CN113595242B - Non-invasive load identification method based on depth CNN-HMM - Google Patents
Non-invasive load identification method based on depth CNN-HMM Download PDFInfo
- Publication number
- CN113595242B CN113595242B CN202110719855.XA CN202110719855A CN113595242B CN 113595242 B CN113595242 B CN 113595242B CN 202110719855 A CN202110719855 A CN 202110719855A CN 113595242 B CN113595242 B CN 113595242B
- Authority
- CN
- China
- Prior art keywords
- load
- mutual information
- cnn
- state
- equal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 39
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000012545 processing Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 230000007704 transition Effects 0.000 claims description 7
- 238000012216 screening Methods 0.000 claims description 3
- 230000002194 synthesizing effect Effects 0.000 abstract description 2
- 238000005457 optimization Methods 0.000 description 8
- 238000012544 monitoring process Methods 0.000 description 7
- 230000015556 catabolic process Effects 0.000 description 5
- 238000006731 degradation reaction Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000008034 disappearance Effects 0.000 description 3
- 238000007637 random forest analysis Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000008030 elimination Effects 0.000 description 2
- 238000003379 elimination reaction Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Image Analysis (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a non-invasive load identification method based on depth CNN-HMM, comprising the following steps: the mutual information algorithm is adopted to perform characteristic selection on the total load data collected at the power inlet, and the characteristic expression capacity is enhanced by adopting a differential processing mode; determining the running state of the load to be identified according to the effective current value of the load to be identified, and generating a tag data set of the equipment; synthesizing the load characteristic data and the label data into a training data set of the model; training a deep Convolutional Neural Network (CNN); training parameters of a Hidden Markov Model (HMM) according to a primary identification result obtained by the deep CNN network; and obtaining the structure and parameters of the complete depth CNN-HMM model, and realizing the identification of the power load. The invention can effectively extract the main data characteristics of the power utilization load, weaken the influence of irrelevant characteristics and redundant characteristics, optimize the problem of discontinuity in the identification result of the CNN algorithm and realize high-precision identification of the power utilization load in industry and commerce.
Description
Technical Field
The invention belongs to the technical field of non-invasive load identification, and particularly relates to a depth CNN-HMM-based non-invasive load identification method.
Background
Advanced and effective demand side management technology can provide security and reliability guarantee for smart grid development, and in order to improve the energy utilization efficiency of user terminal load, reduce electric quantity consumption and power demand, promote the user to participate in demand side management or realize remote optimal control, the on-line monitoring of power load is indispensable. The method for monitoring the power load mainly comprises two types of invasive load monitoring and non-invasive load monitoring, and the traditional invasive load monitoring has the problems of high input cost, high maintenance difficulty, low resident acceptance degree and the like, so that the non-invasive load monitoring with a simple hardware structure and convenient installation and maintenance has obvious advantages.
Load identification is the core content of load monitoring, and the load identification methods can be divided into three main categories, namely an optimization method, a supervised learning method and an unsupervised learning method. The optimization method is the most direct method for load identification, the extracted load characteristic data is compared with characteristic data in a template library, an optimization algorithm is adopted to minimize errors, and the state combination of the electrical equipment with the highest matching degree is found, wherein the common optimization algorithm comprises a Dynamic Time Warping (DTW) algorithm, a genetic algorithm, a particle swarm optimization algorithm and the like, but the load identification of the optimization method generally requires that the load characteristic template library is established in advance according to historical information. The supervised learning method needs specific operation information of known electrical equipment to perform initial training of a model, the performance of an algorithm depends on a data tag of model learning to a great extent, and basically, the problem of load identification is converted into a problem of classification learning to be solved, and the main methods include a Bayesian classifier, a Support Vector Machine (SVM) and an Artificial Neural Network (ANN). The unsupervised learning method can be run without any prior information, reducing the interference of the training steps compared to the supervised learning method, and the main algorithms are Hidden Markov Models (HMMs) and their extensions, such as Factor Hidden Markov Models (FHMMs) and hidden semi-markov models (HSMMs), which are applied to load recognition. The Convolutional Neural Network (CNN) has very strong feature extraction capability, and is often applied to load identification, but as the CNN itself does not have time sequence modeling capability, the perception range of time sequence data can only stay within the selected time window length, so that the problem of discontinuity often exists in the identification result of the CNN, as shown in fig. 1.
Disclosure of Invention
Aiming at the problem of non-continuity in the CNN algorithm identification result, the invention provides a non-invasive load identification method based on depth CNN-HMM. Firstly, extracting multidimensional features of the load through depth CNN, and performing primary identification of the load; and then according to the time dependence of the load state, the two classification states of the load are identified again by utilizing the HMM model, so that the state breakpoint phenomenon in the CNN identification result is reduced, and the identification accuracy is improved.
The invention solves the problems by the following technical means:
a non-invasive load identification method based on depth CNN-HMM comprises the following steps:
step 1: the mutual information algorithm is adopted to perform feature selection on the total load data, influence of irrelevant features and redundant features on the accuracy of the load identification model is weakened, and differential processing is performed to strengthen the expression capability of the features;
step 2: determining the running state of the load to be identified according to the current effective value of the load to be identified, and generating tag data of the equipment;
step 3: combining the load characteristic data obtained in the step 1 with the label data obtained in the step 2 to form a training data set of the model, selecting a residual neural network ResNet-50 as an infrastructure of a deep Convolutional Neural Network (CNN) for solving the gradient elimination and degradation problems in the deep network, and training a deep CNN model;
step 4: taking the initial identification result of the deep CNN network as an observation sequence, and training parameters of a Hidden Markov Model (HMM) so as to optimize the problem of discontinuity in the identification result of the CNN algorithm;
step 5: and obtaining a complete depth CNN-HMM non-invasive load identification model, and realizing effective identification of industrial and commercial power loads.
Further, the method for selecting the features by using the mutual information algorithm in the step 1 is as follows:
the formula for calculating the correlation of the mutual information is as follows:
wherein I (x, y) is the mutual information value size of the random variables x and y, the random variable x= { x 1 ,x 2 ,…,x T The time series of differences of any one characteristic in the total load data is represented by a random variable y= { y 1 ,y 2 ,…,y T The sequence of states of operation of the load to be identified is represented by p (x) and p (y) being the edge probability density functions of the random variables x and y, respectively, and p (x, y) being the joint probability density functions of x and y. The larger the mutual information value, the stronger the correlation between random variables.
The specific steps of the mutual information feature selection include:
(1) Calculating a mutual information value between each load characteristic and the load operation state to be identified;
(2) Ordering all the features according to the size of the mutual information from big to small;
(3) Selecting the first n features with the largest mutual information to form an initial feature set { f } of load identification 1 ,f 2 ,…,f n };
(4) Calculating f i And f j ,(i<j) The mutual information of the two, if f i And a certain characteristic f j The mutual information value of (a) is greater than the threshold mu th Will f j Removing the feature subset and adding new features in sequence;
(5) And recalculating the mutual information values among the features in the new feature subset, repeatedly screening until the mutual information values among the features are smaller than a threshold value, and constructing a new feature set.
Further, any load characteristic data acquired in the step 1 can be used as a time sequence r= { R 0 ,R 1 ,R 2 …,R T And (2) representing the time length of the load data, wherein T is the time length of the load data, and the difference characteristic sequence x= { x of the load can be obtained by differential processing 1 ,x 2 ,…,x T X, where x t =R t -R t-1 (1.ltoreq.t.ltoreq.T). Given the time window length w and the number n of load characteristics, input= { H of input load data of a model can be generated 1 ,H 2 ,…,H L }, wherein Time series of differences of the ith window representing the jth load profile,/th window representing the jth load profile>Wherein i is more than or equal to 1 and less than or equal to T-w+1, j is more than or equal to 1 and less than or equal to n.
Further, the load data to be identified collected in the step 2 is based on the current effective value IrmsGenerating a tag state set target= { y of a device 1 ,y 2 ,…,y L -wherein y i Representing the operating state of the load at the i + w/2 point in time within the i-th time window,
further, the parameter λ= { U, V, a, B, pi } of the HMM model in step 4 is defined as follows:
implicit state set u= { U 1 ,u 2 ,…,u N N is the number of hidden states, the hidden state Q of the system at time t t ∈U;
Observation state set v= { V 1 ,v 2 ,…,v M -where M is the number of different observations that each implicit state may output, the observed state S of the system at time t t ∈V;
State transition probability matrix a= { a ij },a ij Indicating a state u at time t i Under the condition of (1), the state is u at the time t+1 j Probability of a) ij =P(Q t+1 =u j |Q t =u i ) Wherein 1 is less than or equal to i, j is less than or equal to N,
observed state transition probability matrix b= { B i (k)},b i (k) Indicating that the implicit state at time t is u i Under the condition of (1) that the observation state is v k Probability of b i (k)=P(S t =v k |Q t =u i ) Wherein i is more than or equal to 1 and less than or equal to N, k is more than or equal to 1 and less than or equal to M,
initial state probability matrix pi= { pi i Probability distribution of initial state at time t=1, pi i =P(Q 1 =u i ) Wherein i is more than or equal to 1 and less than or equal to N,
compared with the prior art, the invention has the beneficial effects that at least:
the invention discloses a non-invasive load identification method based on depth CNN-HMM, after total load data of an electric power system is obtained, firstly, a mutual information algorithm is adopted to perform feature selection, irrelevant features and redundant features in the load feature data are removed, the accuracy of load identification is improved, on the other hand, the feature selection reduces the data dimension to a great extent, and the calculation amount of a model is reduced; then constructing a deep CNN network architecture by adopting a residual neural network, and solving the problems of gradient disappearance and degradation in the deep network training process; further, the problem of discontinuity in the CNN identification result is optimized by adopting the HMM model, and the combination of the deep CNN network and the HMM model completes the extraction of the deep hidden characteristics of the load and the optimization of the discontinuity of the identification result, so that the industrial and commercial power loads can be effectively identified, and the method has better accuracy and identification precision.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic diagram of a discontinuity problem in CNN recognition results.
Fig. 2 is an algorithm flow chart of the deep CNN-HMM model.
FIG. 3 is an effect of time window length and feature number on load recognition accuracy.
Fig. 4 is an active power graph of an elevator load.
FIG. 5 is a graph of the convergence of the loss function during ResNet-50 network training.
Fig. 6 is a load recognition result of the comparison algorithm.
Detailed Description
In order to make the above objects, features and advantages of the present invention more comprehensible, the following detailed description of the technical solution of the present invention refers to the accompanying drawings and specific embodiments. It should be noted that the described embodiments are only some embodiments of the present invention, and not all embodiments, and that all other embodiments obtained by persons skilled in the art without making creative efforts based on the embodiments in the present invention are within the protection scope of the present invention.
The invention provides a non-invasive load identification method based on depth CNN-HMM, which adopts mutual information to perform feature selection, so that the data dimension can be reduced, the operand is reduced, irrelevant features and redundant features in load feature data can be removed, and the accuracy of load identification is improved; on one hand, based on the strong feature extraction capability of the deep Convolutional Neural Network (CNN), the implicit features of the load are extracted, the nonlinear mapping relation between the total load data and the load state is fitted, and on the other hand, the transition probability description of the Hidden Markov Model (HMM) on the load state is considered aiming at the state breakpoint problem existing in the deep CNN identification result, so that the discontinuity of the deep CNN identification result is optimized, and the high-precision identification of the industrial and commercial load is realized.
The following applies to embodiments of the complete method according to the invention:
fig. 2 is a flowchart of a non-invasive load identification method based on deep CNN-HMM according to an embodiment of the present invention, the identification method includes the following steps:
step 1: and the mutual information algorithm is adopted to perform feature selection on the total load data, influence of irrelevant features and redundant features on the accuracy of the load identification model is weakened, and differential processing is performed to strengthen the expression capability of the features.
The calculation formula of the mutual information is as follows:
wherein I (x, y) is the mutual information value size of the random variables x and y, the random variable x= { x 1 ,x 2 ,…,x T The time series of differences of any one characteristic in the total load data is represented by a random variable y= { y 1 ,y 2 ,…,y T The sequence of states of operation of the load to be identified is represented by p (x) and p (y) being the edge probability density functions of the random variables x and y, respectively, and p (x, y) being the joint probability density functions of x and y. The larger the mutual information value, the stronger the correlation between random variables.
The specific steps of adopting mutual information algorithm to select the characteristics include:
(1) Calculating a mutual information value between each load characteristic and the load operation state to be identified;
(2) Ordering all the features according to the size of the mutual information from big to small;
(3) Selecting the first n features with the largest mutual information to form an initial feature set { f } of load identification 1 ,f 2 ,…,f n };
(4) Calculating f i And f j ,(i<j) The mutual information of the two, if f i And a certain characteristic f j The mutual information value of (a) is greater than the threshold mu th Will f j Removing the feature subset and adding new features in sequence;
(5) And recalculating the mutual information values among the features in the new feature subset, repeatedly screening until the mutual information values among the features are smaller than a threshold value, and constructing a new feature set.
In this embodiment, one of the purposes of performing feature selection by using the mutual information algorithm is to reduce the data dimension and reduce the computation of the model. Table 1 records the relationship between the number of feature choices and the training time per turn of the ResNet-50 network, with the training time of the model decreasing as the number of features decreases.
TABLE 1 training time for each round of ResNet-50 network
Number of features | 30 | 40 | 50 | 60 | 70 |
Training time | 2′57″ | 3′12″ | 3′33″ | 3′50″ | 4′24″ |
In this embodiment, the second purpose of performing feature selection by using the mutual information algorithm is to reject irrelevant features and redundant features, so as to improve the accuracy of load identification. FIG. 3 shows the relationship between the time window length and the number of load features and the model identification accuracy, and the finally selected parameter values are respectively: the time window length w=60, the number of load features n=40, and the load identification model can obtain the best identification effect.
In this embodiment, the purpose of processing the model input data in a differential manner is to enhance the expressive power of the load characteristics. Any characteristic data in the total load data, the available time sequence R= { R 0 ,R 1 ,R 2 …,R T And (2) representing the time length of the load data, wherein T is the time length of the load data, and the difference characteristic sequence x= { x of the load can be obtained by differential processing 1 ,x 2 ,…,x T X, where x t =R t -R t-1 (1.ltoreq.t.ltoreq.T). Given the time window length w and the number n of load characteristics, input= { H of input load data of a model can be generated 1 ,H 2 ,…,H L }, wherein Time series of differences of the ith window representing the jth load profile,/th window representing the jth load profile>Wherein i is more than or equal to 1 and less than or equal to T-w+1, j is more than or equal to 1 and less than or equal to n.
Step 2: and determining the running state of the load to be identified according to the current effective value of the load to be identified, and generating the label data of the equipment.
In this embodiment, an elevator load is used as the load to be identified, and fig. 4 is an active power graph of the elevator load. Generating a tag state set target= { y of the device based on the magnitude of the current effective value Irms of the elevator load 1 ,y 2 ,…,y L -wherein y i Representing the operating state of the load at the i + w/2 point in time within the i-th time window,
step 3: and (3) synthesizing the load characteristic data obtained in the step (1) and the label data obtained in the step (2) into a training data set of the model, selecting a residual neural network ResNet-50 as an infrastructure of a deep Convolutional Neural Network (CNN) in order to solve the gradient elimination and degradation problems in the deep network, and training the deep CNN model to learn a nonlinear mapping relation between the fitting characteristic data and the label.
In the embodiment, the purpose of adopting the ResNet-50 network is to relieve the gradient disappearance and degradation problems existing in the deep network training process and improve the accuracy of the load identification algorithm. 80% of the training dataset was used for model training and the remaining 20% was used for model evaluation, thus selecting the best performing model. In the training process, an Adam optimizer is adopted to accelerate the convergence rate of the model, the initial learning rate is set to be 0.001, and the cross entropy function is used as a loss function to calculate errors. Fig. 5 is a plot of the convergence of the loss function value with the number of iterations.
Step 4: and taking the initial identification result of the deep CNN network as an observation sequence, and training parameters of a Hidden Markov Model (HMM) so as to optimize the problem of discontinuity in the identification result of the CNN algorithm.
The five-tuple parameter λ= { U, V, a, B, pi } of the HMM model is defined as follows:
implicit state set u= { U 1 ,u 2 ,…,u N N is the number of hidden states, the hidden state Q of the system at time t t E U; for the elevator load, its implicit state number n=2, only active and inactive states, i.e. u= {0,1}.
Observation state set v= { V 1 ,v 2 ,…,v M -where M is the number of different observations that each implicit state may output, the observed state S of the system at time t t E, V; for elevator load, the number of observed states m=2, v= {0,1}.
State transition probability matrix a= { a ij },a ij Indicating a state u at time t i Under the condition of (1), the state is u at the time t+1 j Probability of a) ij =P(Q t+1 =u j |Q t =u i ) Wherein 1 is less than or equal to i, j is less than or equal to N,
observed state transition probability matrix b= { B i (k)},b i (k) Indicating that the implicit state at time t is u i Under the condition of (1) that the observation state is v k Probability of b i (k)=P(S t =v k |Q t =u i ) Wherein i is more than or equal to 1 and less than or equal to N, k is more than or equal to 1 and less than or equal to M,
initial state probability matrix pi= { pi i Probability distribution of initial state at time t=1, pi i =P(Q 1 =u i ) Wherein i is more than or equal to 1 and less than or equal toN,
In this embodiment, the purpose of introducing the HMM model is to improve the accuracy of load identification, and the specific flow of the HMM algorithm is as follows:
(1) Initializing model parameters lambda of HMM 0 '={A,B,π};
(2) Generating an observation sequence S= { S according to the primary identification result of the depth CNN 1 ,S 2 ,…,S L };
(3) Iteratively updating the parameter lambda 'until the probability P (S|lambda') of the occurrence of the observation sequence reaches a maximum value;
(4) The state sequence q= { Q that maximizes the probability P (q|s, λ ') is found by the optimum parameter λ' and the observation sequence S using the viterbi algorithm 1 ,Q 2 ,…,Q L And (3) obtaining the final identification result of the elevator load.
Step 5: and obtaining a complete depth CNN-HMM non-invasive load identification model, and realizing effective identification of industrial and commercial power loads.
In this embodiment, several common non-invasive load recognition algorithms are used as comparison algorithms to perform load recognition as well, including KNN algorithm, random Forest (RF), HMM model, long-short-term memory network (LSTM), CNN, and CNN-LSTM model. Table 2 shows the results of load identification, including the accuracy Acc, accuracy P, recall R, and harmonic mean evaluation index F of load identification score Fig. 6 shows the identification results of several algorithms in the same period of time. Table 2 and FIG. 6 show that the method provided by the invention has higher precision, and meanwhile, the discontinuity of the identification result caused by the identification error of the CNN algorithm is effectively reduced. Although the CNN-HMM algorithm is not optimal in terms of accuracy P and recall R, it is not optimal in terms of F that comprehensively reflects the degree of reliability of the recognition result score The highest value of 0.9879 is obtained, and compared with other comparison algorithms, the method has the highest identification accuracy.
Table 2 compares algorithm load identification results
Algorithm | P/% | R/% | F score | Acc/% |
KNN | 93.41 | 91.60 | 0.9249 | 86.33 |
RF | 91.35 | 98.36 | 0.9473 | 90.75 |
HMM | 93.73 | 93.76 | 0.9375 | 88.23 |
LSTM | 96.98 | 96.88 | 0.9693 | 94.00 |
CNN | 100.00 | 95.43 | 0.9766 | 95.60 |
CNN-LSTM | 96.94 | 99.22 | 0.9806 | 96.50 |
CNN-HMM | 98.50 | 99.07 | 0.9879 | 97.90 |
The invention adopts the mutual information algorithm to select the characteristics, removes irrelevant characteristics and redundant characteristics in the load characteristic data, improves the accuracy of load identification, simultaneously reduces the data dimension to a great extent, and reduces the operation amount of the model. The method solves the gradient disappearance and degradation problems in the deep network training process by adopting the residual neural network, optimizes the discontinuity problem in the CNN identification result by adopting the HMM model, combines the extraction capability of the deep CNN network to the deep hidden features of the load and the optimization capability of the HMM model to the discontinuity of the identification result, can effectively identify the industrial and commercial power load, and has better accuracy and identification precision.
The foregoing examples illustrate only a few embodiments of the invention and are described in detail herein without thereby limiting the scope of the invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.
Claims (3)
1. A depth CNN-HMM based non-intrusive load identification method, comprising the steps of:
step 1: adopting a mutual information algorithm to perform feature selection on the total load data, and performing differential processing to strengthen the expression capability of the features;
step 2: determining the running state of the load to be identified according to the current effective value of the load to be identified, and generating tag data of the equipment;
step 3: combining the load characteristic data obtained in the step 1 with the tag data obtained in the step 2 to form a training data set of the model, selecting a residual neural network ResNet-50 as an infrastructure of the depth CNN, and training the depth CNN model;
step 4: taking the primary identification result of the deep CNN network as an observation sequence, training parameters of an HMM model, and thus optimizing the problem of discontinuity in the identification result of the CNN algorithm;
step 5: obtaining a complete depth CNN-HMM non-invasive load identification model, and realizing effective identification of industrial and commercial power loads;
the method for selecting the characteristics by adopting the mutual information algorithm in the step 1 is as follows:
the formula for calculating the correlation of the mutual information is as follows:
wherein I (x, y) is the mutual information value size of the random variables x and y, the random variable x= { x 1 ,x 2 ,…,x T The time series of differences of any one characteristic in the total load data is represented by a random variable y= { y 1 ,y 2 ,…,y T The sequence of running states of the load to be identified is represented by p (x) and p (y) being the edge probability density functions of the random variables x and y, respectively, and p (x, y) being the joint probability density functions of x and y; the larger the mutual information value, the phase between random variables is shownThe stronger the relationship;
the specific steps of the mutual information feature selection include:
(1) Calculating a mutual information value between each load characteristic and the load operation state to be identified;
(2) Ordering all the features according to the size of the mutual information from big to small;
(3) Selecting the first n features with the largest mutual information to form an initial feature set { f } of load identification 1 ,f 2 ,…,f n };
(4) Calculating f i And f j The mutual information between (i < j), if f i And a certain characteristic f j The mutual information value of (a) is greater than the threshold mu th Will f j Removing the feature subset and adding new features in sequence;
(5) Re-calculating the mutual information values among the features in the new feature subset, repeatedly screening until the mutual information values among the features are smaller than a threshold value, and constructing a new feature set;
the parameter λ= { U, V, a, B, pi } of the HMM model in step 4 is defined as follows:
implicit state set u= { U 1 ,u 2 ,…,u N N is the number of hidden states, the hidden state Q of the system at time t t ∈U;
Observation state set v= { V 1 ,v 2 ,…,v M -where M is the number of different observations that each implicit state may output, the observed state S of the system at time t t ∈V;
State transition probability matrix a= { a ij },a ij Indicating a state u at time t i Under the condition of (1), the state is u at the time t+1 j Probability of a) ij =P(Q t+1 =u j |Q t =u i ) Wherein 1 is less than or equal to i, j is less than or equal to N,
observed state transition probability matrix b= { B i (k)},b i (k) Indicating implicit shape at time tState u i Under the condition of (1) that the observation state is v k Probability of b i (k)=P(S t =v k |Q t =u i ) Wherein i is more than or equal to 1 and less than or equal to N, k is more than or equal to 1 and less than or equal to M,
initial state probability matrix pi= { pi i Probability distribution of initial state at time t=1, pi i =P(Q 1 =u i ) Wherein i is more than or equal to 1 and less than or equal to N,
2. the depth CNN-HMM-based non-invasive load recognition method according to claim 1, wherein any load data collected in step 1 is available in a time sequence r= { R 0 ,R 1 ,R 2 …,R T And (2) representing the time length of the load data, wherein T is the time length of the load data, and the difference characteristic sequence x= { x of the load can be obtained by differential processing 1 ,x 2 ,…,x T X, where x t =R t -R t-1 (1.ltoreq.t.ltoreq.T); given the time window length w and the number n of load characteristics, input= { H of input load data of a model can be generated 1 ,H 2 ,…,H L }, wherein H i ={X i (1) ,X i (2) ,…,X i (n) },X i (j) Time series of differences, X, of the ith window representing the jth load signature i (j) =[x i (j) ,x i+1 (j) ,…,x i+w-1 (j) ] T Wherein i is more than or equal to 1 and less than or equal to T-w+1, and j is more than or equal to 1 and less than or equal to n.
3. The non-invasive load identification method based on deep CNN-HMM according to claim 1, wherein the load data to be identified collected in step 2 is used to generate a tag state set target = of the device based on the magnitude of the current effective value Irms thereof{y 1 ,y 2 ,…,y L Where yi represents the operating state of the load at the i + w/2 point in time within the i-th time window,
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110719855.XA CN113595242B (en) | 2021-06-28 | 2021-06-28 | Non-invasive load identification method based on depth CNN-HMM |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110719855.XA CN113595242B (en) | 2021-06-28 | 2021-06-28 | Non-invasive load identification method based on depth CNN-HMM |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113595242A CN113595242A (en) | 2021-11-02 |
CN113595242B true CN113595242B (en) | 2024-02-20 |
Family
ID=78244845
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110719855.XA Active CN113595242B (en) | 2021-06-28 | 2021-06-28 | Non-invasive load identification method based on depth CNN-HMM |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113595242B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117473368A (en) * | 2023-10-30 | 2024-01-30 | 湖南大学 | Non-invasive load monitoring semi-supervised learning method and system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019234533A1 (en) * | 2018-06-06 | 2019-12-12 | Gridimp Limited | Electrical load monitoring system |
CN111783684A (en) * | 2020-07-03 | 2020-10-16 | 湖南大学 | Bi-GRU-SA-based household electrical load decomposition method and system |
CN111864896A (en) * | 2019-04-29 | 2020-10-30 | 清华大学 | Power load monitoring method and system |
US10983496B1 (en) * | 2020-10-13 | 2021-04-20 | King Abdulaziz University | Apparatus and method for non-intrusive load monitoring aided hybrid high and low frequency approaches |
CN112821559A (en) * | 2021-01-22 | 2021-05-18 | 西安理工大学 | Non-invasive household appliance load depth re-identification method |
-
2021
- 2021-06-28 CN CN202110719855.XA patent/CN113595242B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019234533A1 (en) * | 2018-06-06 | 2019-12-12 | Gridimp Limited | Electrical load monitoring system |
CN111864896A (en) * | 2019-04-29 | 2020-10-30 | 清华大学 | Power load monitoring method and system |
CN111783684A (en) * | 2020-07-03 | 2020-10-16 | 湖南大学 | Bi-GRU-SA-based household electrical load decomposition method and system |
US10983496B1 (en) * | 2020-10-13 | 2021-04-20 | King Abdulaziz University | Apparatus and method for non-intrusive load monitoring aided hybrid high and low frequency approaches |
CN112821559A (en) * | 2021-01-22 | 2021-05-18 | 西安理工大学 | Non-invasive household appliance load depth re-identification method |
Non-Patent Citations (1)
Title |
---|
一种基于LSTM模型的电力负荷辨识方法;刘恒勇;刘永礼;邓世聪;史帅彬;闵若琳;周东国;;电测与仪表(23);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN113595242A (en) | 2021-11-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110795690A (en) | Wind power plant operation abnormal data detection method | |
CN113408341A (en) | Load identification method and device, computer equipment and storage medium | |
CN115344693B (en) | Clustering method based on fusion of traditional algorithm and neural network algorithm | |
Wang et al. | Big data analytics for price forecasting in smart grids | |
CN113595242B (en) | Non-invasive load identification method based on depth CNN-HMM | |
CN116226689A (en) | Power distribution network typical operation scene generation method based on Gaussian mixture model | |
CN113762355B (en) | User abnormal electricity behavior detection method based on non-invasive load decomposition | |
Wang et al. | DiffLoad: uncertainty quantification in load forecasting with diffusion model | |
CN111898694B (en) | Non-invasive load identification method and device based on random tree classification | |
CN112215490B (en) | Power load cluster analysis method based on correlation coefficient improved K-means | |
CN113624998A (en) | Electric boiler heat supplementing and heat storing cost optimization method and device based on electric power big data | |
CN117493922A (en) | Power distribution network household transformer relation identification method based on data driving | |
CN112149052A (en) | Daily load curve clustering method based on PLR-DTW | |
CN111026741A (en) | Data cleaning method and device based on time series similarity | |
Zhang et al. | Dynamic conditional score model-based weighted incremental fuzzy clustering of consumer power load data | |
CN116681497A (en) | Fund risk identification method based on graph neural network, computer device and computer readable storage medium | |
Liu et al. | Embedded scenario clustering for wind and photovoltaic power, and load based on multi-head self-attention | |
Chen et al. | A GPU-accelerated approximate algorithm for incremental learning of Gaussian mixture model | |
CN115423091A (en) | Conditional antagonistic neural network training method, scene generation method and system | |
Flasiński et al. | Multi-Derivational Parsing of Vague Languages-the New Paradigm of Syntactic Pattern Recognition | |
CN110942089B (en) | Multi-level decision-based keystroke recognition method | |
CN114168822A (en) | Method for establishing time series data clustering model and time series data clustering | |
Fan et al. | A reduced weighted wang-mendel algorithm using the clustering algorithm to build fuzzy system | |
Zhang et al. | Load prediction based on depthwise separable convolution model | |
CN112231933B (en) | Feature selection method for radar electromagnetic interference effect analysis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |