CN111783684A - Bi-GRU-SA-based household electrical load decomposition method and system - Google Patents

Bi-GRU-SA-based household electrical load decomposition method and system Download PDF

Info

Publication number
CN111783684A
CN111783684A CN202010636012.9A CN202010636012A CN111783684A CN 111783684 A CN111783684 A CN 111783684A CN 202010636012 A CN202010636012 A CN 202010636012A CN 111783684 A CN111783684 A CN 111783684A
Authority
CN
China
Prior art keywords
load decomposition
gru
data
neural network
load
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010636012.9A
Other languages
Chinese (zh)
Inventor
何洪英
罗滇生
黎灿兵
周斌
蒋宇翔
高思远
尹希浩
赵友琳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hunan University
Original Assignee
Hunan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hunan University filed Critical Hunan University
Priority to CN202010636012.9A priority Critical patent/CN111783684A/en
Publication of CN111783684A publication Critical patent/CN111783684A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • 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
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Signal Processing (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a household electrical load decomposition method and a system based on Bi-GRU-SA, wherein the method comprises the following steps: collecting household power data of a user and power data of single equipment in a period of time; establishing a load decomposition model of a bidirectional gating recurrent neural network based on a self-attention mechanism; dividing the household power consumption data into training set data and test set data; training the load break model using the training set; the gated neural network can well avoid gradient explosion and local optimization; the bidirectional mechanism can fully utilize the information of the future time; the self-attention mechanism enables important features to be highlighted; therefore, the method for decomposing the household power load can greatly improve the accuracy of load decomposition.

Description

Bi-GRU-SA-based household electrical load decomposition method and system
Technical Field
The invention relates to the technical field of intelligent power utilization, in particular to a Bi-GRU-SA-based household power utilization load decomposition method and system.
Background
At present, both in the production field and the consumption field, the reduction of the electric energy consumption is focused on, so that the energy is efficiently utilized, and the sustainable development of the energy is realized. For owners and residents, the power consumption of each household electric device is known, and the targeted power saving behavior is facilitated.
The load monitoring technology is a power consumption detection technology comparing mainstream power equipment, and is divided into intrusive load monitoring and non-intrusive load decomposition. The intrusive load monitoring needs to equip each electric device with a sensor with a communication function, and the sensors acquire the electricity utilization information of each electric device to analyze the electricity utilization behavior of a user.
Non-intrusive load splitting is a technique that requires only collecting total load data at the customer's power input and splitting the customer's total load into various electrical equipment loads. The energy consumption information of the household electrical appliance can be provided for resident users, so that the users are guided to economically and reasonably arrange electricity, and the energy utilization efficiency is improved; the method is more beneficial to guiding a power grid company to reasonably arrange and schedule, and realizes peak elimination and valley filling.
In addition, compared with an invasive monitoring technology, the non-invasive load decomposition technology does not need a resident user to install a data acquisition unit at each equipment end, so that the privacy safety of the user is protected, the installation and detection cost is reduced, and the practicability is higher.
The accuracy of the non-intrusive load splitting technique depends on the validity of the extracted load features and the validity of the identification method used. The existing non-intrusive load decomposition technology extracts the characteristics of the current operation time of the electrical equipment, and the characteristics of the operation state of the electrical equipment are time-related, namely the current operation state of the equipment is related to the operation state of the equipment at the previous time, and the operation state of the equipment at the later time is also influenced. Only the characteristics of the current running time are adopted, the time relevance of the characteristics is not considered, and the importance difference between different characteristics is not considered, so that the accuracy of the final decomposition result is to be improved.
Disclosure of Invention
The invention mainly aims to provide a Bi-GRU-SA-based household electrical load decomposition method and system, and aims to solve the problem that the accuracy of a final decomposition result of a non-invasive load decomposition technology in the prior art needs to be improved.
The invention provides a household electrical load decomposition method based on Bi-GRU-SA, which comprises the following steps:
collecting household power data of a user and power data of single equipment in a period of time;
establishing a load decomposition model based on a bidirectional gating recurrent neural network introducing a self-attention mechanism;
dividing the household power consumption data into training set data and test set data;
training the load break model using the training set data;
and applying the load decomposition model to carry out household power load decomposition on the user.
Preferably, the establishing a load decomposition model based on a bidirectional gated recurrent neural network introducing a self-attention mechanism includes:
establishing a bidirectional gated recurrent neural network, wherein the expression of an output variable of the bidirectional gated recurrent neural network is as follows:
Figure BDA0002569292630000021
Figure BDA0002569292630000022
Figure BDA0002569292630000023
wherein, YtFor the output variable, the bidirectional gated recurrent neural network includes a forward layer for processing information at a next time and a backward layer for processing information at a previous time, xtFor the purpose of the time series of inputs,
Figure BDA0002569292630000024
for the output concealment factor of the forward layer,
Figure BDA0002569292630000025
is an output hiding factor, Y, of the backward layertIn order to output the variable, the output variable,
Figure BDA0002569292630000026
wb、w'b、wfand w'fAll are weight vectors;
and taking the bidirectional gated recurrent neural network as a load decomposition model.
Preferably, the establishing a load decomposition model based on a bidirectional gated recurrent neural network introducing a self-attention mechanism further includes:
establishing a self-attention mechanism layer, wherein an expression of an environment vector of the self-attention mechanism layer is as follows:
Figure BDA0002569292630000027
Figure BDA0002569292630000031
Figure BDA0002569292630000032
wherein f isattIn order to be a function of the attention,vaand waAre all learnable network parameters, HtFor hidden state sequences at the present moment, HiA sequence of hidden states at time i, and having Hi＝{h1,h2,...,hn}，{h1,h2,...,hn}itanh represents the activation function as a hyperbolic tangent function as a hidden state vector at time i αiTo focus on weight, αiRepresenting attention weights generated by the hidden state vectors at the historical moments, wherein C is an environment vector and represents a linear weighted combination of the attention weights and the hidden state sequence at the current moment;
establishing a full connection layer and a normalization index function layer;
and constructing the bidirectional gated recurrent neural network, the self-attention mechanism layer, the full connection layer and the normalized exponential function layer as the load decomposition model.
Preferably, the training the load split model using the training set data includes:
training the non-invasive load decomposition model by adopting a supervised learning algorithm to obtain model parameters;
preferably, the training of the load decomposition model using the training set data further includes:
and carrying out accuracy test on an output result generated by decomposing the test set data by using the trained load decomposition model.
Preferably, the acquiring the household power data of the user further includes:
and denoising and normalizing the household power consumption data.
Preferably, the denoising and normalizing the household power consumption data includes:
and denoising the household power consumption data by adopting a wavelet denoising method based on Bayesian estimation.
Preferably, the denoising processing of the household power data by using a wavelet denoising method based on bayesian estimation further includes:
and (3) normalizing the denoised household power data by adopting a (0,1) normalization algorithm.
The invention also provides a Bi-GRU-SA-based household electrical load decomposition system, which is applied to any one of the Bi-GRU-SA-based household electrical load decomposition methods, and the system comprises:
the system comprises an acquisition module, a storage module and a processing module, wherein the acquisition module is used for acquiring household power consumption data of a user and power data of a single device in a period of time, and dividing the household power consumption data into training set data and test set data;
the load decomposition module is used for establishing a load decomposition model based on a bidirectional gating recurrent neural network introducing a self-attention mechanism;
the training and testing module is used for training the load decomposition model by using the training set data;
and the application module is used for applying the load decomposition model to carry out household power load decomposition on the user.
Through above-mentioned technical scheme, can realize following beneficial effect:
the invention provides a household electrical load decomposition method based on Bi-GRU-SA; the core of the method is to establish a load decomposition model, and the model is based on Bi-GRU-SA; the Bi-GRU-SA is (Bi-directional Gated Recurrent Unit-SelfAttention mechanism), namely a bidirectional gating recursion Unit neural network introducing a self-attention mechanism; namely, a Bi-directional mechanism (Bi-directional) and a self-attention mechanism (self-attention mechanism) are introduced into the gated recurrent neural network GRU.
The accuracy of the load decomposition depends on the effectiveness of the extracted load features and the effectiveness of the identification method used; the existing non-intrusive load decomposition technology extracts the characteristics of the current operation time of the electrical equipment, and the characteristics of the operation state of the electrical equipment are time-related, namely the current operation state of the equipment is related to the operation state of the equipment at the previous time, and the operation state of the equipment at the later time is also influenced. Only the characteristics of the current running time are adopted, the time relevance of the characteristics is not considered, and the importance difference between different characteristics is not considered, so that the accuracy of the final decomposition result is to be improved.
Gated neural networks have significant advantages in processing time series, and have enjoyed great success in the processing of time series data such as speech, text, financial data, audio, video, weather, and the like. The invention utilizes the bidirectional mechanism of the bidirectional gated neural network (Bi-GRU) to automatically bring the characteristics of the equipment at the current time, the previous time and the later time into the characteristic extraction, thereby improving the accuracy of load decomposition.
Because the gated neural network adopts the transmission and the update of two pieces of gate control information, the gated neural network is used as a variant of a Long Short-Term Memory network (LSTM) and has a simpler structure compared with the Long Short-Term Memory network (LSTM), thereby the load decomposition has a faster speed.
In addition, the attention mechanism (Attentionmechanism) can calculate the importance degree of each element in the sequence and combine all the elements according to the importance degree; in view of this, the invention adds a Self-Attention (Self-Attention) mechanism to the constructed gated neural network, so as to highlight important features and further improve the accuracy of measurement.
In conclusion, the gated neural network introduces the information of the current operation time, the previous time and the next time into the calculation, so that gradient explosion and local optimization can be well avoided; the self-attention mechanism enables important features to be highlighted; therefore, the accuracy of load decomposition can be greatly improved by the household power load decomposition method.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the structures shown in the drawings without creative efforts.
Fig. 1 is a flow chart of a first embodiment of a Bi-GRU-SA-based household electrical load decomposition method according to the present invention;
fig. 2 is a schematic structural diagram of a bidirectional gated recurrent neural network in a second embodiment of the Bi-GRU-SA-based household electrical load decomposition method according to the present invention;
FIG. 3 is a schematic structural diagram of a load decomposition model of a Bi-directional gated recursive unit neural network (Bi-GRU-SA) based on a self-attention mechanism in a third embodiment of the Bi-GRU-SA-based household electrical load decomposition method according to the present invention;
FIG. 4 is a comparison diagram of the power load decomposition results of 4 different load decomposition models for a dishwasher with electricity type in a third embodiment of a Bi-GRU-SA based household electrical load decomposition method provided by the present invention;
FIG. 5 is a schematic diagram showing the comparison of the power load decomposition results of a refrigerator with 4 different load decomposition models according to a third embodiment of the Bi-GRU-SA-based household electrical load decomposition method of the present invention;
FIG. 6 is a comparison diagram of the power load decomposition results of 4 different load decomposition models for the lamp as the electricity consumption type in the third embodiment of the method for decomposing the household electrical load based on the Bi-GRU-SA according to the present invention;
FIG. 7 is a schematic diagram showing the comparison of the power load decomposition results of a microwave oven with 4 different load decomposition models for the power consumption type in the third embodiment of the household power load decomposition method based on Bi-GRU-SA according to the present invention;
fig. 8 is a table comparing accuracy of decomposition results of 4 different load decomposition models for each load type in the third embodiment of the Bi-GRU-SA based household electrical load decomposition method according to the present invention.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention provides a household electric load decomposition method and system based on Bi-GRU-SA.
As shown in fig. 1, in a first embodiment of the Bi-GRU-SA based household electrical load decomposition method of the present invention, the method includes the following steps:
step S110: collecting household power data of a user and power data of a single device in a period of time.
Specifically, in this embodiment, the household power consumption data of the user is collected through the intelligent household electricity meter, the power data includes power data of each household electrical device, and the power data is stored in a time series manner. The power data of each electric device is collected through a temporarily installed branch ammeter and power data of each electric device in a period of time, wherein the power data of each electric device is contained in the power data of each electric device, and the power data are stored in a time series mode.
Step S120: and establishing a load decomposition model based on a bidirectional gated recurrent neural network introducing a self-attention mechanism.
In particular, in recent years, artificial intelligence technology has been developed vigorously, and has been applied successfully in the fields of speech recognition, image recognition, medical diagnosis, and the like. Artificial intelligence techniques typified by machine learning have also received great attention in power systems.
The high-speed and high-accuracy decomposition of the load is realized by utilizing the efficient capability of a gated recurrent neural network GRU (gated recurrent unit) in processing time series.
Step S130: and dividing the household power consumption data into training set data and test set data.
Step S140: training the load break model using the training set data.
Step S150: and applying the load decomposition model to carry out household power load decomposition on the user.
The invention provides a household electrical load decomposition method based on Bi-GRU-SA; the core of the method is to establish a load decomposition model, and the model is based on Bi-GRU-SA; the Bi-GRU-SA is (Bi-directional Gated Recurrent Unit-SelfAttention mechanism), namely a bidirectional gating recursion Unit neural network introducing a self-attention mechanism; namely, a Bi-directional mechanism (Bi-directional) and an attention-seeking mechanism (Attentionmechanism) are introduced into the gated recurrent neural network GRU.
The accuracy of load decomposition depends on the effectiveness of the extracted load characteristics and the effectiveness of the used artificial intelligence identification method, namely, the characteristics of the current operation time of the electrical equipment are extracted by the existing non-invasive load decomposition technology, and the operation state characteristics of the electrical equipment are time-related, namely, the current operation state of the equipment is related to the operation state of the previous time, and the operation state of the latter time is influenced. Only the characteristics of the current running time are adopted, the time relevance of the characteristics is not considered, and the importance difference between different characteristics is not considered, so that the accuracy of the final decomposition result is to be improved.
Gated neural networks have significant advantages in processing time series, and have enjoyed great success in the processing of time series data such as speech, text, financial data, audio, video, weather, and the like. The invention utilizes the bidirectional mechanism of the bidirectional gated neural network (Bi-GRU) to automatically bring the characteristics of the equipment at the current time, the previous time and the later time into the characteristic extraction, thereby improving the accuracy of load decomposition.
Since the gated neural network employs two kinds of gate control information transfer and update, as a variation of LMST, it has a simpler structure than conventional LMST, thereby allowing load decomposition at a faster speed.
In addition, the attention mechanism (Attentionmechanism) can calculate the importance degree of each element in the sequence and combine all the elements according to the importance degree; in view of this, the invention adds a Self-Attention (Self-Attention) mechanism to the constructed gated neural network, so as to highlight important features and further improve the accuracy of measurement.
In conclusion, the gated neural network introduces the information of the current operation time, the previous time and the next time into the calculation, so that gradient explosion and local optimization can be well avoided; the self-attention mechanism enables important features to be highlighted; therefore, the accuracy of load decomposition can be greatly improved by the household power load decomposition method.
In a second embodiment of the Bi-GRU-SA based household electrical load decomposition method proposed by the present invention, based on the first embodiment, step S120 includes the following steps:
step S210: establishing a bidirectional gated recurrent neural network, wherein an output variable Y of the bidirectional gated recurrent neural networktThe expression of (a) is:
Figure BDA0002569292630000071
Figure BDA0002569292630000081
Figure BDA0002569292630000082
wherein, YtFor the output variable, the bidirectional gated recurrent neural network includes a forward layer for processing information at a next time and a backward layer for processing information at a previous time, xtFor the purpose of the time series of inputs,
Figure BDA0002569292630000083
for the output concealment factor of the forward layer,
Figure BDA0002569292630000084
is an output hiding factor, Y, of the backward layertIn order to output the variable, the output variable,
Figure BDA0002569292630000085
wb、w'b、wfand w'fAre all weight vectors.
In particular, the gated recurrent neural network (GRU) is a variation of the long-short term memory neural network (LSTM) that selects information by updating and resetting gates, and has a simpler structure than the long-short term memory neural network (LSTM) that selects information by using input gates, forgetting gates, and output gates, while retaining the advantages of the long-short term memory neural network (LSTM) in preventing the disappearance of gradients or gradient explosions.
Step S220: and taking the bidirectional gated recurrent neural network as a load decomposition model.
Specifically, the bidirectional gated recurrent neural network (BI-GRU) adopted by the invention has two hierarchies of a forward layer and a backward layer, wherein the forward layer processes information at the next moment, and the backward layer processes information at the previous moment.
As shown in fig. 2, the forward layer on the left learns the information at the previous moment, and the information fed back by the forward layer is in a forward sequence; the backward layer on the right side learns the information at the next moment; the information fed back is reverse-sequenced. Two hidden factors are cascaded with each other and iterated layer by layer to finally obtain an output variable Yt. The bidirectional gated recurrent neural network (BI-GRU) is a bidirectional structure that enables time series data to be processed more efficiently than a general gated recurrent neural network (GRU), resulting in an increase in the accuracy of load decomposition.
In a third embodiment of the Bi-GRU-SA based household electrical load decomposition method provided by the present invention, based on the second embodiment, step S120 further includes the following steps:
step S310: establishing a self-attention mechanism layer, wherein an expression of an environment vector of the self-attention mechanism layer is as follows:
Figure BDA0002569292630000086
Figure BDA0002569292630000087
Figure BDA0002569292630000091
wherein f isattAs a function of attention, vaAnd waAre all learnable network parameters, HtFor hidden state sequences at the present moment, HiA sequence of hidden states at time i, and having Hi＝{h1,h2,...,hn}，{h1,h2,...,hn}itanh represents the activation function as a hyperbolic tangent function as a hidden state vector at time i αiTo focus on weight, αiAnd C represents a linear weighted combination of the attention weight and the hidden state sequence at the current moment.
Specifically, with the increasing of input data and the increasing of input sequences, the information acquisition capacity of the whole network model is limited, and the output variable Y is obtained through a bidirectional gated recurrent neural network (BI-GRU)tBeing fixed to a certain length due to the limitation of the original model, the method may cause information defects, so that the prediction result generates a large error, and therefore, a self-attention mechanism is introduced.
The self-attention mechanism distinguishes the importance degree of data by applying attention to target data, and can quickly extract important features of sparse data; the attention degree of the information is emphasized by weighting the target data, and the weighting factor is automatically adjusted along with the increase of the acquired information.
The invention weights the characteristic data through a self-attention (self-attention) mechanism, highlights important information, ignores noise and redundant information, and further improves the accuracy of load decomposition.
Self-attention (self-attention), also called internal attention, is an improvement in the self-attention mechanism that reduces reliance on external information and is better at capturing internal correlations of data or features. f. ofattIs an alignment model, which reflects the alignment degree of the source terminal and the target terminal, i.e. the attention function.
Step S320: and establishing a full connection layer and a normalized exponential function layer.
Step S330: and constructing the bidirectional gated recurrent neural network, the self-attention mechanism layer, the full connection layer and the normalized exponential function layer as the load decomposition model.
Specifically, the finally obtained model is a load decomposition model of a Bi-directional gated recursion unit neural network (Bi-GRU-SA) based on the self-attention mechanism, and a structural schematic diagram of the model is shown in fig. 3; the self-attention (self-attention) mechanism in the present invention employs a global attention mode, allowing the decoder to pay attention to each location until all locations including the location are paid attention, thus allowing the results of the load decomposition model to have high accuracy. The self-attention (self-attention) mechanism is characterized in that the difference of position spaces can be ignored, the dependency relationship among the characteristics can be directly calculated, parallel calculation can be realized, and the self-attention (self-attention) mechanism has high timeliness.
In order to compare the final application effect of the load decomposition model provided by the invention, the invention uses other 3 different load decomposition models (respectively a load decomposition model based on a gated recursion unit neural network, a load decomposition model based on a gated recursion unit neural network introducing a self-attention mechanism and a load decomposition model based on a bidirectional gated recursion unit neural network) together with the load decomposition model provided by the invention (based on a bidirectional gated recursion unit neural network introducing a self-attention mechanism) to carry out load decomposition on the household electric power of an actual user, and compares the final decomposition result with a real result.
The power consumption of the actual user household comprises four types of loads of a refrigerator, a microwave oven, an electric lamp and a dishwasher, and the decomposition conditions of 4 different models for the 4 types of power consumption loads are shown in the attached figures 4-8 (black curves in the figures represent the decomposition results of the 4 different load decomposition models, and gray is the actual load power of the electric appliance). It can be known from the figure that the load decomposition model (based on the load decomposition model of the bidirectional gated recursion unit neural network introducing the self-attention mechanism) provided by the invention effectively and accurately decomposes results for 4 loads, has high matching degree with real load power waveforms, can accurately distinguish load power models of different household appliances, has high adaptability to different household load decompositions, and particularly has very high accuracy on identification of load models of dishwashers and refrigerators.
In a fourth embodiment of the Bi-GRU-SA based household electrical load decomposition method proposed by the present invention, based on the first embodiment, step S140 includes the following steps:
step S410: and training the load decomposition model by adopting a supervised learning algorithm to obtain model parameters.
Specifically, the weight vector in the bidirectional gated recurrent neural network (BI-GRU) is
Figure BDA0002569292630000101
wb、w'b、wfAnd w'fAnd weight vector W in the self attention mechanism layeraAnd VaIt is obtained through learning and training.
The invention trains the network by adopting a deep learning method, and the deep learning method comprises the steps of supervised learning and unsupervised learning. The supervised learning method trains a network through the corresponding relation between the existing input data and the output data, and adopts an objective function to guide network parameters to realize optimization.
In a fifth embodiment of the Bi-GRU-SA based household electrical load decomposition method proposed by the present invention, based on the first embodiment, step S140, then comprises the following steps:
step S510: and carrying out accuracy test on an output result generated by decomposing the test set data by using the trained load decomposition model.
Specifically, the accuracy check can be completed by comparing the output result with the power data of the single device collected in the first embodiment.
In a sixth embodiment of the Bi-GRU-SA based household electrical load decomposition method according to the present invention, based on the first embodiment, step S110, then further includes the following steps:
step S610: and denoising and normalizing the household power consumption data.
Specifically, noise interference inevitably exists in the power data obtained by sampling due to the existence of line transmission noise and electronic noise of a sampling circuit. The invention needs to carry out denoising and normalization processing on the acquired power data.
In a seventh embodiment of the Bi-GRU-SA based household electrical load decomposition method provided by the present invention, based on the seventh embodiment, step S610 further includes the following steps:
step S710: and denoising the household power consumption data by adopting a wavelet denoising method based on Bayesian estimation.
Specifically, a wavelet denoising method based on Bayesian estimation is adopted to denoise noise, so that subsequent feature extraction and correct load decomposition are facilitated.
In an eighth embodiment of the Bi-GRU-SA based household electrical load decomposition method according to the present invention, based on the seventh embodiment, step S710 is followed by the following steps:
step S810: and (3) normalizing the denoised household power data by adopting a (0,1) normalization algorithm.
Specifically, because the power ranges of the household appliance loads may be different, some have large power and some have small power, the effect of data of low-power equipment may be weakened by adopting the original index for calculation, so that the decomposition has accuracy and the precision loss is reduced, the patent performs normalization processing on the data in advance.
The formula of the normalization processing is as follows:
Figure BDA0002569292630000111
the invention also provides a Bi-GRU-SA-based household electrical load decomposition system, which is applied to any one embodiment of the Bi-GRU-SA-based household electrical load decomposition method, and the system comprises:
the system comprises an acquisition module, a storage module and a processing module, wherein the acquisition module is used for acquiring household power consumption data of a user and power data of a single device in a period of time, and dividing the household power consumption data into training set data and test set data;
the load decomposition module is used for establishing a load decomposition model based on a bidirectional gating recurrent neural network introducing a self-attention mechanism;
the training and testing module is used for training the load decomposition model by using the training set data;
and the application module is used for applying the load decomposition model to carry out household power load decomposition on the user.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, wherein the software product is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk), and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (9)

1. A household electrical load decomposition method based on Bi-GRU-SA is characterized by comprising the following steps:
collecting household power data of a user and power data of single equipment in a period of time;
establishing a load decomposition model based on a bidirectional gating recurrent neural network introducing a self-attention mechanism;
dividing the household power consumption data into training set data and test set data;
training the load break model using the training set data;
and applying the load decomposition model to carry out household power load decomposition on the user.
2. The Bi-GRU-SA based domestic electrical load decomposition method according to claim 1, wherein the establishing of the load decomposition model based on the bidirectional gated recurrent neural network introducing the self-attention mechanism comprises:
establishing a bidirectional gated recurrent neural network, wherein an output variable Y of the bidirectional gated recurrent neural networktThe expression of (a) is:
Figure FDA0002569292620000011
Figure FDA0002569292620000012
Figure FDA0002569292620000013
wherein, YtFor the output variable, the bidirectional gated recurrent neural network includes a forward layer for processing information at a next time and a backward layer for processing information at a previous time, xtFor the purpose of the time series of inputs,
Figure FDA0002569292620000014
for the output concealment factor of the forward layer,
Figure FDA0002569292620000015
is an output hiding factor, Y, of the backward layertIn order to output the variable, the output variable,
Figure FDA0002569292620000016
wb、w'b、wfand w'fAll are weight vectors;
and taking the bidirectional gated recurrent neural network as a load decomposition model.
3. The Bi-GRU-SA based domestic electrical load decomposition method according to claim 1, wherein the establishing of the load decomposition model based on the bidirectional gated recurrent neural network introducing the self-attention mechanism further comprises:
establishing a self-attention mechanism layer, wherein an expression of an environment vector of the self-attention mechanism layer is as follows:
Figure FDA0002569292620000017
Figure FDA0002569292620000021
Figure FDA0002569292620000022
wherein f isattAs a function of attention, vaAnd waAre all learnable network parameters, HtFor hidden state sequences at the present moment, HiA sequence of hidden states at time i, and having Hi＝{h1,h2,...,hn}，{h1,h2,...,hn}itanh represents the activation function as a hyperbolic tangent function as a hidden state vector at time i αiTo focus on weight, αiRepresenting attention weights generated by the hidden state vectors at the historical moments, wherein C is an environment vector and represents a linear weighted combination of the attention weights and the hidden state sequence at the current moment;
establishing a full connection layer and a normalization index function layer;
and constructing the bidirectional gated recurrent neural network, the self-attention mechanism layer, the full connection layer and the normalized exponential function layer as the load decomposition model.
4. The Bi-GRU-SA based home electrical load splitting method of claim 1, wherein the training the load splitting model using the training set data comprises:
and training the load decomposition model by adopting a supervised learning algorithm to obtain model parameters.
5. The Bi-GRU-SA based home electrical load splitting method of claim 1, wherein the training the load splitting model using the training set data further comprises:
and carrying out accuracy test on an output result generated by decomposing the test set data by using the trained load decomposition model.
6. The Bi-GRU-SA based household electrical load decomposition method according to claim 1, wherein the collecting household electrical power data of the user further comprises:
and denoising and normalizing the household power consumption data.
7. The Bi-GRU-SA based household electrical load decomposition method of claim 1, wherein the denoising and normalizing the household electrical power data comprises:
and denoising the household power consumption data by adopting a wavelet denoising method based on Bayesian estimation.
8. The Bi-GRU-SA based household electrical load decomposition method according to claim 7, wherein the denoising processing is performed on the household electrical power data by using a wavelet denoising method based on bayesian estimation, and then further comprising:
and (3) normalizing the denoised household power data by adopting a (0,1) normalization algorithm.
9. A Bi-GRU-SA-based household electrical load decomposition system, applied to the Bi-GRU-SA-based household electrical load decomposition method according to any one of claims 1 to 8, the system comprising:
the system comprises an acquisition module, a storage module and a processing module, wherein the acquisition module is used for acquiring household power consumption data of a user and power data of a single device in a period of time, and dividing the household power consumption data into training set data and test set data;
the load decomposition module is used for establishing a load decomposition model based on a bidirectional gating recurrent neural network introducing a self-attention mechanism;
the training and testing module is used for training the load decomposition model by using the training set data;
and the application module is used for applying the load decomposition model to carry out household power load decomposition on the user.
CN202010636012.9A 2020-07-03 2020-07-03 Bi-GRU-SA-based household electrical load decomposition method and system Pending CN111783684A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010636012.9A CN111783684A (en) 2020-07-03 2020-07-03 Bi-GRU-SA-based household electrical load decomposition method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010636012.9A CN111783684A (en) 2020-07-03 2020-07-03 Bi-GRU-SA-based household electrical load decomposition method and system

Publications (1)

Publication Number Publication Date
CN111783684A true CN111783684A (en) 2020-10-16

Family

ID=72758569

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010636012.9A Pending CN111783684A (en) 2020-07-03 2020-07-03 Bi-GRU-SA-based household electrical load decomposition method and system

Country Status (1)

Country Link
CN (1) CN111783684A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112348789A (en) * 2020-11-03 2021-02-09 哈尔滨市科佳通用机电股份有限公司 Train lock catch loss fault detection method, system and device
CN113393025A (en) * 2021-06-07 2021-09-14 浙江大学 Non-invasive load decomposition method based on Informer model coding structure
CN113434493A (en) * 2021-06-28 2021-09-24 湘潭大学 Non-invasive load decomposition method based on Transformer
CN113469412A (en) * 2021-06-02 2021-10-01 国核电力规划设计研究院有限公司 Real-time operation strategy optimization method and system for comprehensive energy system
CN113595242A (en) * 2021-06-28 2021-11-02 湖南大学 Non-invasive load identification method based on deep CNN-HMM
CN114037178A (en) * 2021-11-23 2022-02-11 天津大学 Non-invasive load decomposition method based on unsupervised pre-training neural network

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108416695A (en) * 2018-02-24 2018-08-17 合肥工业大学 Electric load probability density prediction technique based on deep learning and system, medium
CN108510985A (en) * 2017-02-24 2018-09-07 百度（美国）有限责任公司 System and method for reducing the principle sexual deviation in production speech model
US10199038B1 (en) * 2012-11-29 2019-02-05 Google Llc System and method for speech recognition using deep recurrent neural networks
CN109754113A (en) * 2018-11-29 2019-05-14 南京邮电大学 Load forecasting method based on dynamic time warping Yu length time memory
CN110188826A (en) * 2019-05-31 2019-08-30 湖南科技大学 Household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data
CN110619420A (en) * 2019-07-31 2019-12-27 广东工业大学 Attention-GRU-based short-term residential load prediction method

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10199038B1 (en) * 2012-11-29 2019-02-05 Google Llc System and method for speech recognition using deep recurrent neural networks
CN108510985A (en) * 2017-02-24 2018-09-07 百度（美国）有限责任公司 System and method for reducing the principle sexual deviation in production speech model
CN108416695A (en) * 2018-02-24 2018-08-17 合肥工业大学 Electric load probability density prediction technique based on deep learning and system, medium
CN109754113A (en) * 2018-11-29 2019-05-14 南京邮电大学 Load forecasting method based on dynamic time warping Yu length time memory
CN110188826A (en) * 2019-05-31 2019-08-30 湖南科技大学 Household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data
CN110619420A (en) * 2019-07-31 2019-12-27 广东工业大学 Attention-GRU-based short-term residential load prediction method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
徐雅芸等: "面向家居用户行为预测的BiGRU-DAtt 模型研究", pages 1 - 8, Retrieved from the Internet <URL:http://kns.cnki.net/kcms/detail/11.2127.tp.20190524.0848.014.html> *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112348789A (en) * 2020-11-03 2021-02-09 哈尔滨市科佳通用机电股份有限公司 Train lock catch loss fault detection method, system and device
CN113469412A (en) * 2021-06-02 2021-10-01 国核电力规划设计研究院有限公司 Real-time operation strategy optimization method and system for comprehensive energy system
CN113469412B (en) * 2021-06-02 2024-04-09 国核电力规划设计研究院有限公司 Real-time operation strategy optimization method and system for comprehensive energy system
CN113393025A (en) * 2021-06-07 2021-09-14 浙江大学 Non-invasive load decomposition method based on Informer model coding structure
CN113434493A (en) * 2021-06-28 2021-09-24 湘潭大学 Non-invasive load decomposition method based on Transformer
CN113595242A (en) * 2021-06-28 2021-11-02 湖南大学 Non-invasive load identification method based on deep CNN-HMM
CN113595242B (en) * 2021-06-28 2024-02-20 湖南大学 Non-invasive load identification method based on depth CNN-HMM
CN114037178A (en) * 2021-11-23 2022-02-11 天津大学 Non-invasive load decomposition method based on unsupervised pre-training neural network

Similar Documents

Publication Publication Date Title
CN111783684A (en) Bi-GRU-SA-based household electrical load decomposition method and system
Wu et al. An adaptive deep transfer learning method for bearing fault diagnosis
Angelis et al. NILM applications: Literature review of learning approaches, recent developments and challenges
Pan et al. Sequence-to-subsequence learning with conditional gan for power disaggregation
Munshi et al. Unsupervised nonintrusive extraction of electrical vehicle charging load patterns
CN112434799B (en) Non-invasive load identification method based on full convolution neural network
CN114723040A (en) LSTM and sequence-to-point translation model-based non-invasive load monitoring method
CN114037178A (en) Non-invasive load decomposition method based on unsupervised pre-training neural network
Humala et al. Universalnilm: A semi-supervised energy disaggregation framework using general appliance models
Javaid et al. Using GANCNN and ERNET for detection of non technical losses to secure smart grids
Jiang et al. Remaining useful life estimation combining two-step maximal information coefficient and temporal convolutional network with attention mechanism
CN110837934B (en) Smart grid short-term residential load prediction method based on deep learning
Dash et al. Attention-based multitask probabilistic network for nonintrusive appliance load monitoring
CN113762355B (en) User abnormal electricity behavior detection method based on non-invasive load decomposition
Han et al. Non‐intrusive load monitoring by voltage–current trajectory enabled asymmetric deep supervised hashing
Xie et al. Unsupervised abnormal detection using VAE with memory
Yang et al. An adaptive lightweight seq2subseq model for non‐intrusive load monitoring
CN117421638A (en) Load monitoring network model construction method, load monitoring method and system
CN116756575A (en) Non-invasive load decomposition method based on BGAIN-DD network
Wang et al. An Automatic Identification Framework for Complex Power Quality Disturbances Based on Ensemble CNN
Wang et al. Stockwell‐transform and random‐forest based double‐terminal fault diagnosis method for offshore wind farm transmission line
CN116008747A (en) Yogi-mLSTM cable partial discharge identification method and diagnosis system based on wavelet threshold denoising
CN115758273A (en) Method, device, equipment and medium for detecting time sequence data abnormity
CN113094931B (en) Non-invasive load decomposition method, device and equipment
Liu et al. Enhancing short-term wind power forecasting accuracy for reliable and safe integration into power systems: A gray relational analysis and optimized support vector regression machine approach

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