Non-intrusive load identification method and system

Abstract

The invention relates to the technical field of advanced measurement systems in the intelligent power utilization technology, and the method comprises the steps that in view of the advantages of variational mode decomposition in blind source separation and feature extraction, load identification is achieved in combination with VMD-FastICA load decomposition and a VMD-Entropy-LSTM identification algorithm. Decomposing the total load power signal by using VMD to obtain a plurality of modal components; and according to a kurtosis criterion and singular value decomposition, reconstructing a decomposed modal component, virtualizing single-channel blind source separation into multi-channel blind source separation input, and then separating a load signal by using rapid independent component analysis. A multi-dimensional feature matrix is constructed by calculating and decomposing energy and energy entropy of a load waveform modal component, and load classification and identification are performed by adopting a long-short-term memory network. According to the method, an event monitoring process is not needed, the integrity of the separated load waveform can be ensured, the load operation state can be identified, and data support is provided for energy consumption management of a demand side.

Description

Technical Field

[0001] The present invention relates to the technical field of advanced measurement systems in intelligent power consumption technologies, and in particular to a non-invasive load identification method and system. Background Art

[0002] As one of the indispensable parts of the Advanced Metering Infrastructure (AMI) in the smart grid, load monitoring has important practical significance. Its purpose is to obtain the working status information of the load and effectively identify the load according to the difference in working status. It is the first step to realize the smart grid and an important part of the smart power consumption technology system. It can analyze the load composition of users in detail and obtain various degrees of user power consumption information. It is crucial for the management of demand-side response and has great significance for users, power companies and other parties. Among them, non-intrusive load monitoring has the advantages of no interference, simple installation and maintenance, high cost-effectiveness, real-time monitoring and data acquisition compared with intrusive load monitoring. At present, most non-intrusive load monitoring is based on event detection, which requires high-frequency feature extraction of load characteristics and requires too high equipment accuracy. Load monitoring based on non-event detection reduces the process of event detection and realizes load monitoring by first decomposing and then identifying the load signal. However, how to ensure that the similarity coefficient of load decomposition waveform is higher and the load identification accuracy is more accurate under low-frequency sampling has become an urgent problem to be solved. Summary of the invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides a non-invasive load identification method, which can solve the shortcomings of the existing non-invasive load decomposition, such as the lack of a complete and comprehensive decomposition method for independent loads, resulting in the integrity of electricity consumption information cannot be guaranteed, and the problem that load characteristics of non-invasive load identification are prone to overlap under low-frequency sampling, resulting in low identification accuracy.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: a non-intrusive load identification method, comprising: collecting the total power of user electricity consumption, performing median filtering, de-averaging and whitening processing on the original data to eliminate impact noise in the sampled data and eliminate the correlation between redundant information and load characteristics; performing signal decomposition on the total power consumption after data preprocessing according to the VMD algorithm to form multiple modal components; performing optimal reconstruction on the decomposed multi-modal components, and reconstructing the decomposed modal components according to the kurtosis criterion and singular value decomposition; according to the reconstructed multi-modal components, converting the one-dimensional power characteristics into a multi-dimensional feature matrix, and inputting it into the FastICA separation algorithm for load decomposition; performing VMD decomposition on the independent load waveform after load decomposition and extracting the energy entropy of each modal component, forming a multi-feature matrix and inputting it into a pre-trained LSTM-based non-intrusive load identification model for load identification.

[0006] As a preferred solution of the non-intrusive load identification method described in the present invention, wherein: the collecting and processing of raw data includes: data collection refers to collecting low-frequency data of the operating load, the frequency of which is 1 Hz;

[0007] According to the median filter, the impact noise of the original data and the influence of missing values ​​and abnormal values ​​caused by equipment sampling can be reduced. The specific formula is:

[0008] g(x,y)=med{f(xk,y-1),(k,1)∈W}

[0009] Among them, f(x, y) is the original waveform, g(x, y) is the processed image, W is the two-dimensional template, and the dimensional differences between the data are eliminated according to the centralization process to reduce the computational complexity of the objective function in the algorithm. The specific formula is:

[0010] X'(t)=X(t)-E(X(t))

[0011] Among them, E(X(t)) represents the solution of the signal mean, and the correlation in the data is eliminated by whitening processing.

[0012] As a preferred solution of the non-intrusive load identification method described in the present invention, the signal decomposition to form multiple modal components includes decomposing the original power signal P(t) after data preprocessing into K intrinsic modal components IMF is an amplitude-frequency modulated signal, performing Hilbert transform on each mode separation to obtain a single-sided spectrum of each component, and calculating the center frequency w corresponding to each modal function. k The exponential term aliasing modulates the single-sided spectrum of each mode to the corresponding baseband. Each modal component is around its center frequency. The sum of the modal components is equal to the input signal f(t). The minimum sum of the bandwidths of the modal components is the constraint condition. The specific variational constraint problem is:

[0013]

[0014] Among them, U k Each eigenmode component; w k is the center frequency corresponding to each modal component, δ(t) is the unit pulse function, u k (t) is the Kth IMF component of the decomposition;

[0015] The constrained variational problem is transformed into an unconstrained variational problem to solve the objective function. VMD introduces a quadratic penalty factor α and a Lagrange multiplication operator λ. The specific expression is:

[0016]

[0017] Use Fourier isometric transform to transform each component into frequency domain, and iterate each modal component U k and the center frequency w k and the Lagrange multiplier λ, the updated expression is:

[0018]

[0019] Continue to iterate and update until the termination condition is met:

[0020]

[0021] As a preferred solution of the non-intrusive load identification method described in the present invention, the optimal reconstruction of the multimodal components includes determining the K value of VMD decomposition based on the kurtosis criterion, taking the maximum kurtosis as the optimization criterion, when K=n, when the kurtosis value of the IMF component with the largest correlation coefficient with the source signal is the largest, K=n is the optimal value, and the specific expression of kurtosis is:

[0022]

[0023] Where μ is the mean of signal x, and σ is the standard deviation of signal x;

[0024] The mutual correlation coefficient can express the similarity between the modal component IMF and the source signal. The specific expression is:

[0025]

[0026] As a preferred embodiment of the non-intrusive load identification method of the present invention, the singular value decomposition includes estimating the number of source signals based on the singular value decomposition, constructing a sample covariance matrix Rx=E[x imf (t),x imf (t) T ], R xThe specific expression of the singular value decomposition is:

[0027]

[0028] Among them, Λ s are the k main eigenvalues, Λ s =diag{λ1≥λ2≥…≥λ k},Λ b is Qk noise eigenvalues, Λ b =diag{λ k+1 ≥λ k+2 ≥…≥λ Q}, sort the eigenvalue matrix in descending order [λ1,λ2,…,λ k ], find the ratio Δλ between two adjacent eigenvalues 1,2 =λ1 / λ2, Δλ 2,3 =λ2 / λ3,…,Δλ k-1,k =λ k-1 / λ k , select two adjacent eigenvalues ​​maxΔλ n-1,n , corresponding to the index number n, that is, the number of source signals, the IMFs with larger correlation coefficients with the source signals are retained, and the number is the same as the estimated number of source signals.

[0029] As a preferred solution of the non-intrusive load identification method described in the present invention, the load decomposition includes converting the underdetermined blind source separation problem into a positive blind source problem and inputting it into FastICA to realize load decomposition, and evaluating the decomposition model by the similarity coefficient index. The specific formula is:

[0030]

[0031] Among them, x is the real load signal, y is the separated load signal, and M is the sampling point.

[0032] As a preferred solution of the non-intrusive load identification method described in the present invention, the load identification includes: when the load electrical appliances are operating independently, taking power data of each electrical appliance power signal in different time periods, performing VMD decomposition, and calculating the energy E of each IMF. i and the energy entropy H of the load signal EN Construct the corresponding feature vector T = [E1, E2, ..., Ek, H EN ] Input LSTM for training, the specific expressions of energy and energy entropy are:

[0033]

[0034] Among them, P i =E i / E is the proportion of the i-th intrinsic mode IMF component to the total energy The ratio of

[0035] The load decomposition signal based on VMD-FastICA is collected, and the feature vector T' is constructed according to the same method as the input of the LSTM classifier. The type of load electrical appliance is determined through the output of the long short-term memory network classifier to identify the load.

[0036] The load identification model is evaluated by accuracy, precision, recall, and F1:

[0037]

[0038] Among them, TP means that the state of the appliance is running and the appliance is predicted to be in the running state, FP means that the actual state of the appliance is off and the appliance is predicted to be in the running state, TN means that the state of the appliance is off and the appliance is predicted to be off, and FN means that the state of the appliance is running and the appliance is predicted to be off.

[0039] Another object of the present invention is to provide a non-invasive load identification system, which can reduce the acquisition accuracy and acquisition equipment cost, and can analyze the energy consumption of electrical appliances in a specific time period based on low-frequency load data, providing data support for subsequent energy consumption management.

[0040] As a preferred solution of the non-intrusive load identification system described in the present invention, it includes: a data acquisition module, a data preprocessing module, a signal decomposition module, a modal component reconstruction module, a feature conversion and load decomposition module, and a load identification module;

[0041] The data collection module collects the total power consumed by the user, and records the power usage data at a certain frequency through a smart meter or data collection device;

[0042] The data preprocessing module performs median filtering, de-averaging and whitening on the original data to eliminate the impact noise, redundant information and the correlation between load characteristics, thereby improving the data quality;

[0043] The signal decomposition module performs variational mode decomposition on the preprocessed total power consumption, and decomposes the signal into multiple intrinsic mode functions;

[0044] The modal component reconstruction module selects and reconstructs the decomposed modal components according to the kurtosis criterion and singular value decomposition, and retains the most valuable load characteristic information;

[0045] The feature conversion and load decomposition module converts the reconstructed multi-modal components into one-dimensional power features into a multi-dimensional feature matrix, and inputs the FastICA separation algorithm to perform load decomposition to separate independent load signals;

[0046] The load identification module performs VMD decomposition on the independent load waveform after load decomposition, and extracts the energy entropy of each modal component to form a multi-feature matrix which is input into a pre-trained non-intrusive load identification model based on a long short-term memory (LSTM) network to perform load identification.

[0047] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and is characterized in that the processor implements the steps of a non-intrusive load identification method when executing the computer program.

[0048] A computer-readable storage medium stores a computer program, wherein the computer program implements the steps of a non-intrusive load identification method when executed by a processor.

[0049] Beneficial effects of the present invention: The present invention is based on a load decomposition method for blind source separation, and applies variational mode decomposition in blind source separation to load decomposition, abandoning the traditional event detection process, reducing trivial feature extraction, improving the similarity coefficient of load decomposition, and retaining the integrity of the decomposed load signal. Based on the characteristic source of active power, the present invention uses VMD decomposition combined with energy entropy to form a multi-dimensional feature matrix, fully explores the characteristics of each load, better reflects the differences between different electrical appliances, and combines with an LSTM classifier that is good at learning time series features to improve the load identification accuracy. The present invention overcomes the problem of low similarity coefficient of load waveform when low-frequency data is used for load decomposition, the problem of low recognition accuracy caused by easy overlap of load features under low-frequency sampling, and the problem that traditional load identification can only identify the load state but cannot retain the complete load waveform. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0051] Figure 1 A schematic flow chart of a non-intrusive load identification method provided by an embodiment of the present invention.

[0052] Figure 2 A load decomposition model diagram of a non-intrusive load identification method provided by an embodiment of the present invention.

[0053] Figure 3 A load decomposition flow chart of a non-intrusive load identification method provided by an embodiment of the present invention.

[0054] Figure 4A load identification model diagram of a non-invasive load identification method provided by an embodiment of the present invention.

[0055] Figure 5 A load identification flow chart of a non-intrusive load identification method provided by an embodiment of the present invention.

[0056] Figure 6 A comparison diagram of a load decomposition source signal and a separation signal of a non-intrusive load identification method provided by an embodiment of the present invention.

[0057] Figure 7 A load identification confusion diagram of a non-intrusive load identification method provided by an embodiment of the present invention.

[0058] Figure 8 A schematic diagram of working modules of a non-intrusive load identification system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0060] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0061] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0062] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0063] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0064] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0065] Example 1

[0066] Reference Figure 1-Figure 5 , which is the first embodiment of the present invention, provides a non-intrusive load identification method, including:

[0067] S1: Collect the total power consumed by users, perform median filtering, de-averaging and whitening on the raw data to eliminate impact noise in the sampled data and eliminate the correlation between redundant information and load characteristics.

[0068] Furthermore, the collecting and processing of raw data includes that data collection refers to collecting low-frequency data of the operating load, with a frequency of 1 Hz;

[0069] According to the median filter, the impact noise of the original data and the influence of missing values ​​and abnormal values ​​caused by equipment sampling can be reduced. The specific formula is:

[0070] g(x,y)=med{f(xk,y-1),(k,1)∈W}

[0071] Among them, f(x, y) is the original waveform, g(x, y) is the processed image, and W is a two-dimensional template, usually 3*3, 5*5.

[0072] The difference in dimensions between data is eliminated by centralization processing, which reduces the computational complexity of the objective function in the algorithm. The specific formula is:

[0073] X'(t)=X(t)-E(X(t))

[0074] Among them, E(X(t)) represents the solution of the signal mean, and the correlation in the data is eliminated by whitening processing.

[0075] S2: The total power consumption after data preprocessing is decomposed according to the VMD algorithm to form multiple modal components.

[0076] Furthermore, the original power signal P(t) after data preprocessing is decomposed into K intrinsic mode components IMF, which is an amplitude modulation frequency modulation (AM-FM) signal. The specific expression is:

[0077] U k (t) = A k (t)cos(φ k (t))

[0078] In the formula, A k is the signal U k (t) envelope amplitude, φ k (t) is the instantaneous phase.

[0079] Calculate the bandwidth of each mode. Perform Hilbert transform on each mode separation to obtain the single-sided spectrum of each component. The specific formula is:

[0080]

[0081] For each mode function, the corresponding center frequency w k The exponential term aliasing modulates the single-sided spectrum of each mode to the corresponding baseband. The specific formula is:

[0082]

[0083] Furthermore, the decomposition of the signal into multiple modal components includes decomposing the original power signal P(t) after data preprocessing into K intrinsic modal components IMF, which is an amplitude-frequency modulated signal, performing Hilbert transform on each modal separation to obtain a single-sided spectrum of each component, and calculating the center frequency w corresponding to each modal function. k The exponential term aliasing modulates the single-sided spectrum of each mode to the corresponding baseband. Each modal component is around its center frequency. The sum of the modal components is equal to the input signal f(t). The minimum sum of the bandwidths of the modal components is the constraint condition. The specific variational constraint problem is:

[0084]

[0085] Among them, U k Each eigenmode component; w k is the center frequency corresponding to each modal component, δ(t) is the unit pulse function, u k (t) is the Kth IMF component of the decomposition;

[0086] The constrained variational problem is transformed into an unconstrained variational problem to solve the objective function. VMD introduces a quadratic penalty factor α and a Lagrange multiplication operator λ. The specific expression is:

[0087]

[0088] Use Fourier isometric transform to transform each component into frequency domain, and iterate each modal component U k and the center frequency w k and the Lagrange multiplier λ, the updated expression is:

[0089]

[0090]

[0091] Continuously iterate and update until the termination condition is met:

[0092]

[0093] S3: The decomposed multimodal components are optimally reconstructed and the decomposed modal components are reconstructed according to the kurtosis criterion and singular value decomposition.

[0094] Furthermore, the K value of VMD decomposition is determined based on the kurtosis criterion, and the maximum kurtosis value is used as the optimization criterion. When K = n, the kurtosis value of the IMF component with the largest correlation coefficient with the source signal is the largest, and K = n is the optimal value. The specific expression of kurtosis is:

[0095]

[0096] Where μ is the mean of signal x, and σ is the standard deviation of signal x;

[0097] The mutual correlation coefficient can express the similarity between the modal component IMF and the source signal. The specific expression is:

[0098]

[0099] Furthermore, the singular value decomposition includes estimating the number of source signals based on the singular value decomposition and constructing a sample covariance matrix Rx=E[x imf (t),x imf (t) T ], R x The specific expression of the singular value decomposition is:

[0100]

[0101] Among them, Λ s are the k main eigenvalues, Λ s =diag{λ1≥λ2≥…≥λk},Λ b is Qk noise eigenvalues, Λ b =diag{λ k+1 ≥λ k+2 ≥…≥λ Q}, sort the eigenvalue matrix in descending order [λ1,λ2,…,λ k ], find the ratio Δλ between two adjacent eigenvalues 1,2 =λ1 / λ2, Δλ 2,3 =λ2 / λ3,…,Δλ k-1,k =λ k-1 / λ k , select two adjacent eigenvalues ​​maxΔλ n-1,n , the corresponding index number n is the number of source signals. The IMFs with larger correlation coefficients with the source signals are retained, and the number is the same as the estimated number of source signals.

[0102] S4: According to the reconstructed multimodal components, the one-dimensional power features are converted into a multidimensional feature matrix and input into the FastICA separation algorithm for load decomposition.

[0103] Furthermore, the load decomposition includes converting the underdetermined blind source separation problem into a positive blind source problem and inputting it into FastICA to realize load decomposition, and evaluating the decomposition model by a similarity coefficient index. The specific formula is:

[0104]

[0105] Among them, x is the real load signal, y is the separated load signal, and M is the sampling point.

[0106] S5: Perform VMD decomposition on the independent load waveform after load decomposition and extract the energy entropy of each modal component to form a multi-feature matrix which is input into the pre-trained LSTM-based non-intrusive load identification model for load identification.

[0107] Furthermore, the load identification includes, when the load electrical appliances are independently operated, obtaining power data of each electrical appliance power signal at different time periods, performing VMD decomposition, and calculating the energy E of each IMF. i and the energy entropy H of the load signal EN Construct the corresponding feature vector T = [E1, E2, ..., Ek, H EN ] Input LSTM for training, the specific expressions of energy and energy entropy are:

[0108]

[0109] Among them, P i =E i / E is the proportion of the i-th intrinsic mode IMF component to the total energy The ratio of

[0110] The load decomposition signal based on VMD-FastICA is collected, and the feature vector T' is constructed according to the same method as the input of the LSTM classifier. The type of load electrical appliance is determined through the output of the long short-term memory network classifier to identify the load.

[0111] The load identification model is evaluated by accuracy, precision, recall, and F1:

[0112]

[0113]

[0114] Among them, TP means that the state of the appliance is running and the appliance is predicted to be in the running state, FP means that the actual state of the appliance is off and the appliance is predicted to be in the running state, TN means that the state of the appliance is off and the appliance is predicted to be off, and FN means that the state of the appliance is running and the appliance is predicted to be off.

[0115] Example 2

[0116] Reference Figure 6 and Figure 7 , which is an embodiment of the present invention, provides a non-invasive load identification method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0117] The FastICA algorithm, which takes the maximization of negative entropy as the objective function, always takes the maximum negative entropy as the search direction during the iterative optimization process to achieve the separation of independent components. The negative entropy expression is:

[0118] J(y)=H(y gauss )-H(y)

[0119] In the formula, y gauss is a Gaussian signal, y is the observed signal, and both have the same covariance matrix; H(·) is the differential entropy.

[0120] In practice, when obtaining the probability density distribution of a vibration signal, it is impossible to solve the negative entropy in the form of differential entropy. The negative entropy can be approximately expressed as:

[0121] J(y) = {E[G(y)] - E[G(y)] gauss )]} 2

[0122] In the formula, E[·] represents mathematical expectation, G(·) is a nonlinear function, and the commonly used G(·) is specifically expressed as:

[0123] G1(y)=(logcosha1y) / a1

[0124]

[0125] In the formula, a1 is a constant, the selection range is between 1-2, and a1=1 is usually taken.

[0126] FastICA introduces the Lagrangian factor, constructs the Lagrangian equation, and uses Newton iteration to solve the Jacobian matrix. The iterative result expression is:

[0127] w←E{zg(w T z)}-E{g′(w T z)}w

[0128] Determine whether w converges (convergence means that the dot product of the two vectors before and after is 1). If w does not converge, continue to iterate. If it converges or meets the maximum number of iterations, output the separation signal.

[0129] The decomposition model is evaluated by the similarity coefficient index. The specific formula is:

[0130]

[0131] In the formula, x is the real load signal, y is the separated load signal, and M is the sampling point. The closer δ is to 1, the more similar the load decomposition signal is to the source signal, and the more ideal the decomposition effect is. Figure 6 Shown is a comparison diagram of the load decomposition source signal and the separated signal of a blind source separation non-intrusive load identification method adapted to demand-side management according to the present invention. It can be seen that the separated signal has a high similarity with the source signal, retains the integrity of the signal, and can further improve the load recognition rate.

[0132] When the load electrical appliances are running independently, 45 sets of power data are collected for each electrical appliance power signal at different time periods, with a sampling length of 3600. VMD decomposition is performed to calculate the energy E of each IMF. i and the energy entropy H of the load signal EN Construct the corresponding feature vector T = [E1, E2, ..., Ek, H EN ] Input LSTM for training. The specific expressions of energy and energy entropy are:

[0133]

[0134] Where P i =E i / E is the proportion of the i-th intrinsic mode IMF component to the total energy ratio.

[0135] The load decomposition signal based on VMD-FastICA is collected, and the feature vector T' is constructed according to the same method as the input of the LSTM classifier. The type of load electrical appliance is determined through the output of the long short-term memory network classifier to identify the load.

[0136] The load identification model is evaluated by accuracy, precision, recall, and F1. The specific formula is:

[0137]

[0138] In the formula, TP means that the state of an appliance is running, and the appliance is predicted to be in the running state; FP means that the actual state of an appliance is off, and the appliance is predicted to be in the running state; TN means that the state of an appliance is off, and the appliance is predicted to be off; FN means that the state of an appliance is running, and the appliance is predicted to be off. Figure 7 This is a load identification confusion diagram of a blind source separation non-intrusive load identification method adapted to demand side management implemented in the present invention. It can be seen that the overall recognition rate of each load of the present invention is relatively high.

[0139] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0140] Example 3

[0141] The third embodiment of the present invention is different from the first two embodiments in that:

[0142] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0143] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0144] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0145] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0146] Example 4

[0147] Reference Figure 8 , which is an embodiment of the present invention, provides a non-intrusive load identification system, characterized by: including a data acquisition module 1, a data preprocessing module 2, a signal decomposition module 3, a modal component reconstruction module 4, a feature conversion and load decomposition module 5, and a load identification module 6.

[0148] The data collection module 1 collects the total power consumed by the user through a smart meter or data collection device, and records the power usage data at a certain frequency.

[0149] The data preprocessing module 2 performs median filtering, de-averaging and whitening on the original data to eliminate the impact noise, redundant information and the correlation between load characteristics, thereby improving the data quality.

[0150] The signal decomposition module 3 performs variational mode decomposition on the preprocessed total power consumption and decomposes the signal into multiple intrinsic mode functions.

[0151] The modal component reconstruction module 4 selects and reconstructs the decomposed modal components according to the kurtosis criterion and singular value decomposition, and retains the most valuable load characteristic information.

[0152] The feature conversion and load decomposition module 5 converts the reconstructed multi-modal components into one-dimensional power features into a multi-dimensional feature matrix, and inputs the FastICA separation algorithm for load decomposition to separate independent load signals.

[0153] The load identification module 6 performs VMD decomposition on the independent load waveform after load decomposition, and extracts the energy entropy of each modal component to form a multi-feature matrix which is input into a pre-trained non-intrusive load identification model based on a long short-term memory (LSTM) network for load identification.

[0154] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A non-intrusive load identification method, characterized in that: include, Collect the total power consumed by users, perform median filtering, de-averaging and whitening on the raw data to eliminate impact noise in the sampled data and eliminate the correlation between redundant information and load characteristics; The total power consumption after data preprocessing is decomposed into multiple modal components according to the VMD algorithm; The decomposed multi-modal components are optimally reconstructed, and the decomposed modal components are reconstructed according to the kurtosis criterion and singular value decomposition; According to the reconstructed multi-modal components, the one-dimensional power characteristics are converted into a multi-dimensional characteristic matrix and input into the FastICA separation algorithm for load decomposition; The independent load waveform after load decomposition is subjected to VMD decomposition and the energy entropy of each modal component is extracted to form a multi-feature matrix which is input into the pre-trained LSTM-based non-intrusive load identification model for load identification.

2. A non-intrusive load identification method as claimed in claim 1, characterized in that: The collecting and processing of the raw data includes: data collection refers to collecting low-frequency data of the operating load, with a frequency of 1 Hz; According to the median filter, the impact noise of the original data and the influence of missing values ​​and abnormal values ​​caused by equipment sampling can be reduced. The specific formula is: g(x,y)=med{f(xk,y-1),(k,1)∈W} Among them, f(x, y) is the original waveform, g(x, y) is the processed image, W is the two-dimensional template, and the dimensional differences between the data are eliminated according to the centralization process to reduce the computational complexity of the objective function in the algorithm. The specific formula is: X'(t)=X(t)-E(X(t)) Among them, E(X(t)) represents the solution of the signal mean, and the correlation in the data is eliminated by whitening processing.

3. A non-intrusive load identification method as claimed in claim 2, characterized in that: The decomposition of the signal into multiple modal components includes decomposing the original power signal P(t) after data preprocessing into K intrinsic modal components IMF, which is an amplitude-frequency modulated signal, performing Hilbert transformation on each modal separation to obtain a single-sided spectrum of each component, and calculating the center frequency w corresponding to each modal function. k The exponential term aliasing modulates the single-sided spectrum of each mode to the corresponding baseband. Each modal component is around its center frequency. The sum of the modal components is equal to the input signal f(t). The minimum sum of the bandwidths of the modal components is the constraint condition. The specific variational constraint problem is: Among them, U k Each eigenmode component, w k is the center frequency corresponding to each modal component, δ(t) is the unit pulse function, u k (t) is the Kth IMF component of the decomposition; The constrained variational problem is transformed into an unconstrained variational problem to solve the objective function. VMD introduces a quadratic penalty factor α and a Lagrange multiplication operator λ. The specific expression is: Use Fourier isometric transform to transform each component into frequency domain, and iterate each modal component U k and the center frequency w k and the Lagrange multiplier λ, the updated expression is: Continue to iterate and update until the termination condition is met:

4. A non-intrusive load identification method as claimed in claim 3, characterized in that: The optimal reconstruction of the multimodal components includes determining the K value of VMD decomposition based on the kurtosis criterion, taking the maximum kurtosis as the optimization criterion, and when K=n, when the kurtosis value of the IMF component with the largest correlation coefficient with the source signal is the largest, K=n is the optimal value, and the specific expression of kurtosis is: Where μ is the mean of signal x, and σ is the standard deviation of signal x; The mutual correlation coefficient can express the similarity between the modal component IMF and the source signal. The specific expression is:

5. A non-intrusive load identification method as claimed in claim 4, characterized in that: The singular value decomposition includes estimating the number of source signals based on the singular value decomposition and constructing a sample covariance matrix Rx=E[x imf (t),x imf (t) T ], R x The specific expression of the singular value decomposition is: Among them, s are k main eigenvalues, Λ s =diag{λ1≥λ2≥…≥λ k };Λ b is Qk noise eigenvalues, Λ b =diag{λ k+1 ≥λ k+2 ≥…≥λ Q }, sort the eigenvalue matrix in descending order [λ1,λ2,…,λ k ], find the ratio Δλ between two adjacent eigenvalues 1,2 =λ1 / λ2, Δλ 2,3 =λ2 / λ3,…,Δλ k-1,k =λ k-1 / λ k , select two adjacent eigenvalues ​​maxΔλ n-1,n , corresponding to the index number n, that is, the number of source signals, the IMFs with larger correlation coefficients with the source signals are retained, and the number is the same as the estimated number of source signals.

6. A non-intrusive load identification method as claimed in claim 5, characterized in that: The load decomposition includes converting the underdetermined blind source separation problem into a positive blind source problem and inputting it into FastICA to realize load decomposition, and evaluating the decomposition model by the similarity coefficient index. The specific formula is: Among them, x is the real load signal, y is the separated load signal, and M is the sampling point.

7. A non-intrusive load identification method as claimed in claim 6, characterized in that: The load identification includes taking power data of each electrical appliance power signal in different time periods for VMD decomposition under the condition that the load electrical appliances are independently operated, and calculating the energy E of each IMF. i and the energy entropy H of the load signal EN Construct the corresponding feature vector T = [E1, E2, ..., Ek, H EN ] Input LSTM for training, the specific expressions of energy and energy entropy are: Among them, P i =E i / E is the proportion of the i-th intrinsic mode IMF component to the total energy The ratio of Collect the load decomposition signal based on VMD-FastICA, construct the feature vector T' in the same way as the input of the LSTM classifier, and determine the type of load electrical appliances through the output of the long short-term memory network classifier to identify the load; The load identification model is evaluated by accuracy, precision, recall, and F1: Among them, TP means that the state of the appliance is running and the appliance is predicted to be in the running state, FP means that the actual state of the appliance is off and the appliance is predicted to be in the running state, TN means that the state of the appliance is off and the appliance is predicted to be off, and FN means that the state of the appliance is running and the appliance is predicted to be off.

8. A system using a non-intrusive load identification method as claimed in any one of claims 1 to 7, characterized in that: It includes data acquisition module, data preprocessing module, signal decomposition module, modal component reconstruction module, feature conversion and load decomposition module, and load identification module; The data collection module collects the total power consumed by the user, and records the power usage data at a certain frequency through a smart meter or data collection device; The data preprocessing module performs median filtering, de-averaging and whitening on the original data to eliminate the impact noise, redundant information and the correlation between load characteristics, thereby improving the data quality; The signal decomposition module performs variational mode decomposition on the preprocessed total electric power to decompose the signal into multiple intrinsic mode functions; The modal component reconstruction module selects and reconstructs the decomposed modal components according to the kurtosis criterion and singular value decomposition, and retains the most valuable load characteristic information; The feature conversion and load decomposition module converts the reconstructed multi-modal components into one-dimensional power features into a multi-dimensional feature matrix, and inputs the FastICA separation algorithm to perform load decomposition to separate independent load signals; The load identification module performs VMD decomposition on the independent load waveform after load decomposition, extracts the energy entropy of each modal component, constructs a multi-feature matrix and inputs it into a pre-trained non-intrusive load identification model based on a long short-term memory network to perform load identification.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images