Power sensing data completion method and device based on nonlinear Tucker decomposition coupling

Abstract

The invention discloses an electric power perception data completion method and device based on nonlinear Tucker decomposition coupling. The method comprises the following steps: receiving an electric power perception data completion instruction; acquiring historical power sensing data based on the completion instruction; constructing a power sensing data tensor according to the collected historical power sensing data; nonlinear Tucker decomposition coupling is carried out on historical power sensing data, and a power sensing data core tensor and a hidden feature matrix are constructed; and on the basis of the power sensing data tensor, the power sensing data core tensor and the hidden feature matrix, constructing a target loss function and performing iterative optimization, and on the basis of the target loss function after iterative optimization, performing data completion on the power sensing data to obtain a power sensing data completion value. The method is used for processing the power sensing data, and high-precision data completion can be carried out based on a statistical rule, so that the problem of missing of dynamic power sensing data containing time sequence information is solved.

Description

Technical Field

[0001] The present invention relates to the field of power sensing data completion, and in particular to a power sensing data completion method based on nonlinear Tucker decomposition coupling Background Art

[0002] As the digitalization and intelligentization of power systems continue to advance, a large amount of power perception data is collected in real time from various sensors and smart devices to monitor the operating status of power systems, detect anomalies, and optimize scheduling decisions. However, in actual applications, due to sensor failures, communication delays, data loss, and other reasons, missing data often occurs in power perception data. Missing data not only affects the system's accurate assessment of the current state, but also reduces the reliability of prediction models and optimization algorithms. Therefore, how to effectively complete power perception data has become a key issue in ensuring the safe and stable operation of power systems.

[0003] At present, many researchers have proposed models based on matrix decomposition to solve the problem of power sensing data completion. These models usually construct historical power sensing data into a static two-dimensional matrix and predict and complete the missing data by decomposing the matrix. However, these methods often ignore the dynamic characteristics and time dependence of power sensing data. Power sensing data usually changes over time and is affected by multiple factors such as external environment, load fluctuations, and equipment status, showing obvious temporal and complex nonlinear characteristics. Traditional matrix decomposition methods simplify data into static structures, making it difficult to capture the trend and periodic changes of data over time, resulting in low accuracy of the completion results.

[0004] Therefore, how to introduce time series dynamics into the process of power sensing data completion, fully consider the time dimension and potential change trend of the data, and improve the prediction accuracy is a technical problem to be solved. Summary of the invention

[0005] Based on this, the technical problem to be solved by the present invention is to provide a method and device for completing power sensing data based on nonlinear Tucker decomposition coupling.

[0006] In a first aspect, an embodiment of the present application provides a method for completing power sensing data based on nonlinear Tucker decomposition coupling, comprising the following steps:

[0007] S1: receiving a completion instruction for power sensing data;

[0008] S2: Based on the completion instruction, collect historical power sensing data;

[0009] S3: constructing a power sensing data tensor according to the collected historical power sensing data;

[0010] S4: performing nonlinear Tucker decomposition coupling on the historical power sensing data to construct a core tensor and a latent feature matrix of the power sensing data;

[0011] S5: Based on the power sensing data tensor, the power sensing data core tensor and the latent feature matrix, a target loss function is constructed and iterative optimization is performed, and based on the iteratively optimized target loss function, data completion is performed on the power sensing data to obtain a power sensing data completion value.

[0012] Preferably, the historical power sensing data is stored in the form of a four-tuple, and the four-tuple is represented by Q=(l, d, p, v);

[0013] Wherein, l represents the measured parameter, d represents the measurement time point, p represents the measurement day, and v represents the power perception value of the measured parameter l measured at the measurement time point d on the measurement day p.

[0014] Preferably, constructing a power sensing data tensor according to the collected historical power sensing data includes:

[0015] S31: Divide all the four-tuples Q=(l, d, p, v) of the historical power sensing data into data of K measurement days according to the measurement day p;

[0016] S32: Quadruplet Q based on p=1 (1) =(l,d,1,v) to get Q (1) Data (l, d, v), and according to Q (1) Construct the slice matrix P with measurement day p = 1 (1) ;

[0017] S33: Select the four-tuple Q from the four-tuple Q = (l, d, p, v) in turn based on the above steps according to the remaining measurement days p (2) , Q (3) , …, Q (K) , respectively construct the slice matrix P (2) , P (3) ,…,P (K) ;

[0018] S34: Arrange the K slice matrices in order from front to back according to the divided time periods to construct a power sensing data tensor Ψ∈R I×J×K ;

[0019] Among them, P (1) The size of the slice matrix P is I rows and J columns, where I is the number of measurement parameters and J is the number of measurement time points. (1) The element P in (1) ijrepresents the power perception value of the i-th measurement parameter measured at the j-th measurement time point on the 1st measurement day, 1≤i≤I, 1≤j≤J; R represents a real number set.

[0020] Preferably, nonlinear Tucker decomposition coupling is performed on the historical power sensing data to construct a core tensor and a latent feature matrix of the power sensing data, including:

[0021] S41: constructing a power sensing data core tensor G based on the historical power sensing data;

[0022] S42: Based on the historical power sensing data, construct a latent feature matrix A representing all measurement parameters l, a latent feature matrix B representing all measurement time points d, and a latent feature matrix C representing all measurement days p.

[0023] Preferably, based on the power sensing data tensor, the power sensing data core tensor and the latent feature matrix, a target loss function is constructed and iteratively optimized, and based on the iteratively optimized target loss function, data completion is performed on the power sensing data to obtain the completed power sensing data, including:

[0024] S51: Initializing process parameters involved in the power sensing data completion process;

[0025] S52: constructing an initial target loss function for power sensing data completion according to the power sensing data tensor, the power sensing data core tensor, the latent feature matrices A, B, C and the process parameters;

[0026] S53: Iteratively optimize the initial objective function to obtain an intermediate objective loss function;

[0027] S54: Determine whether the intermediate target loss function satisfies the iteration termination condition. If so, terminate the iteration and output the intermediate target loss function as the target loss function after iterative optimization; if not, continue the iteration;

[0028] S55: completing the power sensing data based on the iteratively optimized target loss function to obtain completed power sensing data.

[0029] Preferably, the process parameters include: power sensing data tensor Ψ, power sensing data core tensor G, latent feature matrices A, B, C; latent feature dimensions M, N, O; maximum number of training iterations L; iteration number control variable r during training; and convergence termination threshold τ.

[0030] Preferably, the initial target loss function ε of power sensing data completion is constructed according to formula (1):

[0031]

[0032] in, represents the variable value when the objective function ε takes the minimum value; (i, j, k)∈Λ represents the subscript set of known elements contained in the power sensing data tensor Ψ; y ijk represents the value of the power sensing data of the measured parameter l at the measurement time point d on the measurement day p; Φ() represents the nonlinear activation function; g mno represents the element of the nth row and oth column of the mth slice in the core tensor G of the power sensing data, 1≤m≤M, 1≤n≤N, 1≤o≤O, the size of G is M×N×O, M, N and O represent the dimensions of the latent features of the measurement parameter l, the measurement time point d and the measurement day p respectively, M, N and O are positive integers; a im represents the value of the i-th row and m-th column in the latent feature matrix A of the measurement parameter; b jn represents the value of the jth row and nth column in the latent feature matrix B at the measurement time point; c ko Represents the value of the kth row and oth column in the latent feature matrix C of the measurement days;

[0033] Among them, A (i) represents the hidden feature of the i-th measurement parameter, the size of A is I×M, 1≤i≤I; B (j) represents the hidden features at the jth measurement time point, the size of B is J×N, where 1≤j≤J; C (k) Represents the latent feature of the kth measurement day, and the size of C is K×O, where 1≤k≤K.

[0034] Preferably, the Euclidean distance is used as the optimization target, and the initial target loss function is iteratively optimized using the non-negative stochastic gradient descent rule. The formula for iterative optimization is as follows:

[0035]

[0036] Wherein, (i,j,k)∈Λ, (i,j,k)∈Λ(i), (i,j,k)∈Λ(j), and (i,j,k)∈Λ(k) represent the subscript set of known elements in the power sensing data tensor Ψ, the subscript set of known elements related to i, the subscript set of known elements related to j, and the subscript set of known elements related to k, respectively; is the power sensing data completion value, expressed by formula (6):

[0037]

[0038] Preferably, the iteration termination condition is that r reaches the maximum number of training iterations L or the absolute difference between the target loss function after the current iteration and the target loss function of the previous iteration is less than τ.

[0039] In a second aspect, an embodiment of the present application provides a power sensing data completion device based on nonlinear Tucker decomposition coupling, characterized in that the device includes:

[0040] A power sensing data completion instruction receiving module receives a power sensing data completion instruction;

[0041] A power sensing data collection module, used to collect historical power sensing data based on the completion instruction;

[0042] A power sensing data tensor construction module, used to construct a power sensing data tensor according to the collected historical power sensing data;

[0043] A power sensing data nonlinear Tucker decomposition coupling module is used to perform nonlinear Tucker decomposition coupling on the historical power sensing data to construct a core tensor and a latent feature matrix of the power sensing data;

[0044] The power sensing data completion module is used to construct a target loss function and perform iterative optimization based on the power sensing data tensor, the power sensing data core tensor and the latent feature matrix, and complete the power sensing data based on the iteratively optimized target loss function to obtain a power sensing data completion value.

[0045] Compared with the prior art, the present invention has the following beneficial effects: the present invention uses historical power sensing data to construct a power sensing data tensor, and through a nonlinear Tucker decomposition coupling model, constructs a power sensing data core tensor and a latent feature matrix, and defines a target loss function to measure the difference between the observed power sensing data and the completion value. In the iterative optimization process, the non-negative random gradient descent rule is used to update the target loss function, and the data completion value is dynamically adjusted, thereby accelerating the convergence process of the model and improving the accuracy of data completion. The present invention is specifically used to process power sensing data, and can perform high-precision data completion based on statistical laws to solve the problem of missing dynamic power sensing data containing time series information. The method of the present invention can be widely used in power system monitoring, smart grid, traffic flow prediction and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] By referring to the following drawings, the exemplary embodiments of the present invention can be more completely understood. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present invention and do not constitute a limitation of the present invention. In the drawings, the same reference numerals generally represent the same parts or steps.

[0047] Figure 1A flowchart of a power sensing data completion method based on nonlinear Tucker decomposition coupling according to an exemplary embodiment of the present application;

[0048] Figure 2 It is a structural schematic diagram of a power sensing data completion device based on nonlinear Tucker decomposition coupling according to an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0049] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0050] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by technicians in the field to which this application belongs.

[0051] In addition, the terms "first" and "second" etc. are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices.

[0052] Example 1

[0053] The embodiment of the present application provides a method for completing power sensing data based on nonlinear Tucker decomposition coupling, comprising the following steps:

[0054] S1: receiving a completion instruction for power sensing data;

[0055] Specifically, the server may send instructions to the power sensing data acquisition module by periodically sending instructions, or based on notification from the device, or based on notification from a server, requesting the sending of instructions to complete the power sensing data.

[0056] S2: Based on the completion instruction, collect historical power sensing data;

[0057] Specifically, the historical power sensing data is stored in the form of a four-tuple, which is represented by Q = (l, d, p, v);

[0058] Wherein, l represents the measured parameter, d represents the measurement time point, p represents the measurement day, and v represents the power perception value of the measured parameter l measured at the measurement time point d on the measurement day p.

[0059] S3: construct a power sensing data tensor based on the collected historical power sensing data;

[0060] Specifically, step S3 includes:

[0061] S31: Divide all the quaternary groups Q=(l, d, p, v) of the historical power sensing data into data of K measurement days according to the measurement day p;

[0062] S32: Quadruplet Q based on p=1 (1) =(l,d,1,v) to get Q (1) Data (l, d, v), and according to Q (1) Construct the slice matrix P with measurement day p = 1 (1) ;

[0063] S33: Select the four-tuple Q from the four-tuple Q = (l, d, p, v) in turn based on the above steps according to the remaining measurement days p (2) , Q (3) , …, Q (K) , respectively construct the slice matrix P (2) , P (3) ,…,P (K) ;

[0064] S34: Arrange the K slice matrices in order from front to back according to the divided time periods to construct the power sensing data tensor Ψ∈R I×J×K ;

[0065] Among them, P (1) The size of the slice matrix P is I rows and J columns, where I is the number of measurement parameters and J is the number of measurement time points. (1) The element P in (1) ij represents the power perception value of the i-th measurement parameter measured at the j-th measurement time point on the 1st measurement day, 1≤i≤I, 1≤j≤J; R represents a real number set.

[0066] S4: Perform nonlinear Tucker decomposition coupling on historical power sensing data to construct the core tensor and latent feature matrix of power sensing data;

[0067] Specifically, step S4 includes:

[0068] S41: Based on the historical power sensing data, construct the power sensing data core tensor G;

[0069] S42: Based on the historical power sensing data, construct a latent feature matrix A representing all measurement parameters l, a latent feature matrix B representing all measurement time points d, and a latent feature matrix C representing all measurement days p.

[0070] Among them, g mno A represents the element of the nth row and oth column of the mth slice in the core tensor G of the power sensing data, and the size of G is M×N×O, where M, N, and O are all positive integers, representing the dimensions of the latent features of the measurement parameter l, the measurement time point d, and the measurement day p, respectively. (i) represents the hidden feature of the i-th measurement parameter, and the size of A is I×M, where 1≤i≤I, and M is a positive integer, representing the dimension of the hidden feature of the measurement parameter l. (j) represents the latent feature at the jth measurement time point, and the size of B is J×N, where 1≤j≤J, and N is a positive integer representing the dimension of the latent feature at the measurement time point d.

[0071] S5: Based on the power sensing data tensor, the power sensing data core tensor and the latent feature matrix, a target loss function is constructed and iteratively optimized, and the power sensing data is completed based on the iteratively optimized target loss function to obtain a power sensing data completion value.

[0072] Specifically, step S5 includes:

[0073] S51: Initializing process parameters involved in the power sensing data completion process;

[0074] Specifically, initialize the power sensing data tensor Ψ, the power sensing data core tensor G, and the latent feature matrices A, B, and C; initialize the factor matrix latent feature dimensions M, N, and O; initialize the maximum number of training iterations L; initialize the iteration number control variable r during training; and initialize the convergence termination threshold τ.

[0075] In this embodiment, the specific operations of initializing each parameter are as follows:

[0076] (1) The latent feature dimension M of the factor matrix determines the latent feature space dimension of the latent feature matrix A and is initialized to a positive integer;

[0077] (2) The latent feature dimension N of the factor matrix determines the latent feature space dimension of the latent feature matrix B and is initialized to a positive integer;

[0078] (3) The latent feature dimension O of the factor matrix determines the latent feature space dimension of the latent feature matrix C, which is initialized to a positive integer;

[0079] (4) The size of the core tensor G is determined by the dimensions M, N, and O of the latent feature matrix; the sizes of the latent feature matrices A, B, and C are determined by the values ​​of each dimension of the corresponding power sensing data tensor Ψ and the latent feature dimensions M, N, and O of the factor matrix, that is, A is a factor matrix with I rows and M columns, B is a factor matrix with J rows and N columns, and C is a factor matrix with K rows and O columns. The core tensor and the three latent feature matrices are initialized with small random positive numbers respectively;

[0080] (5) The maximum number of training iterations L is a variable that controls the upper limit of the iteration process and is initialized to a large positive integer;

[0081] (6) The iteration number control variable r is initialized to 0;

[0082] (7) The convergence termination threshold τ is a parameter used to determine whether the iterative process has converged and is initialized with a very small positive number.

[0083] S52: constructing an initial target loss function for power sensing data completion according to the power sensing data tensor, the power sensing data core tensor, the latent feature matrix and the process parameters;

[0084] In this embodiment, according to the core tensor G and three latent feature matrices A, B, C created by the constructed power sensing data tensor Ψ and the power sensing data nonlinear Tucker decomposition coupling module and the related parameters involved in the initialized power sensing data completion process, the target loss function ε is constructed for the known data set Λ of the power sensing data tensor Ψ to train and obtain the latent eigenvalues ​​of the core tensor G and the three latent feature matrices A, B, C of the historical power sensing data, and the initial target loss function ε of the power sensing data completion is constructed according to formula (1):

[0085]

[0086] in, represents the variable value when the objective function ε takes the minimum value; (i, j, k)∈Λ represents the subscript set of known elements contained in the power sensing data tensor Ψ; y ijk represents the value of the power sensing data of the measured parameter l at the measurement time point d on the measurement day p; Φ() represents the nonlinear activation function; g mno represents the element of the nth row and oth column of the mth slice in the core tensor G of the power sensing data, 1≤m≤M, 1≤n≤N, 1≤o≤O, the size of G is M×N×O, M, N and O represent the dimensions of the latent features of the measurement parameter l, the measurement time point d and the measurement day p respectively, M, N and O are positive integers; a im represents the value of the i-th row and m-th column in the latent feature matrix A of the measurement parameter; b jn represents the value of the jth row and nth column in the latent feature matrix B at the measurement time point; c koRepresents the value of the kth row and oth column in the latent feature matrix C of the measurement days;

[0087] Among them, A (i) represents the hidden feature of the i-th measurement parameter, the size of A is I×M, 1≤i≤I; B (j) represents the hidden features at the jth measurement time point, the size of B is J×N, where 1≤j≤J; C (k) Represents the latent feature of the kth (k)th measurement day, and the size of C is K×O, where 1≤k≤K.

[0088] S53: Iteratively optimize the initial objective function to obtain an intermediate objective loss function;

[0089] In this embodiment, the Euclidean distance is used as the optimization target. To ensure the non-negativity in the update process, the target loss function ε is iteratively optimized using the non-negative stochastic gradient descent rule to minimize the value of the target loss function ε. The formula for training iteration is as follows:

[0090]

[0091]

[0092] Among them, (i,j,k)∈Λ, (i,j,k)∈Λ(i), (i,j,k)∈Λ(j) and (i,j,k)∈Λ(k) respectively represent the subscript set of known elements in the power sensing data tensor Ψ, the subscript set of known elements related to i, the subscript set of known elements related to j and the subscript set of known elements related to k.

[0093] S54: Determine whether the intermediate target loss function satisfies the iteration termination condition. If so, terminate the iteration and output the intermediate target loss function as the target loss function after iterative optimization; if not, continue the iteration;

[0094] In this step, there are two situations in which the loss function ε reaches the termination condition during the training iteration process on Λ. The first is that for each iteration of ε, the value of the training iteration control variable r is increased by 1. When the value of r reaches the maximum training iteration number L, ε stops training; the second is that during the ε training process, when the absolute difference between the ε value calculated after the end of this round of iteration and the ε value of the previous round is less than the convergence termination threshold τ, ε stops training.

[0095] S55: completing the power sensing data based on the iteratively optimized target loss function to obtain completed power sensing data.

[0096] Specifically, when the target loss function ε converges on the known data set Λ, the core tensor and three latent feature matrices A, B, and C obtained by training when the target loss function ε reaches the minimum value are used to calculate the power sensing data completion value of the highest accuracy measurement parameter l at the measurement time point d on the measurement day p It is expressed by formula (6):

[0097]

[0098] Example 2

[0099] In the above embodiment, a method is provided, and correspondingly, the present application also provides a device. The device provided in the embodiment of the present application can implement the above method, and the device can be implemented by software, hardware, or a combination of software and hardware. For example, the device may include integrated or separate functional modules or units to perform the corresponding steps in the above methods.

[0100] In some implementations of the embodiments of the present application, the device 10 provided in the embodiments of the present application is based on the same inventive concept as the method provided in the aforementioned embodiments of the present application and has the same beneficial effects.

[0101] Please refer to Figure 2 , which shows a schematic diagram of a device provided by some embodiments of the present application. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described below is only illustrative.

[0102] like Figure 2 As shown, the apparatus 10 may include:

[0103] A power sensing data completion instruction receiving module 110 receives a power sensing data completion instruction;

[0104] A power sensing data collection module 120, configured to collect historical power sensing data based on the completion instruction;

[0105] A power sensing data tensor construction module 130, configured to construct a power sensing data tensor according to the collected historical power sensing data;

[0106] The power sensing data nonlinear Tucker decomposition coupling module 140 is used to perform nonlinear Tucker decomposition coupling on the historical power sensing data to construct a core tensor and a latent feature matrix of the power sensing data;

[0107] The power sensing data completion module 150 is used to construct a target loss function and perform iterative optimization based on the power sensing data tensor, the power sensing data core tensor and the latent feature matrix, and to complete the power sensing data based on the iteratively optimized target loss function to obtain a power sensing data completion value.

[0108] The power sensing data acquisition module 120 includes a power sensing data storage unit 121 and a power sensing supplementary data storage unit 122. The power sensing data storage unit 121 is used to store the received historical power sensing data in the form of a quaternary. The representation of the quaternary is Q = (l, d, p, v), where l represents the measurement parameter, d represents the measurement time point, p represents the measurement day, and v represents the power sensing value measured at the measurement parameter l at the measurement time point d on the measurement day p. The power sensing supplementary data storage unit 122 is used to store the supplemented power sensing data, wherein the supplemented data is also stored in the form of a quaternary.

[0109] In this embodiment, the power sensing data completion module 150 includes an initialization unit 151 , a training unit 152 , and a power sensing data completion value calculation unit 153 .

[0110] Among them, the initialization unit 151 is used to initialize the relevant parameters involved in the power sensing data completion process. The relevant parameters have been explained in the above-mentioned embodiments and will not be repeated here.

[0111] Among them, the training unit 152 is used to combine the constructed initial power sensing data tensor, the initial power sensing data core tensor, the initial latent feature matrix and the relevant parameters initialized by the initialization unit 151 to train and calculate the core tensor Ψ and latent feature matrices A, B, and C of the power sensing data.

[0112] The power sensing data supplement value calculation unit 153 is used to calculate the power sensing data supplement value according to the core tensor and latent feature matrix of the historical power sensing data obtained by the training unit 152. Then the calculated supplement data is stored in the power sensing supplement data storage unit 122.

[0113] The iterative optimization process and parameter-related definitions involved in this embodiment have been explained in the previous embodiments and will not be repeated here.

[0114] It should be noted that the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0115] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0116] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0117] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0118] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0119] 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 application, 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, which is stored in a storage medium and includes 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 application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application, and they should all be included in the scope of the claims and specification of the present application.

Claims

1. A power sensing data completion method based on nonlinear Tucker decomposition coupling, characterized in that: The steps include: S1: receiving a completion instruction for power sensing data; S2: Based on the completion instruction, collect historical power sensing data; S3: constructing a power sensing data tensor according to the collected historical power sensing data; S4: performing nonlinear Tucker decomposition coupling on the historical power sensing data to construct a core tensor and a latent feature matrix of the power sensing data; S5: Based on the power sensing data tensor, the power sensing data core tensor and the latent feature matrix, a target loss function is constructed and iterative optimization is performed, and based on the iteratively optimized target loss function, data completion is performed on the power sensing data to obtain a power sensing data completion value.

2. The method according to claim 1, characterized in that The historical power sensing data is stored in the form of a four-tuple, and the four-tuple is represented by Q=(l, d, p, v); Wherein, l represents the measured parameter, d represents the measurement time point, p represents the measurement day, and v represents the power perception value of the measured parameter l measured at the measurement time point d on the measurement day p.

3. The method according to claim 2, characterized in that The step of constructing a power sensing data tensor according to the collected historical power sensing data includes: S31: Divide all the four-tuples Q=(l, d, p, v) of the historical power sensing data into data of K measurement days according to the measurement day p; S32: Quadruplet Q based on p=1 (1) =(l,d,1,v) to get Q (1) Data (l, d, v), and according to Q (1) Construct the slice matrix P with measurement day p = 1 (1) ; S33: Select the four-tuple Q from the four-tuple Q = (l, d, p, v) in turn based on the above steps according to the remaining measurement days p (2) , Q (3) ,…,Q (K) , respectively construct the slice matrix P (2) , P (3) ,…,P (K) ; S34: Arrange the K slice matrices in order from front to back according to the divided time periods to construct a power sensing data tensor Ψ∈R I×J×K ; Among them, P (1) The size of the slice matrix P is I rows and J columns, where I is the number of measurement parameters and J is the number of measurement time points. (1) The element P in (1) ij represents the power perception value of the i-th measurement parameter measured at the j-th measurement time point on the 1st measurement day, 1≤i≤I, 1≤j≤J; R represents a real number set.

4. The method according to claim 3, characterized in that The historical power sensing data is subjected to nonlinear Tucker decomposition coupling to construct a core tensor and a latent feature matrix of the power sensing data, including: S41: constructing a power sensing data core tensor G based on the historical power sensing data; S42: Based on the historical power sensing data, construct a latent feature matrix A representing all measurement parameters l, a latent feature matrix B representing all measurement time points d, and a latent feature matrix C representing all measurement days p.

5. The method according to claim 4, characterized in that Based on the power sensing data tensor, the power sensing data core tensor and the latent feature matrix, a target loss function is constructed and iteratively optimized, and based on the iteratively optimized target loss function, data completion is performed on the power sensing data to obtain the completed power sensing data, including: S51: Initializing process parameters involved in the power sensing data completion process; S52: constructing an initial target loss function for power sensing data completion according to the power sensing data tensor, the power sensing data core tensor, the latent feature matrices A, B, C and the process parameters; S53: Iteratively optimize the initial objective function to obtain an intermediate objective loss function; S54: Determine whether the intermediate target loss function satisfies the iteration termination condition. If so, terminate the iteration and output the intermediate target loss function as the target loss function after iterative optimization; if not, continue the iteration; S55: completing the power sensing data based on the iteratively optimized target loss function to obtain completed power sensing data.

6. The method according to claim 5, characterized in that The process parameters include: power sensing data tensor Ψ, power sensing data core tensor G, latent feature matrices A, B, C; latent feature dimensions M, N, O; maximum training iteration number L; iteration number control variable r during training; and convergence termination threshold τ.

7. The method according to claim 6, characterized in that According to formula (1), the initial target loss function ε of power sensing data completion is constructed: in, represents the variable value when the objective function ε takes the minimum value; (i, j, k)∈Λ represents the subscript set of known elements contained in the power sensing data tensor Ψ; y ijk represents the value of the power sensing data of the measured parameter l at the measurement time point d on the measurement day p; Φ() represents the nonlinear activation function; g mno represents the element of the nth row and oth column of the mth slice in the core tensor G of the power sensing data, 1≤m≤M, 1≤n≤N, 1≤o≤O, the size of G is M×N×O, M, N and O represent the dimensions of the latent features of the measurement parameter l, the measurement time point d and the measurement day p respectively, M, N and O are positive integers; a im represents the value of the i-th row and m-th column in the latent feature matrix A of the measurement parameter; b jn represents the value of the jth row and nth column in the latent feature matrix B at the measurement time point; c ko Represents the value of the kth row and oth column in the latent feature matrix C of the measurement days; Among them, A (i) represents the hidden feature of the i-th measurement parameter, the size of A is I×M, 1≤i≤I; B (j) represents the hidden features at the jth measurement time point, the size of B is J×N, where 1≤j≤J; C (k) Represents the latent feature of the kth measurement day, and the size of C is K×O, where 1≤k≤K.

8. The method according to claim 7, characterized in that Taking the Euclidean distance as the optimization target, the initial target loss function is iteratively optimized using the non-negative stochastic gradient descent rule. The iterative optimization formula is as follows: Wherein, (i,j,k)∈Λ, (i,j,k)∈Λ(i), (i,j,k)∈Λ(j), and (i,j,k)∈Λ(k) represent the subscript set of known elements in the power sensing data tensor Ψ, the subscript set of known elements related to i, the subscript set of known elements related to j, and the subscript set of known elements related to k, respectively; is the power sensing data completion value, expressed by formula (6):

9. The method according to claim 8, characterized in that The iteration termination condition is that r reaches the maximum number of training iterations L or the absolute difference between the target loss function after the current iteration and the target loss function of the previous iteration is less than τ.

10. A power sensing data completion device based on nonlinear Tucker decomposition coupling, characterized in that: The device includes: A power sensing data completion instruction receiving module receives a power sensing data completion instruction; A power sensing data collection module, used to collect historical power sensing data based on the completion instruction; A power sensing data tensor construction module, used to construct a power sensing data tensor according to the collected historical power sensing data; A power sensing data nonlinear Tucker decomposition coupling module is used to perform nonlinear Tucker decomposition coupling on the historical power sensing data to construct a core tensor and a latent feature matrix of the power sensing data; The power sensing data completion module is used to construct a target loss function and perform iterative optimization based on the power sensing data tensor, the power sensing data core tensor and the latent feature matrix, and complete the power sensing data based on the iteratively optimized target loss function to obtain a power sensing data completion value.

Images