrical appliance in the training phase of the conditional countermeasure generating network.
2. The method as claimed in claim 1, wherein a corresponding conditional countermeasure generation network is trained for each target appliance, and a training set corresponding to each target appliance is prepared for the training of the target appliance.
3. The method of claim 1, wherein the step of inputting the total data sequence of the electricity meters into the pre-trained conditional countermeasure generating network, and the pre-trained conditional countermeasure generating network outputting the data sequence of the target electrical appliance, further comprises:
selecting a training data set and a testing data set from a historical database of total data of the electric meters; training data sets including total electric meter data and a real data sequence of a certain target electric appliance;
inputting the total data of the electric meters in the training data set into a generation network of the conditional countermeasure generation network for training; judging whether the synthetic data sequence generated by the generating network is the real data sequence of the current target electrical appliance or not through the judging network;
if the judgment network judges that the synthetic data sequence generated by the generation network is the real data sequence of the current target electrical appliance, the training is finished;
testing the generated model through the test data set to obtain a corresponding test result; and if the test result meets the set requirement, obtaining a trained conditional countermeasure generation network.
4. The method as claimed in claim 3, wherein if the discrimination network determines that the synthesized data sequence generated by the generation network is not the real data sequence of the current target electrical appliance, the parameters of the generation network are adjusted and then the training is continued.
5. The method of claim 3, wherein the step of preparing the training data set comprises:
selecting a data set, wherein the data set comprises a total data sequence of electric meters and a plurality of electric appliance electricity utilization data sequences in the same time period; the electric meter total data sequence and the electricity utilization sequence both comprise: the method comprises the steps of numbering electrical appliances, collecting timestamps of power of the electrical appliances and power values of the electrical appliances corresponding to the timestamps;
performing missing value processing and abnormal value processing on the data in the selected data set;
eliminating noise data from the data sequence after processing the missing value and the abnormal value;
sequences with noise data eliminated are divided into two categories: a data sequence in an operating state and a data sequence in a non-operating state;
selecting a certain electric appliance, and selecting the power utilization data sequence of the electric appliance according to the name of the electric appliance, wherein the power utilization data sequence comprises a data sequence of the electric appliance in a working state and a data sequence of the electric appliance in a non-working state;
selecting a total data sequence of electric meters in the same power utilization time range according to the power utilization time range of the electric appliance; the total data sequence of the electric meters in the same electricity utilization time range comprises a data sequence in an operating state and a data sequence in a non-operating state of all the electric appliances in the electricity utilization range;
and aligning the time stamps of the electricity utilization data sequence of the electric appliance to the total electricity utilization data sequence of the same electricity utilization time range.
6. The method of claim 5, wherein the missing value processing step comprises:
step (11): scanning the electric meter total data sequence or the electric power utilization sequence from front to back, calculating the difference value of every two adjacent time stamps, and judging whether each difference value is in a set range;
step (12): if the current time is within the set range, continuing to execute the step (11); if the difference is not in the set range, comparing the difference with the set threshold, and if the difference is greater than the set threshold, continuing to execute the step (11); if the value is smaller than the set threshold value, entering the step (13);
step (13): calculating the mean value of the values corresponding to the two timestamps which are currently processed, and using the changed mean value as the reconstruction value of the missing data;
reconstructing the timestamp corresponding to the missing value: the starting time stamp of the missing interval is marked as s, the ending time stamp is marked as e, the number of the reconstructed time stamps is
The number of the reconstructed time stamps is recorded as c, the reconstructed time stamps are added with q each time from the starting time stamp s, and the process is repeated for c times;
step (15) of inserting the reconstruction value generated in step (13) and the reconstruction timestamp generated in step (14) into the original sequence in order;
step (16) repeats step (11) until the entire sequence of scans is complete.
7. The method of claim 5, wherein the outlier processing step comprises:
21) scanning the total data sequence or the power utilization sequence of the electric meter according to the sequence from front to back, and executing the step 22) if the value of the current sampling point is higher than the threshold value t for the data points except the beginning and the ending, otherwise, continuing to scan the value of the next sampling point;
22) subtracting the numerical values of two adjacent sampling points from the numerical value of the current sampling point, and respectively recording the obtained difference values as b and f;
23) if both b and f are larger than the set threshold, indicating that the value of the current sampling point is an abnormal value, and performing step 24); otherwise, step 21) is executed;
24) and using the average value of two points before and after the current sampling point as a reconstruction value, replacing the value of the current sampling point with the reconstruction value, and performing the step 21) until the sequence traversal is finished.
8. A non-intrusive power load monitoring sequence generation system is characterized by comprising:
an acquisition module configured to acquire a total data sequence of the electricity meter;
the output module is configured to input the total data sequence of the electric meter into a pre-trained conditional countermeasure generation network, and the pre-trained conditional countermeasure generation network outputs the data sequence of the target electric appliance; wherein the conditional countermeasure generating network comprises a generating network and a discriminating network; the generating network is used for generating a synthetic data sequence similar to the real data sequence of the target electrical appliance, and the judging network is used for judging whether the synthetic data sequence generated by the generating network is the real data sequence of the target electrical appliance in the training phase of the conditional countermeasure generating network.
9. An electronic device comprising a memory and a processor and computer instructions stored on the memory and executable on the processor, the computer instructions when executed by the processor performing the steps of the method of any of claims 1 to 7.
10. A computer-readable storage medium storing computer instructions which, when executed by a processor, perform the steps of the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010588708.9A CN111738521B (en) | 2020-06-24 | 2020-06-24 | Non-invasive power load monitoring sequence generation method, system, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010588708.9A CN111738521B (en) | 2020-06-24 | 2020-06-24 | Non-invasive power load monitoring sequence generation method, system, equipment and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111738521A true CN111738521A (en) | 2020-10-02 |
| CN111738521B CN111738521B (en) | 2024-04-05 |
Family
ID=72650930
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010588708.9A Active CN111738521B (en) | 2020-06-24 | 2020-06-24 | Non-invasive power load monitoring sequence generation method, system, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111738521B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113378655A (en) * | 2021-05-24 | 2021-09-10 | 电子科技大学 | Antagonistic energy decomposition method based on deep neural network |
| CN114095241A (en) * | 2021-11-18 | 2022-02-25 | 中国电信股份有限公司 | Detection method, detection device and computer-readable storage medium |
| WO2022143021A1 (en) * | 2020-12-29 | 2022-07-07 | 沃太能源股份有限公司 | Neural network generating method and apparatus, power forecasting method and apparatus, and storage medium |
| CN115617816A (en) * | 2022-11-22 | 2023-01-17 | 北京国科伟业电力科技有限公司 | System for acquiring specified ammeter data table |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107392391A (en) * | 2017-08-10 | 2017-11-24 | 天津求实智源科技有限公司 | A kind of electric load electricity consumption decomposition method based on deep learning |
| CN108229660A (en) * | 2017-12-30 | 2018-06-29 | 天津求实智源科技有限公司 | The electric energy decomposition method screened through background based on deep learning |
| CN109840691A (en) * | 2018-12-31 | 2019-06-04 | 天津求实智源科技有限公司 | Non-intrusion type subitem electricity estimation method based on deep neural network |
| CN110376457A (en) * | 2019-06-28 | 2019-10-25 | 同济大学 | Non-intrusion type load monitoring method and device based on semi-supervised learning algorithm |
-
2020
- 2020-06-24 CN CN202010588708.9A patent/CN111738521B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107392391A (en) * | 2017-08-10 | 2017-11-24 | 天津求实智源科技有限公司 | A kind of electric load electricity consumption decomposition method based on deep learning |
| CN108229660A (en) * | 2017-12-30 | 2018-06-29 | 天津求实智源科技有限公司 | The electric energy decomposition method screened through background based on deep learning |
| CN109840691A (en) * | 2018-12-31 | 2019-06-04 | 天津求实智源科技有限公司 | Non-intrusion type subitem electricity estimation method based on deep neural network |
| CN110376457A (en) * | 2019-06-28 | 2019-10-25 | 同济大学 | Non-intrusion type load monitoring method and device based on semi-supervised learning algorithm |
Non-Patent Citations (2)
| Title |
|---|
| YUNGANG PAN ETAL: "《sequence-to-subsequence learning with conditional gan for power disaggregation》", 《IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING》 * |
| 张宇帆等: "《基于生成对抗网络的负荷序列随机场景生成方法》", 《供用电》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022143021A1 (en) * | 2020-12-29 | 2022-07-07 | 沃太能源股份有限公司 | Neural network generating method and apparatus, power forecasting method and apparatus, and storage medium |
| CN113378655A (en) * | 2021-05-24 | 2021-09-10 | 电子科技大学 | Antagonistic energy decomposition method based on deep neural network |
| CN113378655B (en) * | 2021-05-24 | 2022-04-19 | 电子科技大学 | Antagonistic energy decomposition method based on deep neural network |
| CN114095241A (en) * | 2021-11-18 | 2022-02-25 | 中国电信股份有限公司 | Detection method, detection device and computer-readable storage medium |
| CN115617816A (en) * | 2022-11-22 | 2023-01-17 | 北京国科伟业电力科技有限公司 | System for acquiring specified ammeter data table |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111738521B (en) | 2024-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111738521A (en) | Non-invasive power load monitoring sequence generation method, system, equipment and medium | |
| Kaltenecker et al. | Distance-based sampling of software configuration spaces | |
| CN111367961B (en) | Time sequence data event prediction method and system based on graph convolution neural network and application thereof | |
| Hoyer et al. | Nonlinear causal discovery with additive noise models | |
| CN106685674B (en) | Method and device for predicting network event and establishing network event prediction model | |
| CN113255835A (en) | Hydropower station pump equipment anomaly detection method | |
| CN110059845B (en) | Metering device clock error trend prediction method based on time sequence evolution gene model | |
| CN102521080B (en) | Computer data recovery method for electricity-consumption information collecting system for power consumers | |
| CN111382789B (en) | Power load identification method and system based on machine learning | |
| CN109740648A (en) | Electric load disorder data recognition method, apparatus and computer equipment | |
| CN111861023B (en) | Mixed wind power prediction method and device based on statistics | |
| CN112819082B (en) | Satellite spectrum sensing data reconstruction method based on deep learning | |
| Dorn et al. | Mastering uncertainty in performance estimations of configurable software systems | |
| González‐Martínez et al. | Multisynchro: a novel approach for batch synchronization in scenarios of multiple asynchronisms | |
| CN112785067A (en) | Data prediction method and device, equipment and storage medium | |
| CN114881123A (en) | Method and device for detecting periodic data abnormity, electronic equipment and storage medium | |
| CN110597880A (en) | Method, system and equipment for mining and matching water use patterns | |
| KR102500548B1 (en) | Power consumption prediction system of residential and commercial building using hybrid convolution neural network and method thereof | |
| Wang et al. | Deep blind image quality assessment powered by online hard example mining | |
| Li et al. | Load Profile Inpainting for Missing Load Data Restoration and Baseline Estimation | |
| CN113327008A (en) | Electricity stealing detection method, system and medium based on time sequence automatic encoder | |
| WO2024093207A1 (en) | Time series data processing method and apparatus, device, and nonvolatile readable storage medium | |
| CN117251672A (en) | Regional industry multi-dimensional energy carbon emission calculation method and device based on electric power data | |
| WO2016200413A1 (en) | Application session analysis and recommendation system | |
| CN111124816B (en) | HDP algorithm-based server log analysis method and system |
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 |