CN110188826A - Household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data - Google Patents
Household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data Download PDFInfo
- Publication number
- CN110188826A CN110188826A CN201910468488.3A CN201910468488A CN110188826A CN 110188826 A CN110188826 A CN 110188826A CN 201910468488 A CN201910468488 A CN 201910468488A CN 110188826 A CN110188826 A CN 110188826A
- Authority
- CN
- China
- Prior art keywords
- data
- lstm
- network
- matrix
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—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
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data, comprising the following steps: acquire the electricity consumption general power data of subscriber household and the consumption power data of pre- electricity measurer;Electric operation state is labeled, and the power data of acquisition is normalized;Build deep learning network model;The LSTM deep learning network built is trained, trained LSTM network model is obtained;Trained LSTM network model is tested, neural network forecast accuracy is examined;It acquires any family to register one's residence intelligent electric meter power data, as the input of LSTM network model, detection identifies the operating status of multiple household electrical appliance.The present invention has good effect to the running state recognition of load, compared with common single prediction network, the training time of network greatly reduces, and can also reach good prediction effect using transfer learning for different areas, has very high value in production and living.
Description
Technical field
The present invention relates to household electricity field, in particular to a kind of household electrical appliance based on intelligent electric meter data run shape
State non-invasive inspection methods.
Background technique
Energy problem is one of maximum challenge of facing mankind in recent decades, in development in science and technology, people's living standard
While raising, we also exceedingly develop earth resource.The use of electric power be energy consumption sharply increase it is very big because
Element shows according to american energy consumption data handbook, 40 percent primary energy consumption and 70% electric power resource consumption
From interior, have a high potential so saving indoor electric consumption.
Innovation Input and construction by many decades, national grid have basically reached wanting for ubiquitous Internet of Things in net side
How the data asked, but generated in the user terminal of power grid, nearly 400,000,000 intelligent electric meters, do not obtain sufficiently effective utilization
It is the service of electric power energy management value using these data, is the key that the ubiquitous electric power Internet of Things construction of development.Utilize intelligent electricity
The operating status of the main electrical appliance of family can be obtained using electric load decomposition technique in the household electricity power data of table acquisition
And detailed power consumption situation, demand side management can be carried out for electric system and analysis provides foundation, can such as provide in detail
Thin electricity charge inventory helps user's using electricity wisely and detection failure electric appliance, monitors behaviour to look after and is easy to happen danger
Crowd, and the accuracy of determination etc. for helping grid company to improve electric load distribution, to realize more effective electric power energy pipe
Reason and value service provide safeguard.
Electric load decomposition technique is divided into two major classes: intrusive electric load decomposes (Intrusive load
Decomposition, ILM) and non-intrusive electrical load decomposition (Non-intrusive load decomposition,
NILM).ILM refers to is installing sensor in power load, detects the power consumption of electric appliance.Although the identification of this method electric appliance is quasi-
True rate is higher, but it involves great expense, is easy to influence resident's daily life, not convenient for safeguarding.NILM refers on user's main circuit
Intelligent electric meter is installed and obtains the data such as general power and total current, the work of all kinds of electrical equipments of user is identified according to these information
State.Its advantages are safety economies, easy to spread, the disadvantage is that recognition accuracy is not high.In recent years, NILM becomes power load
The mainstream research direction in lotus decomposition direction.
NILM is earliest by Hart of the Massachusetts Institute of Technology et al. proposition, subsequent bayesian algorithm, Hidden Markov Model
(HMM), support vector machines (SVM), neural network model (ANN) scheduling algorithm are used for the field NILM.With rising abruptly for deep learning
It rises, convolutional neural networks (CNN) and Recognition with Recurrent Neural Network (RNN) are used for electric load and decompose field, and shot and long term remembers net
Network (LSTM) as an improvement after Recognition with Recurrent Neural Network, he, which can solve RNN and can not handle the dependence of long range, asks
Topic has good recognition effect to the data of timing input.The ordinal number when input data of NILM is the general power of household electricity
According to, modeled using the powerful memory function of LSTM network, for non-intrusive electrical load decompose NILM have it is good
Effect.
Summary of the invention
That in order to solve the above technical problem, the present invention provides a kind of algorithms is simple, detection accuracy is high based on intelligent electric meter
The household electrical appliance operating status non-invasive inspection methods of data.
Technical proposal that the invention solves the above-mentioned problems is: a kind of household electrical appliance operation shape based on intelligent electric meter data
State non-invasive inspection methods, comprising the following steps:
Step 1: acquisition data: acquiring the electricity consumption general power data of multiple subscriber households whithin a period of time and prediction electricity
The consumption power data of device, is divided into the data set of acquisition to obtain training set and test set;
Step 2: data mark and pretreatment: electric operation state is labeled according to the consumption power of electric appliance, and
The power data of acquisition is normalized;
Step 3: building deep learning network model: LSTM deep learning network is built using TensorFlow framework,
Input parameter, the network number of plies of adjustment network obtain optimal LSTM network model;
Step 4: trained and test deep learning network model: the training set generated using step 2 is to the LSTM built
Deep learning network is trained, and continues to optimize network, is improved network training precision, is obtained trained LSTM network model;
Step 5: testing trained LSTM network model using the test data set that step 2 generates, and calculates
Neural network forecast accuracy rate;
It registers one's residence intelligent electric meter power data Step 6: acquiring any family, as the input of LSTM network model, detection
Identify the operating status of multiple household electrical appliance.
The above-mentioned household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data, in the step 1,
Subscriber household electricity consumption general power data are acquired using home intelligent ammeter, the sample frequency of ammeter is 1/6Hz;In household electrical appliance
Upper installation intelligent socket, the consumption power data of M pre- electricity measurers of acquisition.
The above-mentioned household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data, in the step 2,
Electric operation state is labeled according to the consumption power of electric appliance, power threshold is arranged to pre- electricity measurer, will be more than threshold value
Consumption power is labeled as ' 1 ', is labeled as ' 0 ' lower than threshold value;The power data of acquisition is pre-processed, according to power number
According to mark state electric operation state is set, the data of training set test set are normalized, while making data
Format meets the input/output format of LSTM network model;
Normalized handles general power data, specific formula using deviation standardized method are as follows:
Wherein xiFor i-th of value of certain general power data, m is the quantity of general power data, xminFor all general powers
Minimum value, xmaxFor the maximum value of all general powers, xi' for normalization after data.
The above-mentioned household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data, in the step 3,
It is first directed to processed general power data and correlation data, network hyper parameter is set;Then LSTM neural network is built,
In LSTM neural network hidden layer, each LSTM neuron is respectively there are three door is controlled: forgeing door, input gate and output
Door;Input variable enters in the neuron of LSTM the operation for passing through three doors:
ft=σ (Wf×[Ht-1,Xt']+bf)
In formula: ftThe output of door, X are forgotten for t momentt' indicate t moment LSTM neural network input matrix, Ht-1It indicates
The output of t-1 moment entire neuron, bracket indicate that two vectors are connected and merge that σ is sigmoid function, WfTo forget door
Weight, bfFor the bias matrix for forgeing door;
ht=σ (Wh×[Ht-1,Xt']+bh)
In formula: htIndicate the output of t moment input gate,Indicate the candidate matrices of input gate, tanh indicates tanh
Function, Wh、bhIt indicates to calculate htWeight and bias matrix, Wc、bcIt indicates to calculateWeight and bias matrix;
The location mode C of t momenttThe output of door and the product of last moment state are forgotten plus the defeated of input gate for t moment
Out with the product of candidate matrices:
ot=σ (Wo×[Ht-1,Xt']+bo)
In formula: otFor the output of t moment out gate, WoFor the bias matrix of out gate, boFor the weight matrix of out gate;
The output of the entire neuron of t moment are as follows:
Ht=ot×tanh(Ct)
Each LSTM neuron t moment input variable are as follows: t moment input matrix Xt', the location mode C at t-1 momentt-1
With the output H at t-1 momentt-1, each LSTM neuron t moment output are as follows: location mode CtH is exported with neuront。
The above-mentioned household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data, in the step 4,
According to the output H of entire LSTM neural network hidden layert' come calculate prediction value matrix Y 'M:
Y'M=σ (H 't×Wu+bu)
In formula: Y'MFor neural network forecast value matrix, WuFor the bias matrix of output layer, buFor the weight matrix of output layer;
Pass through Y'MWith comparison matrix YMComparison is done to calculate the prediction error of LSTM neural network, select loss function and
Optimizer, the LSTM neural network deconditioning after loss of network drops to 0.0001, and save LSTM neural network.
The above-mentioned household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data, in the step 4,
Loss function is selected as the cross entropy loss function of sigmoid, formula are as follows:
Optimizer is selected as Adam algorithm optimization device.
The above-mentioned household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data, the step 5 tool
Body step are as follows:
In LSTM neural network output layer, the neuron of the second hidden layer is mapped into M output, entire LSTM nerve net
The output valve of network are as follows:
yi′'=Ht×W2+b2, i'=1,2 ..., M
In formula: yi′' be LSTM neural network network single output, W2For the weight matrix of the second hidden layer, b2It is
The bias matrix of two hidden layers;
The prediction value matrix Y' of entire LSTM neural networkMAre as follows:
Y'M=[y '1,y'2,...,y'M]
By the prediction value matrix Y' of LSTM neural networkM, pass through function:
Prediction result is become ' 0 ', ' 1 ' matrix and comparison matrix YMIt compares, calculates the standard of LSTM neural network prediction
True rate, to examine the prediction effect of network.
The above-mentioned household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data, to different areas,
The power of electric appliance and operating status will be different due to weather and people's living habit, if directly by LSTM nerve
Network model is applied to the pre- electricity measurer in different regions, and prediction result will not known, if resurveying data to train new net
Network, it will a large amount of manpower and material resources are expended, in this regard, coming the bad area of prediction effect using the method for transfer learning, it may be assumed that use
A small amount of household electricity general power data and pre- electricity measurer in home intelligent ammeter and intelligent socket acquisition prediction area consume power
Then data are labeled and pre-process to data, carry out transfer learning to trained LSTM network with low volume data, instruction
White silk finishes the electric operation state that can predict this area.
The beneficial effects of the present invention are: the household electrical appliance proposed by the invention based on intelligent electric meter data run shape
State non-invasive inspection methods have good effect to the running state recognition of load, with common single prediction network phase
Than the present invention can disposably predict the operating status of multiple electric appliances, and during electric appliance prediction, one family is often needed
Predict multiple electric appliances, some is even had tens kinds, is predicted using single network, and data processing, network establishment take time and effort,
And method of the invention is used, data handling procedure will simplify, not need to be handled one by one according to each electric appliance, can be with
Multiple electric appliance batch processings, the training time of network greatly reduce, and can also reach good using transfer learning for different areas
Good prediction effect has very high value in production and living.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Fig. 2 is the structural schematic diagram of LSTM neural network of the invention.
Fig. 3 is the LSTM network establishment figure under tensorflow frame of the invention.
Specific embodiment
The present invention is further illustrated with reference to the accompanying drawings and examples.
As shown in Figure 1, a kind of household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data, packet
Include following steps:
Step 1: acquisition data.
Subscriber household electricity consumption general power data are acquired using home intelligent ammeter, the sample frequency of ammeter is 1Hz;It is in
Intelligent socket is installed on electrical appliance, acquires the consumption power data of pre- electricity measurer.The present embodiment acquires altogether 4 different electricity
Device consumes power, is hot-water bottle, micro-wave oven, dish-washing machine and washing machine respectively, and intelligent socket sample frequency is 1/6Hz.It has recorded
The power consumption data of 1 family, record total time is more than 20 days, and total amount of data 1700000, the text of data CSV format
Part saves, and is divided into the data set of acquisition to obtain training set and test set.
Step 2: data mark and pretreatment.
Electric operation state is labeled according to the consumption power of electric appliance, to hot-water bottle, micro-wave oven, dish-washing machine and is washed
Power threshold is arranged in clothing machine, is 2000W, 200W, 10W and 20W respectively, will be more than that the consumption power of threshold value is labeled as ' 1 ', low
' 0 ' is labeled as in threshold value;Then the data marked are handled, adjustment general power data frequency to 1/6Hz uses
Sliding window extracts general power and appliance power data, and data is made to be concentrated with 50% data electric appliance in a situation of use
Data.100,000 datas are extracted altogether, and every data has the sampled point of 100 general powers and the operating status of 400 different electric appliances
Sampled point (100 sampled points of each electric appliance), i.e., 10 minutes sampled datas, it includes general power data and 4 in 10 minutes
The running state data of a electric appliance.In 100 sampled points of each electric appliance, if occurring continuous 3 ' 1 ', by this number of segment
It according to operating status is identified as, is indicated with ' 1 ', if not occurring, is identified as not running state, is indicated with ' 0 '.Every right
Than the combination that data are 4 ' 0 ' and ' 1 ', correlation data total number is identical with general power number of data.By training set test
The data of collection are normalized, while data format being made to meet the input/output format of LSTM network model;
Normalized handles general power data, specific formula using deviation standardized method are as follows:
Wherein xiFor i-th of value of certain general power data, m is the quantity of general power data, xminFor all general powers
Minimum value, xmaxFor the maximum value of all general powers, xi' for normalization after data.
If training data shares 100, input data matrix are as follows:
Xi'=[x1',x2',...,x100'];
Processed general power data set and correlation data collection are saved to csv file.
Step 3: building deep learning network model: LSTM deep learning network is built using TensorFlow framework,
Input parameter, the network number of plies of adjustment network obtain optimal LSTM network model.Detailed process are as follows:
It is first directed to processed general power data and correlation data, network hyper parameter is set, every 50 data is one
Batch is inputted, every data inputs 100 total power values, exports 4 predicted values, learning rate 0.0001, network concealed layer is
2 layers, first layer has 128 neurons, and the second layer has 256 neurons;Then LSTM neural network, LSTM nerve net are built
In network hidden layer, each LSTM neuron is respectively there are three door is controlled: forgeing door, input gate and out gate;Input variable
Into in the neuron of LSTM pass through three doors operation:
ft=σ (Wf×[Ht-1,Xt']+bf)
In formula: ftThe output of door, X are forgotten for t momentt' indicate t moment LSTM neural network input matrix, Ht-1It indicates
The output of t-1 moment entire neuron, bracket indicate that two vectors are connected and merge that σ is sigmoid function, WfTo forget door
Weight, bfFor the bias matrix for forgeing door;
ht=σ (Wh×[Ht-1,Xt']+bh)
In formula: htIndicate the output of t moment input gate,Indicate the candidate matrices of input gate, tanh indicates tanh
Function, Wh、bhIt indicates to calculate htWeight and bias matrix, Wc、bcIt indicates to calculateWeight and bias matrix;t
The location mode C at momenttOutput and candidate for the output of t moment forgetting door and the product of last moment state plus input gate
The product of matrix:
ot=σ (Wo×[Ht-1,Xt']+bo)
In formula: otFor the output of t moment out gate, WoFor the bias matrix of out gate, boFor the weight matrix of out gate;
The output of the entire neuron of t moment are as follows:
Ht=ot×tanh(Ct)
Each LSTM neuron t moment input variable are as follows: t moment input matrix Xt', the location mode C at t-1 momentt-1
With the output H at t-1 momentt-1, each LSTM neuron t moment output are as follows: location mode CtH is exported with neuront。
Step 4: trained and test deep learning network model: the training set generated using step 2 is to the LSTM built
Deep learning network is trained, and continues to optimize network, is improved network training precision, is obtained trained LSTM network model.
Detailed process are as follows:
According to the output H of entire LSTM neural network hidden layert' come calculate prediction value matrix Y'M:
Y'M=σ (H 't×Wu+bu)
In formula: Y'MFor neural network forecast value matrix, WuFor the bias matrix of output layer, buFor the weight matrix of output layer;
Pass through Y'MWith comparison matrix YMComparison is done to calculate the prediction error of LSTM neural network, select loss function and
Optimizer, the LSTM neural network deconditioning after loss of network drops to 0.0001, and save LSTM neural network.
Loss function is selected as the cross entropy loss function of sigmoid, formula are as follows:
Optimizer is selected as Adam algorithm optimization device.
Step 5: testing trained LSTM network model using the test data set that step 2 generates, and calculates
Neural network forecast accuracy rate.Specific steps are as follows:
In LSTM neural network output layer, the neuron of the second hidden layer is mapped into 4 outputs, entire LSTM nerve net
The output valve of network are as follows:
yi′'=Ht×W2+b2, i'=1,2 ..., 4
In formula: yi′' be LSTM neural network network single output, W2For the weight matrix of the second hidden layer, b2It is
The bias matrix of two hidden layers;
The prediction value matrix Y ' of entire LSTM neural network4Are as follows:
Y′4=[y '1,y'2,...,y'4]
By the prediction value matrix Y ' of LSTM neural network4, pass through function:
Prediction result is become ' 0 ', ' 1 ' matrix and comparison matrix YMIt compares, calculates the standard of LSTM neural network prediction
True rate, to examine the prediction effect of network.
It registers one's residence intelligent electric meter power data Step 6: acquiring any family, as the input of LSTM network model, detection
Identify the operating status of multiple household electrical appliance such as hot-water bottle, micro-wave oven, dryer and washing machine.Table 1 is LSTM in the present embodiment
The predictablity rate of neural network.
Table 1
Predictablity rate | |
Hot-water bottle | 95.6% |
Micro-wave oven | 91.8% |
Dish-washing machine | 87.6% |
Washing machine | 92.7% |
To different areas, the power of electric appliance and operating status can be due to weather and people's living habit
Difference, if LSTM neural network model is directly applied to the pre- electricity measurer in different regions, prediction result will not known, if again
Data are acquired to train new network, it will a large amount of manpower and material resources are expended, in this regard, predicting using the method for transfer learning
The bad area of effect, it may be assumed that with a small amount of household electricity general power number of home intelligent ammeter and intelligent socket acquisition prediction area
Power data is consumed according to pre- electricity measurer, then data are labeled and are pre-processed, with low volume data to trained LSTM
Network carries out transfer learning, and training finishes the electric operation state that can predict this area, and table 2 is LSTM in the present embodiment
The training time table of neural network model.
Table 2
Single output network superposition | Multi output network | |
Training time (min) | 1416.8 | 347.86 |
As can be seen from the results it is proposed by the invention based on the Household appliance switch recognition methods of LSTM neural network to load
Running state recognition have good effect.Compared with common single prediction network, the training time of network subtracts significantly
It is few, good prediction effect can also be reached using transfer learning for different areas, this network has very in production and living
High value.
Claims (8)
1. a kind of household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data, comprising the following steps:
Step 1: acquisition data: acquiring multiple subscriber households electricity consumption general power data whithin a period of time and pre- electricity measurer
Power data is consumed, is divided into the data set of acquisition to obtain training set and test set;
Step 2: data mark and pretreatment: being labeled according to the consumption power of electric appliance to electric operation state, and to acquisition
Power data be normalized;
Step 3: building deep learning network model: building LSTM deep learning network using TensorFlow framework, adjust net
Input parameter, the network number of plies of network obtain optimal LSTM network model;
Step 4: trained and test deep learning network model: the training set generated using step 2 is to the LSTM depth built
Learning network is trained, and continues to optimize network, is improved network training precision, is obtained trained LSTM network model;
Step 5: testing trained LSTM network model using the test data set that step 2 generates, and calculates network
Predictablity rate;
It is more to detect identification Step 6: acquiring any family and registering one's residence intelligent electric meter power data as the input of LSTM network model
The operating status of a household electrical appliance.
2. the household electrical appliance operating status non-invasive inspection methods according to claim 1 based on intelligent electric meter data,
It is characterized in that, acquiring subscriber household electricity consumption general power data, the sampling of ammeter using home intelligent ammeter in the step 1
Frequency is 1/6Hz;Intelligent socket, the consumption power data of M pre- electricity measurers of acquisition are installed on household appliances.
3. the household electrical appliance operating status non-invasive inspection methods according to claim 1 based on intelligent electric meter data,
It is characterized in that, being labeled according to the consumption power of electric appliance to electric operation state, being set to pre- electricity measurer in the step 2
Power threshold is set, the consumption power more than threshold value is labeled as ' 1 ', is labeled as ' 0 ' lower than threshold value;To the power data of acquisition
It is pre-processed, electric operation state is arranged according to the mark state of power data, the data of training set test set are carried out
Normalized, while data format being made to meet the input/output format of LSTM network model;
Normalized handles general power data, specific formula using deviation standardized method are as follows:
Wherein xiFor i-th of value of certain general power data, m is the quantity of general power data, xminFor the minimum of all general powers
Value, xmaxFor the maximum value of all general powers, xi' for normalization after data.
4. the household electrical appliance operating status non-invasive inspection methods according to claim 3 based on intelligent electric meter data,
It is characterized in that, being first directed to processed general power data and correlation data in the step 3, network hyper parameter is set;
Then LSTM neural network is built, in LSTM neural network hidden layer, each LSTM neuron is respectively there are three door is controlled:
Forget door, input gate and out gate;Input variable enters in the neuron of LSTM the operation for passing through three doors:
ft=σ (Wf×[Ht-1,Xt']+bf)
In formula: ftThe output of door, X are forgotten for t momentt' indicate t moment LSTM neural network input matrix, Ht-1When indicating t-1
The output of entire neuron is carved, bracket indicates that two vectors are connected and merge that σ is sigmoid function, WfFor the power for forgeing door
Weight, bfFor the bias matrix for forgeing door;
ht=σ (Wh×[Ht-1,Xt']+bh)
In formula: htIndicate the output of t moment input gate,Indicating the candidate matrices of input gate, tanh indicates hyperbolic tangent function,
Wh、bhIt indicates to calculate htWeight and bias matrix, Wc、bcIt indicates to calculateWeight and bias matrix;
The location mode C of t momenttThe output and time of input gate are added for the output of t moment forgetting door and the product of last moment state
Select the product of matrix:
ot=σ (Wo×[Ht-1,Xt']+bo)
In formula: otFor the output of t moment out gate, WoFor the bias matrix of out gate, boFor the weight matrix of out gate;
The output of the entire neuron of t moment are as follows:
Ht=ot×tanh(Ct)
Each LSTM neuron t moment input variable are as follows: t moment input matrix Xt', the location mode C at t-1 momentt-1When with t-1
The output H at quartert-1, each LSTM neuron t moment output are as follows: location mode CtH is exported with neuront。
5. the household electrical appliance operating status non-invasive inspection methods according to claim 4 based on intelligent electric meter data,
It is characterized in that, in the step 4, according to the output H of entire LSTM neural network hidden layert' calculate prediction value matrix
Y'M:
Y'M=σ (H 't×Wu+bu)
In formula: Y'MFor neural network forecast value matrix, WuFor the bias matrix of output layer, buFor the weight matrix of output layer;
Pass through Y'MWith comparison matrix YMComparison is done to calculate the prediction error of LSTM neural network, selects loss function and optimization
Device, the LSTM neural network deconditioning after loss of network drops to 0.0001, and save LSTM neural network.
6. the household electrical appliance operating status non-invasive inspection methods according to claim 4 based on intelligent electric meter data,
It is characterized in that, loss function is selected as the cross entropy loss function of sigmoid, formula in the step 4 are as follows:
Optimizer is selected as Adam algorithm optimization device.
7. the household electrical appliance operating status non-invasive inspection methods according to claim 6 based on intelligent electric meter data,
It is characterized in that, the step 5 specific steps are as follows:
In LSTM neural network output layer, the neuron of the second hidden layer is mapped M and is exported, entire LSTM neural network it is defeated
It is worth out are as follows:
yi′'=Ht×W2+b2, i'=1,2 ..., M
In formula: yi′' be LSTM neural network network single output, W2For the weight matrix of the second hidden layer, b2It is hidden for second
The bias matrix of layer;
The prediction value matrix Y' of entire LSTM neural networkMAre as follows:
Y'M=[y '1,y'2,...,y'M]
By the prediction value matrix Y' of LSTM neural networkM, pass through function:
Prediction result is become ' 0 ', ' 1 ' matrix and comparison matrix YMIt compares, calculates the accuracy rate of LSTM neural network prediction,
To examine the prediction effect of network.
8. the household electrical appliance operating status non-invasive inspection methods according to claim 7 based on intelligent electric meter data,
It is characterized in that, the power of electric appliance and operating status have due to weather and people's living habit to different areas
Institute is different, if LSTM neural network model is directly applied to the pre- electricity measurer in different regions, prediction result will not known, if again
Data are acquired to train new network, it will a large amount of manpower and material resources are expended, in this regard, predicting to imitate using the method for transfer learning
The bad area of fruit, it may be assumed that with a small amount of household electricity general power data of home intelligent ammeter and intelligent socket acquisition prediction area
Power data is consumed with pre- electricity measurer, then data are labeled and are pre-processed, with low volume data to trained LSTM net
Network carries out transfer learning, and training finishes the electric operation state that can predict this area.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910468488.3A CN110188826A (en) | 2019-05-31 | 2019-05-31 | Household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910468488.3A CN110188826A (en) | 2019-05-31 | 2019-05-31 | Household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110188826A true CN110188826A (en) | 2019-08-30 |
Family
ID=67719301
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910468488.3A Pending CN110188826A (en) | 2019-05-31 | 2019-05-31 | Household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110188826A (en) |
Cited By (33)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110852655A (en) * | 2019-11-29 | 2020-02-28 | 湖南科技大学 | Household power load decomposition system with solar power supply system and decomposition method |
CN111126489A (en) * | 2019-12-24 | 2020-05-08 | 广东电网有限责任公司 | Power transmission equipment state evaluation method based on ensemble learning |
CN111242391A (en) * | 2020-03-06 | 2020-06-05 | 云南电网有限责任公司电力科学研究院 | Machine learning model training method and system for power load identification |
CN111338262A (en) * | 2020-03-20 | 2020-06-26 | 湖南科技大学 | ARM-based non-invasive detection device and method for running state of household appliance |
CN111415270A (en) * | 2020-03-03 | 2020-07-14 | 浙江万胜智能科技股份有限公司 | Power load intelligent identification method based on L STM learning |
CN111551773A (en) * | 2020-05-14 | 2020-08-18 | 资明 | Intelligent power management system |
CN111556107A (en) * | 2020-04-16 | 2020-08-18 | 福州和达电子科技有限公司 | Intelligent Internet of things application method based on stress-reflex model |
CN111563827A (en) * | 2020-04-10 | 2020-08-21 | 杭州电子科技大学 | Load decomposition method based on electrical appliance physical characteristics and residential electricity consumption behaviors |
CN111783684A (en) * | 2020-07-03 | 2020-10-16 | 湖南大学 | Bi-GRU-SA-based household electrical load decomposition method and system |
CN111882041A (en) * | 2020-07-31 | 2020-11-03 | 国网重庆市电力公司电力科学研究院 | Power grid attack detection method and device based on improved RNN (neural network) |
CN111917184A (en) * | 2020-05-19 | 2020-11-10 | 国网山东省电力公司济宁供电公司 | Power consumption monitoring method and system based on BP neural network |
CN112001486A (en) * | 2020-08-28 | 2020-11-27 | 河北工业大学 | Load decomposition method based on deep learning |
CN112098715A (en) * | 2020-09-18 | 2020-12-18 | 宁波智诚祥科技发展有限公司 | Electric energy monitoring and early warning system based on 5G and corrected GCN diagram neural network |
CN112365090A (en) * | 2020-11-26 | 2021-02-12 | 南方电网能源发展研究院有限责任公司 | Deep learning-based non-invasive electrical load identification method and device |
CN112418623A (en) * | 2020-11-12 | 2021-02-26 | 国网河南省电力公司郑州供电公司 | Anti-electricity-stealing identification method based on bidirectional long-time and short-time memory network and sliding window input |
CN112434799A (en) * | 2020-12-18 | 2021-03-02 | 宁波迦南智能电气股份有限公司 | Non-invasive load identification method based on full convolution neural network |
CN112653241A (en) * | 2020-12-16 | 2021-04-13 | 南方电网深圳数字电网研究院有限公司 | Power grid monitoring method and device and Internet of things metering unit |
CN112926645A (en) * | 2021-02-22 | 2021-06-08 | 国网四川省电力公司营销服务中心 | Electricity stealing detection method based on edge calculation |
CN113011482A (en) * | 2021-03-10 | 2021-06-22 | 威胜集团有限公司 | Non-invasive load identification method, terminal device and storage medium |
CN113219272A (en) * | 2021-04-06 | 2021-08-06 | 国网江苏省电力有限公司营销服务中心 | Method and system for predicting household appliance aging based on non-invasive monitoring |
CN113363979A (en) * | 2021-06-23 | 2021-09-07 | 中国科学技术大学 | Non-contact power load decomposition sensing method and system |
CN113434493A (en) * | 2021-06-28 | 2021-09-24 | 湘潭大学 | Non-invasive load decomposition method based on Transformer |
CN113568383A (en) * | 2021-06-30 | 2021-10-29 | 宁波三星医疗电气股份有限公司 | Non-invasive identification method suitable for machining center |
CN113837894A (en) * | 2021-08-06 | 2021-12-24 | 国网江苏省电力有限公司南京供电分公司 | Non-invasive resident user load decomposition method based on residual convolution module |
CN113920375A (en) * | 2021-11-01 | 2022-01-11 | 国网新疆电力有限公司营销服务中心(资金集约中心、计量中心) | Fusion characteristic typical load recognition method and system based on combination of Faster R-CNN and SVM |
CN113962314A (en) * | 2021-10-27 | 2022-01-21 | 南京富尔登科技发展有限公司 | Non-invasive enterprise load decomposition method based on federal learning |
CN113971371A (en) * | 2021-10-29 | 2022-01-25 | 金程科技有限公司 | Method for identifying and safely managing and controlling end power grid user equipment |
CN114065653A (en) * | 2022-01-17 | 2022-02-18 | 南方电网数字电网研究院有限公司 | Construction method of power load prediction model and power load prediction method |
CN114118568A (en) * | 2021-11-19 | 2022-03-01 | 国网河南省电力公司经济技术研究院 | Multi-attention-based power consumption prediction method for convolutional neural network |
CN115018217A (en) * | 2022-08-09 | 2022-09-06 | 国网山东省电力公司东营市河口区供电公司 | Photovoltaic transmission management method and system |
CN115112989A (en) * | 2022-08-29 | 2022-09-27 | 四川大学 | Non-invasive load monitoring method based on low-frequency data |
CN117807528A (en) * | 2023-12-29 | 2024-04-02 | 中科南京信息高铁研究院 | Non-invasive household appliance state identification method and medium based on long-short-term memory network |
CN113962314B (en) * | 2021-10-27 | 2024-07-05 | 南京富尔登科技发展有限公司 | Non-invasive enterprise load decomposition method based on federal learning |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107122790A (en) * | 2017-03-15 | 2017-09-01 | 华北电力大学 | Non-intrusion type load recognizer based on hybrid neural networks and integrated study |
CN107239852A (en) * | 2017-05-05 | 2017-10-10 | 南京邮电大学 | A kind of electric quantity consumption Forecasting Methodology based on deep learning |
CN107730003A (en) * | 2017-10-23 | 2018-02-23 | 华中科技大学 | One kind supports more high-precision NILM implementation methods of appliance type |
CN108280551A (en) * | 2018-02-02 | 2018-07-13 | 华北电力大学 | A kind of photovoltaic power generation power prediction method using shot and long term memory network |
CN109245099A (en) * | 2018-10-29 | 2019-01-18 | 南方电网科学研究院有限责任公司 | A kind of electric load discrimination method, device, equipment and readable storage medium storing program for executing |
CN109614885A (en) * | 2018-11-21 | 2019-04-12 | 齐鲁工业大学 | A kind of EEG signals Fast Classification recognition methods based on LSTM |
CA3017229A1 (en) * | 2017-10-20 | 2019-04-20 | Institute For Information Industry | Power consumption analyzing server and power consumption analyzing method thereof |
CN109685252A (en) * | 2018-11-30 | 2019-04-26 | 西安工程大学 | Building energy consumption prediction technique based on Recognition with Recurrent Neural Network and multi-task learning model |
-
2019
- 2019-05-31 CN CN201910468488.3A patent/CN110188826A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107122790A (en) * | 2017-03-15 | 2017-09-01 | 华北电力大学 | Non-intrusion type load recognizer based on hybrid neural networks and integrated study |
CN107239852A (en) * | 2017-05-05 | 2017-10-10 | 南京邮电大学 | A kind of electric quantity consumption Forecasting Methodology based on deep learning |
CA3017229A1 (en) * | 2017-10-20 | 2019-04-20 | Institute For Information Industry | Power consumption analyzing server and power consumption analyzing method thereof |
CN107730003A (en) * | 2017-10-23 | 2018-02-23 | 华中科技大学 | One kind supports more high-precision NILM implementation methods of appliance type |
CN108280551A (en) * | 2018-02-02 | 2018-07-13 | 华北电力大学 | A kind of photovoltaic power generation power prediction method using shot and long term memory network |
CN109245099A (en) * | 2018-10-29 | 2019-01-18 | 南方电网科学研究院有限责任公司 | A kind of electric load discrimination method, device, equipment and readable storage medium storing program for executing |
CN109614885A (en) * | 2018-11-21 | 2019-04-12 | 齐鲁工业大学 | A kind of EEG signals Fast Classification recognition methods based on LSTM |
CN109685252A (en) * | 2018-11-30 | 2019-04-26 | 西安工程大学 | Building energy consumption prediction technique based on Recognition with Recurrent Neural Network and multi-task learning model |
Cited By (45)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110852655A (en) * | 2019-11-29 | 2020-02-28 | 湖南科技大学 | Household power load decomposition system with solar power supply system and decomposition method |
CN110852655B (en) * | 2019-11-29 | 2023-08-15 | 湖南科技大学 | Household electricity load decomposition system with solar power supply system and household electricity load decomposition method |
CN111126489A (en) * | 2019-12-24 | 2020-05-08 | 广东电网有限责任公司 | Power transmission equipment state evaluation method based on ensemble learning |
CN111415270A (en) * | 2020-03-03 | 2020-07-14 | 浙江万胜智能科技股份有限公司 | Power load intelligent identification method based on L STM learning |
CN111242391A (en) * | 2020-03-06 | 2020-06-05 | 云南电网有限责任公司电力科学研究院 | Machine learning model training method and system for power load identification |
CN111338262A (en) * | 2020-03-20 | 2020-06-26 | 湖南科技大学 | ARM-based non-invasive detection device and method for running state of household appliance |
CN111338262B (en) * | 2020-03-20 | 2021-11-19 | 湖南科技大学 | ARM-based non-invasive detection device and method for running state of household appliance |
CN111563827B (en) * | 2020-04-10 | 2023-08-04 | 杭州电子科技大学 | Load decomposition method based on physical characteristics of electrical appliance and electricity utilization behaviors of residents |
CN111563827A (en) * | 2020-04-10 | 2020-08-21 | 杭州电子科技大学 | Load decomposition method based on electrical appliance physical characteristics and residential electricity consumption behaviors |
CN111556107A (en) * | 2020-04-16 | 2020-08-18 | 福州和达电子科技有限公司 | Intelligent Internet of things application method based on stress-reflex model |
CN111551773A (en) * | 2020-05-14 | 2020-08-18 | 资明 | Intelligent power management system |
CN111917184A (en) * | 2020-05-19 | 2020-11-10 | 国网山东省电力公司济宁供电公司 | Power consumption monitoring method and system based on BP neural network |
CN111783684A (en) * | 2020-07-03 | 2020-10-16 | 湖南大学 | Bi-GRU-SA-based household electrical load decomposition method and system |
CN111882041A (en) * | 2020-07-31 | 2020-11-03 | 国网重庆市电力公司电力科学研究院 | Power grid attack detection method and device based on improved RNN (neural network) |
CN112001486B (en) * | 2020-08-28 | 2022-07-22 | 河北工业大学 | Load decomposition method based on deep learning |
CN112001486A (en) * | 2020-08-28 | 2020-11-27 | 河北工业大学 | Load decomposition method based on deep learning |
CN112098715B (en) * | 2020-09-18 | 2023-06-06 | 宁波智诚祥科技发展有限公司 | Electric energy monitoring and early warning system based on 5G and correction GCN graph neural network |
CN112098715A (en) * | 2020-09-18 | 2020-12-18 | 宁波智诚祥科技发展有限公司 | Electric energy monitoring and early warning system based on 5G and corrected GCN diagram neural network |
CN112418623A (en) * | 2020-11-12 | 2021-02-26 | 国网河南省电力公司郑州供电公司 | Anti-electricity-stealing identification method based on bidirectional long-time and short-time memory network and sliding window input |
CN112365090A (en) * | 2020-11-26 | 2021-02-12 | 南方电网能源发展研究院有限责任公司 | Deep learning-based non-invasive electrical load identification method and device |
CN112653241A (en) * | 2020-12-16 | 2021-04-13 | 南方电网深圳数字电网研究院有限公司 | Power grid monitoring method and device and Internet of things metering unit |
CN112434799A (en) * | 2020-12-18 | 2021-03-02 | 宁波迦南智能电气股份有限公司 | Non-invasive load identification method based on full convolution neural network |
CN112434799B (en) * | 2020-12-18 | 2022-07-05 | 宁波迦南智能电气股份有限公司 | Non-invasive load identification method based on full convolution neural network |
CN112926645A (en) * | 2021-02-22 | 2021-06-08 | 国网四川省电力公司营销服务中心 | Electricity stealing detection method based on edge calculation |
CN113011482A (en) * | 2021-03-10 | 2021-06-22 | 威胜集团有限公司 | Non-invasive load identification method, terminal device and storage medium |
CN113011482B (en) * | 2021-03-10 | 2023-06-20 | 威胜集团有限公司 | Non-invasive load identification method, terminal device and storage medium |
CN113219272A (en) * | 2021-04-06 | 2021-08-06 | 国网江苏省电力有限公司营销服务中心 | Method and system for predicting household appliance aging based on non-invasive monitoring |
CN113363979A (en) * | 2021-06-23 | 2021-09-07 | 中国科学技术大学 | Non-contact power load decomposition sensing method and system |
CN113363979B (en) * | 2021-06-23 | 2022-12-30 | 中国科学技术大学 | Non-contact power load decomposition sensing method and system |
CN113434493A (en) * | 2021-06-28 | 2021-09-24 | 湘潭大学 | Non-invasive load decomposition method based on Transformer |
CN113568383A (en) * | 2021-06-30 | 2021-10-29 | 宁波三星医疗电气股份有限公司 | Non-invasive identification method suitable for machining center |
CN113837894B (en) * | 2021-08-06 | 2023-12-19 | 国网江苏省电力有限公司南京供电分公司 | Non-invasive resident user load decomposition method based on residual convolution module |
CN113837894A (en) * | 2021-08-06 | 2021-12-24 | 国网江苏省电力有限公司南京供电分公司 | Non-invasive resident user load decomposition method based on residual convolution module |
CN113962314A (en) * | 2021-10-27 | 2022-01-21 | 南京富尔登科技发展有限公司 | Non-invasive enterprise load decomposition method based on federal learning |
CN113962314B (en) * | 2021-10-27 | 2024-07-05 | 南京富尔登科技发展有限公司 | Non-invasive enterprise load decomposition method based on federal learning |
CN113971371A (en) * | 2021-10-29 | 2022-01-25 | 金程科技有限公司 | Method for identifying and safely managing and controlling end power grid user equipment |
CN113971371B (en) * | 2021-10-29 | 2023-11-07 | 金程科技有限公司 | Terminal power grid user equipment identification and safety control method |
CN113920375A (en) * | 2021-11-01 | 2022-01-11 | 国网新疆电力有限公司营销服务中心(资金集约中心、计量中心) | Fusion characteristic typical load recognition method and system based on combination of Faster R-CNN and SVM |
CN114118568A (en) * | 2021-11-19 | 2022-03-01 | 国网河南省电力公司经济技术研究院 | Multi-attention-based power consumption prediction method for convolutional neural network |
CN114065653A (en) * | 2022-01-17 | 2022-02-18 | 南方电网数字电网研究院有限公司 | Construction method of power load prediction model and power load prediction method |
CN115018217B (en) * | 2022-08-09 | 2022-10-25 | 国网山东省电力公司东营市河口区供电公司 | Photovoltaic transmission management method and system |
CN115018217A (en) * | 2022-08-09 | 2022-09-06 | 国网山东省电力公司东营市河口区供电公司 | Photovoltaic transmission management method and system |
CN115112989B (en) * | 2022-08-29 | 2022-11-22 | 四川大学 | Non-invasive load monitoring method based on low-frequency data |
CN115112989A (en) * | 2022-08-29 | 2022-09-27 | 四川大学 | Non-invasive load monitoring method based on low-frequency data |
CN117807528A (en) * | 2023-12-29 | 2024-04-02 | 中科南京信息高铁研究院 | Non-invasive household appliance state identification method and medium based on long-short-term memory network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110188826A (en) | Household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data | |
Liu et al. | Energy disaggregation of appliances consumptions using ham approach | |
CN109598451B (en) | Non-invasive load identification method based on PCA (principal component analysis) and LSTM (least Square TM) neural network | |
CN106936129B (en) | Power load identification method and system based on multi-feature fusion | |
Lin et al. | Development of an improved time–frequency analysis-based nonintrusive load monitor for load demand identification | |
CN105429135B (en) | The identification decision-making technique and system that a kind of non-intrusive electrical load decomposes | |
Ma et al. | Toward energy-awareness smart building: Discover the fingerprint of your electrical appliances | |
CN109345409A (en) | A kind of resident's comprehensive energy efficiency management method based on bandwidth carrier | |
Yang et al. | Sequence-to-point learning based on temporal convolutional networks for nonintrusive load monitoring | |
CN113837894B (en) | Non-invasive resident user load decomposition method based on residual convolution module | |
Devlin et al. | Non-intrusive load monitoring using electricity smart meter data: A deep learning approach | |
Lemes et al. | Load disaggregation based on time window for HEMS application | |
CN111563827B (en) | Load decomposition method based on physical characteristics of electrical appliance and electricity utilization behaviors of residents | |
CN110309984B (en) | Non-invasive load identification and short-term user behavior prediction method | |
CN111415270A (en) | Power load intelligent identification method based on L STM learning | |
de Diego-Otón et al. | Recurrent LSTM architecture for appliance identification in non-intrusive load monitoring | |
CN109347584A (en) | Frequency spectrum sensing method based on Smoothing Prediction | |
Humala et al. | Universalnilm: A semi-supervised energy disaggregation framework using general appliance models | |
CN114037178A (en) | Non-invasive load decomposition method based on unsupervised pre-training neural network | |
CN112308270A (en) | Long-term electricity load prediction method and device and computer implementation system | |
Wu et al. | Nonintrusive on-site load-monitoring method with self-adaption | |
Delfosse et al. | Deep learning applied to NILM: Is data augmentation worth for energy disaggregation? | |
CN113361454B (en) | Non-supervision optimization-based deep learning non-invasive load monitoring method | |
CN116756575B (en) | Non-invasive load decomposition method based on BGAIN-DD network | |
CN116951780A (en) | Intelligent central water heater management 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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190830 |
|
RJ01 | Rejection of invention patent application after publication |