A method and device for compressing and fusing power distribution data

Abstract

This invention relates to a method and apparatus for compressing and fusing power distribution data. The method comprises the following steps: decomposing the original power distribution information data using a parallelized EEMD algorithm; symbolizing the decomposed principal component power distribution information data time series based on feature-based symbolic clustering approximation (SAX); reconstructing the power distribution information data from the symbolized time series to complete the cleaning of the collected original power distribution information data; compressing the power distribution information data using the SPIHT compression algorithm; and classifying and fusing the compressed power distribution information data using a multi-input single-output network structure. This invention enables accurate collection of power distribution information based on the cleaning, compression, and fusion of the original power distribution information data, fully leveraging the value of the information data and providing important scientific basis for power system operation.

Description

Technical Field

[0001] The present invention belongs to the technical field of power distribution information data collection and processing, and relates to key data processing methods such as decomposition, time series symbolization processing, reconstruction, compression, classification and fusion of original power distribution information data, especially a power distribution data compression and fusion method and device. Background Art

[0002] With the advancement of the construction of ubiquitous power Internet of Things, the scale of the power system is getting larger and larger. A large number of smart meters and their supporting monitoring equipment have been put into use. A wide variety of power big data are collected in a timely manner, including data, audio data and image data in the power production process. These data run through all aspects of power production. They are interconnected and complementary, and together constitute power big data. Power big data is generated in all aspects of power production and contains huge potential value. Power big data is the application of big data concepts, technologies and methods to the power industry, and is one of the important components of big data. Power big data involves power generation, transmission, substation, distribution, power consumption and dispatching. It is a collection of cross-unit, cross-professional and cross-business data analysis and mining, as well as data visualization. Power big data consists of structured data and unstructured data. Structured data mainly includes various real-time power data and historical data that can be monitored by smart meters. Unstructured data is not limited to power data, but also includes audio data, image data collected by equipment monitoring systems, and customer service information of customer service centers. These various types of data and information together constitute power big data. With the application of energy Internet technology and the construction of smart grids, power big data is showing a rapid development trend. There are more and more application environments for big data, and further research is needed on the application of big data in distribution networks, so as to provide guarantees for the application of big data technology in smart grids. Power big data technology is the core to support the safe and stable operation of smart grids.

[0003] In recent years, big data technology has received great attention and importance from governments of various countries and the global academic and industrial circles, and its application in various industries has developed rapidly. In March 2012, the US government issued the "Big Data Research, Development and Innovation Program", and the Chinese government, academic and industrial circles also paid great attention to big data. As far as the power industry is concerned, under the operation mode of smart grid, the amount of data generated is massive. In recent years, data resources have begun to grow rapidly and have formed a certain scale, and there is also a strong demand for big data technology, which is specifically manifested in the development of data volume from TB level to PB level, and high-performance data storage and high scalability face challenges. At present, the existing problems are: the business is developing in the direction of intelligence and leanness, but the processing and real-time performance of complex data are not ideal; the cross-business and cross-platform data processing and analysis capabilities are insufficient; with the development and improvement of the power big data cloud platform, the efficiency and application capabilities of data collection, classification, storage, analysis of power big data are not high. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a method and device for compressing and fusing power distribution data, which can accurately collect power distribution information on the basis of cleaning, compressing and fusing the original power distribution information data, fully explore the value of information data, and provide an important scientific basis for the operation of the power system.

[0005] The present invention solves the technical problem by adopting the following technical solutions:

[0006] A method for compressing and fusing power distribution data comprises the following steps:

[0007] Step 1: Decompose the original power distribution and consumption information data through the parallelized EEMD algorithm to obtain the time series of the main component power distribution and consumption information data;

[0008] Step 2: symbolize the main component distribution power consumption information data time series decomposed in step 1 by using feature-based symbol clustering approximation SAX;

[0009] Step 3, reconstructing the distribution and consumption information data of the main component distribution and consumption information time series after the symbolization processing in step 2, and completing the cleaning of the collected original distribution and consumption information data;

[0010] Step 4: compressing the power distribution and utilization information data cleaned in step 3;

[0011] Step five, classify and fuse the main component power consumption information data compressed in step four.

[0012] Moreover, the specific steps of decomposing the original power distribution and utilization information by using the parallelized EEMD algorithm in step 1 include:

[0013] (1) Set the overall signal X(t), which is composed of the original power signal x(t) superimposed with a group of Gaussian white noise signals v(t);

[0014] (2) According to the set signal X(t), the boundary of the adaptive window is selected based on the AWS algorithm;

[0015] (3) Determine the extension cost η of the above window, and search the extension cost table based on the local stationarity α of the signal at the segment boundary;

[0016] (4) According to the extension cost η, each segment is extended and the waveform record is converted into a waveform data matrix that can be processed by MapReduce as the input of the Map task. The waveform record is segmented using the extended boundary points as the window boundaries, and each segmented record segment is used as a row of the waveform data matrix.

[0017] (5) Divide the distribution and utilization information data in the waveform record matrix into segments and parallelly obtain extreme values. In the Map stage, the computing node receives the data segments in the waveform record matrix as input, and parallelly judges the extreme points of the records in the segments point by point, and records the extreme points; in the Reduce stage, it receives the local extreme value fragments output by the Map task and performs connection operations, outputting the maximum value matrix spmax and the minimum value matrix spmin;

[0018] (6) Solve the envelope of the segmented signal in parallel. The algorithm receives the output spmax and spmin of step (5) as input. In the Map stage, the Mapper task of each data node uses the cubic spline interpolation algorithm to solve the envelope of each segment in parallel, and cuts the envelope according to the previous extension cost η; in the Reduce stage, the received envelope segments are connected and reconstructed, and the output of the algorithm is the upper envelope and lower envelope of the entire signal.

[0019] (7) Solve the envelope mean and repeat steps (2) to (6) to obtain the IMF components of each stage;

[0020] (8) Add different white noise v(t) to the original signal and repeat the above steps; the IMF component a corresponding to the original signal m (t) is shown in formula (1):

[0021]

[0022] Where M represents the number of IMF components;

[0023] (9) The final decomposition result of the original signal x(t) is shown in formula (2):

[0024]

[0025] Among them, r m (t) is the IMF component, H is the number of residual components, a m (t) is the residual component.

[0026] Moreover, the specific steps of the step 1 using the feature-based symbolic clustering approximation (SAX) to symbolize the decomposed 96-point main component distribution power consumption information data time series include:

[0027] (1) First, the decomposed main component distribution power consumption information data time series is normalized, that is, the data is converted into time series data with an average value of 0 and a standard deviation of 1, recorded as D = [d(1), d(2), ..., d(t)], and the normalization formula is:

[0028]

[0029] Where μ is the mean value of all values ​​in the time series data, δ is the standard deviation of all values, x(i) is the original time series data, and d(i) is the normalized time series data.

[0030] (2) The normalized data is then reduced in dimension and processed using the PAA algorithm D = [d(1), d(2), …, d(t)] to obtain the segmented cumulative approximation of D: Where T is the number of segments represented by the time series PAA, that is:

[0031]

[0032] Among them, d j ′ is the data after normalization in the previous step, and T is the maximum number of acquisitions divided by the acquisition period;

[0033] Finally Discretization is performed, that is, the time series represented by PAA is mapped to a finite character set: according to the size of the selected letter set, the interval splitting point β is found in the Gaussian distribution table, and Map to q intervals of equal probability, use the same character to represent the same interval, so as to convert the PAA mapping into the corresponding string, and finally obtain the discretized target string

[0034] Moreover, the specific steps of reconstructing the power distribution and utilization information data of the power distribution and utilization information time series after the symbolization in step 1 include:

[0035] (1) Initialize the parameters of the model. Set the empty set matrix Q to store the basis vectors corresponding to the recovery matrix Z, set the empty set O to store the subscript value of the selected recovery matrix Z, the recovery matrix Y = ΦK, the initial value of the sparse representation coefficient S = 0, the residual R = Y, and the number of iterations b = 0.

[0036] (2) Select the basis vectors of the fast complex matrix T, and select the basis vectors with the largest inner product with the residual ε.

[0037] (3) Solve the sparse representation coefficient S, and use the selected recovery matrix T basis vector to sparsely represent the symbolized power distribution information time series data, and solve its sparse report coefficient vector

[0038] (4) Update residual

[0039] (5) Determine whether the number of iterations has reached the maximum or whether the error condition is met. If so, stop the calculation; if not, continue with step (2).

[0040] Moreover, the specific steps of compressing the cleaned power distribution information data based on the SPIHT compression algorithm in step 2 include:

[0041] (1) Calculate the high-frequency coefficients and their total threshold C in the reconstructed information data matrix, find the largest high-frequency coefficient, and solve the minimum binary digit k corresponding to the high-frequency coefficient, where k = log2C;

[0042] (2) Divide the sub-bands where the high-frequency coefficients are located into diagonal blocks, horizontal blocks, and vertical blocks, and set the thresholds corresponding to the diagonal, horizontal, and vertical blocks;

[0043] (3) A full scan is performed on the subband before the Kth band of high-frequency coefficients (K is one-third of the number of wavelet coefficient decomposition levels), and the subband after the Kth generation is scanned horizontally in blocks; the current threshold of the high-frequency coefficient is compared with the thresholds of all blocks of the subband before the Kth generation. If the total threshold is greater than the block threshold of the subband, the block is not scanned.

[0044] (4) further finely scan the high-frequency coefficients, and output the kth (k=log2C) most important bit of each element except the important element detected by scanning in step (3);

[0045] (5) Update the current total threshold. If C <C min , C min is the lower limit of the set threshold, then stop the current encoding; if the current total threshold C ≥ C min , then the current total threshold is changed to 1 / 2 of its original value, and the process returns to step (3) for the next sequential scan and fine scan.

[0046] Moreover, the classification and fusion module of the compressed distribution and utilization power collection information data using the multi-input-single-output network structure in step 2 includes:

[0047] (1) Select the user power distribution information data samples that need to be classified and fused, set the samples whose expected output meets the current object to 1, and set the expected output that does not belong to the object to 0, and learn the approximation by selecting a neural network model with a multi-input and single-output mode - the Hermite orthogonal basis neural network.

[0048] (2) The user power distribution information data samples that need to be classified and fused are used as input variables of the trained neural network. According to the output results of each neural network, the unknown user power distribution information data samples are mapped to the corresponding categories.

[0049] (3) Statistically analyze the final classification results and classify and identify the electricity consumption information samples of unknown users.

[0050] The advantages and positive effects of the present invention are:

[0051] 1. After completing the cleaning process of the power distribution and utilization information data, the present invention implements a data cleaning process flow with low complexity and high accuracy based on the characteristics of the power distribution and utilization information data of the power system.

[0052] 2. The present invention can adaptively determine the boundaries and extension lengths of signal segments based on the parallel EEMD algorithm, maintains the characteristics of the original EEMD algorithm, and realizes fast EEMD decomposition, which is suitable for processing high sampling rate signals such as the current ubiquitous power Internet of Things distribution and utilization information.

[0053] 3. The present invention analyzes the specific collection process, network topology architecture and characteristics of power distribution information, and after receiving the sparse compressed sampling perception data, constructs a reconstruction model of the sparse compressed sampling perception data of the power distribution information data, and then completes the reconstruction and analysis of the sparse compressed sampling perception data, thereby improving the collection efficiency of the power distribution information Internet of Things.

[0054] 4. The present invention combines the Map Reduce model to realize the classification and fusion parallel computing of large-scale data sets, achieves the purpose of data classification and fusion processing, and meets the real-time and parallel processing requirements of power data. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0056] The embodiments of the present invention are further described in detail below with reference to the accompanying drawings:

[0057] A method for compressing and fusing power distribution data, the innovation of which is: Figure 1 As shown, the following steps are included:

[0058] Step 1: Use the parallelized EEMD algorithm to decompose the original power distribution information data, and symbolize the decomposed main component power distribution information data time series based on the feature-based symbolic clustering approximation (SAX). Then, reconstruct the power distribution information data after symbolization, and complete the cleaning of the collected original power distribution information data.

[0059] The specific method of using the parallelized EEMD algorithm to decompose the original power distribution and utilization information data in step 1 is:

[0060] (1) Set the overall signal X(t), which is composed of the original power signal x(t) superimposed with a group of Gaussian white noise signals v(t);

[0061] (2) According to the set signal X(t), the boundary of the adaptive window is selected based on the AWS algorithm;

[0062] (3) Determine the extension cost η of the above window, and search the extension cost table based on the local stationarity α of the signal at the segment boundary;

[0063] (4) According to the extension cost η, each segment is extended and the waveform record is converted into a waveform data matrix that can be processed by MapReduce as the input of the Map task. The waveform record is segmented using the extended boundary points as the window boundaries, and each segmented record segment is used as a row of the waveform data matrix.

[0064] (5) Divide the distribution and utilization information data in the waveform record matrix into segments and parallelly obtain extreme values. In the Map stage, the computing node receives the data segments in the waveform record matrix as input, and parallelly judges the extreme points of the records in the segments point by point, and records the extreme points; in the Reduce stage, it receives the local extreme value fragments output by the Map task and performs connection operations, outputting the maximum value matrix spmax and the minimum value matrix spmin;

[0065] (6) Solve the envelope of the segmented signal in parallel. The algorithm receives the output spmax and spmin of step (5) as input. In the Map stage, the Mapper task of each data node uses the cubic spline interpolation algorithm to solve the envelope of each segment in parallel, and cuts the envelope according to the previous extension cost η; in the Reduce stage, the received envelope segments are connected and reconstructed, and the output of the algorithm is the upper envelope and lower envelope of the entire signal.

[0066] (7) Solve the envelope mean and repeat steps (2) to (6) to obtain the IMF components of each stage;

[0067] (8) Add different white noise v(t) to the original signal and repeat the above steps; the IMF component a corresponding to the original signal m (t) is shown in formula (1):

[0068]

[0069] Where M represents the number of IMF components;

[0070] (9) The final decomposition result of the original signal x(t) is shown in formula (2):

[0071]

[0072] Among them, r m (t) is the IMF component, H is the number of residual components, a m (t) is the residual component.

[0073] The specific method of performing symbolic processing on the decomposed main component distribution power consumption information data time series based on the symbolic clustering approximation (SAX) of the feature in step 1 is as follows:

[0074] (1) First, the decomposed main component distribution power consumption information data time series is normalized, that is, the data is converted into time series data with an average value of 0 and a standard deviation of 1, recorded as D = [d(1), d(2), ..., d(t)], and the normalization formula is:

[0075]

[0076] Where μ is the mean value of all values ​​in the time series data, δ is the standard deviation of all values, x(i) is the original time series data, and d(i) is the normalized time series data.

[0077] (2) The normalized data is then reduced in dimension and processed using the PAA algorithm D = [d(1), d(2), …, d(t)] to obtain the segmented cumulative approximation of D: Where T is the number of segments represented by the time series PAA, that is:

[0078]

[0079] Among them, d j ′ is the data after normalization in the previous step, and T is the maximum number of acquisitions divided by the acquisition period;

[0080] Finally Discretization is performed, that is, the time series represented by PAA is mapped to a finite character set: according to the size of the selected letter set, the interval splitting point β is found in the Gaussian distribution table, and Map to q intervals of equal probability, use the same character to represent the same interval, so as to convert the PAA mapping into the corresponding string, and finally obtain the discretized target string

[0081] The specific method of reconstructing the power distribution and utilization information data of the symbolized power distribution and utilization information time series in step 1 is:

[0082] (1) Initialize the parameters of the model. Set the empty set matrix Q to store the basis vectors corresponding to the recovery matrix Z, set the empty set O to store the subscript value of the selected recovery matrix Z, the recovery matrix Y = ΦK, the initial value of the sparse representation coefficient S = 0, the residual R = Y, and the number of iterations b = 0.

[0083] (2) Select the basis vectors of the fast complex matrix T, and select the basis vectors with the largest inner product with the residual ε.

[0084] (3) Solve the sparse representation coefficient S, and use the selected recovery matrix T basis vector to sparsely represent the symbolized power distribution information time series data, and solve its sparse report coefficient vector

[0085] (4) Update residual

[0086] (5) Determine whether the number of iterations has reached the maximum or whether the error condition is met. If so, stop the calculation; if not, continue with step (2).

[0087] Step 2: compress the power distribution and utilization information data based on the SPIHT compression algorithm, and use the multi-input-single-output network structure to complete the classification and fusion of the compressed power distribution and utilization information data.

[0088] like Figure 1 As shown, the specific steps of step 2 for compressing the power distribution information data based on the SPIHT compression algorithm are as follows:

[0089] (1) Calculate the high-frequency coefficients and their total threshold C in the reconstructed information data matrix, find the largest high-frequency coefficient, and solve the minimum binary digit k corresponding to the high-frequency coefficient, where k = log2C;

[0090] (2) Divide the sub-bands where the high-frequency coefficients are located into diagonal blocks, horizontal blocks, and vertical blocks, and set the thresholds corresponding to the diagonal, horizontal, and vertical blocks;

[0091] (3) A full scan is performed on the subband before the Kth band of high-frequency coefficients (K is one-third of the number of wavelet coefficient decomposition levels), and the subband after the Kth generation is scanned horizontally in blocks; the current threshold of the high-frequency coefficient is compared with the thresholds of all blocks of the subband before the Kth generation. If the total threshold is greater than the block threshold of the subband, the block is not scanned.

[0092] (4) further finely scan the high-frequency coefficients, and output the kth (k=log2C) most important bit of each element except the important element detected by scanning in step (3);

[0093] (5) Update the current total threshold. If C <C min , C min is the lower limit of the set threshold, then stop the current encoding; if the current total threshold C ≥ C min , then the current total threshold is changed to 1 / 2 of its original value, and the process returns to step (3) for the next sequential scan and fine scan.

[0094] like Figure 1 As shown, the specific steps of step 2 using the multi-input-single-output network structure to complete the classification and fusion of the compressed distribution and utilization information data are as follows:

[0095] (1) Select the user power distribution information data samples that need to be classified and fused, set the samples whose expected output meets the current object to 1, and set the expected output that does not belong to the object to 0, and learn the approximation by selecting a neural network model with a multi-input and single-output mode - the Hermite orthogonal basis neural network.

[0096] (2) The user power distribution information data samples that need to be classified and fused are used as input variables of the trained neural network. According to the output results of each neural network, the unknown user power distribution information data samples are mapped to the corresponding categories.

[0097] (3) Statistically analyze the final classification results and classify and identify the electricity consumption information samples of unknown users.

[0098] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0099] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0100] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0101] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for compressing and fusing power distribution data, characterized in that: The following steps are involved: Step 1: Decompose the original power distribution and consumption information data to obtain the time series of the main power distribution and consumption information data; Step 2: symbolize the time series of the main component distribution power consumption information data decomposed in step 1; Step 3, reconstructing the distribution and consumption information data of the main component distribution and consumption information time series after the symbolization processing in step 2, and completing the cleaning of the collected original distribution and consumption information data; Step 4: compressing the power distribution and utilization information data cleaned in step 3; Step 5: classify and fuse the main component distribution power consumption information data compressed in step 4; The method for decomposing the original power distribution and utilization information data specifically comprises the following steps: Step 1: Set up the overall signal , the overall signal From the original power signal Superimpose a set of Gaussian white noise signals composition; Step 2: According to the set signal , select the boundary of the adaptive window based on the AWS algorithm; Step 3: Determine the extension cost of the above window , local stationarity of the signal based on the segment boundaries , find the extension cost table; Step 4: According to the extension cost , extend each segment, convert the waveform record into a waveform record matrix that MapReduce can process as the input of the Map task; use the extended boundary points as the window boundary to segment the waveform record, and use each segmented record segment as a row of the waveform data matrix to obtain the segmented signal; Step 5, divide the distribution and utilization information data in the waveform record matrix into segments and parallelly obtain the extreme value; in the Map stage, the computing node receives the data segments in the waveform record matrix as input, and parallelly judges the extreme value points of the records in the segments point by point, and records the extreme value points; in the Reduce stage, the local extreme value fragments output by the Map stage task are received and connected, and the maximum value matrix is ​​output spmax And the minimum matrix spmin ; Step 6: Parallelize the envelope of the segmented signal obtained in step 4. The algorithm receives the outputs spmax and spmin of step 5 as input. In the Map stage, the Mapper task of each data node uses the cubic spline interpolation algorithm to solve the envelope of each segment in parallel, and according to the previous extension cost , the envelope is cut; in the Reduce stage, the received envelope segments are connected and reconstructed, and the output of the algorithm is the upper envelope and lower envelope of the entire signal; Step 7, solve the envelope mean according to the upper envelope and the lower envelope in step 6, repeat steps 2 to 6 to obtain the IMF components of each stage; Step 8: Add different white noise to the original signal Repeat the above steps to get the IMF component under noise: Step 9: Using the white noise mean of zero, average the IMF components in step 8 to obtain the IMF components corresponding to the original signal. The calculation of is shown in formula (1): (1) Where M represents the number of IMF components; Step 10, original signal Finally, it can be decomposed into the sum of an IMF component and the signal residual component. The decomposition result is shown in formula (2): (2) Among them, r m (t) is the IMF component, H is the number of residual components, a m (t) is the residual component.

2. The method for compressing and fusing power distribution data according to claim 1, characterized in that: The method for symbolizing the time series of the decomposed main component distribution power consumption information data specifically comprises the following steps: Step 1: Convert the data into time series data with a mean of 0 and a standard deviation of 1, recorded as , the normalized formula is: (3) In the formula is the average value of all values ​​in the time series data, is the standard deviation of all values, is the original time series data, is the normalized time series data; Step 2: Reduce the dimension of the normalized data and process it using the PAA algorithm ,get The piecewise accumulation is approximately expressed as ,in is the number of segments represented by the time series PAA, that is: (4) in, is the data after normalization in the previous step, and T is the maximum number of acquisitions divided by the acquisition period; Step 3: Reduce the dimension of the data Discretization is performed, that is, the time series represented by PAA is mapped to a finite character set: according to the size of the selected letter set, the interval split point is found in the Gaussian distribution table ,Will Mapping to equal probability q intervals, and use the same character to represent the same interval, so as to convert the PAA mapping into the corresponding string, and finally obtain the discretized target string .

3. The method for compressing and fusing power distribution data according to claim 1, characterized in that: The method of reconstructing the distribution and use information data of the main component distribution and use information time series after symbolization processing to complete the cleaning of the collected original distribution and use information data includes: Step 1: Set the empty set matrix Q to store the basis vectors corresponding to the recovery matrix Z, set the empty set set O to store the subscript value of the selected recovery matrix Z, and restore the matrix , the initial value of the sparse representation coefficient S=0, the residual R=Y, the number of iterations b =0; Step 2: According to the residual in step 1, determine the basis vectors of the fast complex matrix T and select The basis vector with the largest inner product; Step 3, solve the sparse representation coefficient S, use the recovery matrix T basis vector selected in step 2 to represent the sparse representation coefficient of the symbolized power distribution information time series data, and solve its sparse reporting coefficient vector ; Step 4: Update the residual using the solution from step 3 ; Step 5, determine whether the number of iterations has reached the maximum or whether the error condition is met. If it has reached the maximum or the error meets the standard, stop the calculation; if not, continue to execute steps 2 to 4.

4. The method for compressing and fusing power distribution data according to claim 1, characterized in that: The SPIHT compression algorithm is used to compress the cleaned power distribution information data, which includes the following steps: Step 1: Calculate the high-frequency coefficients and their total thresholds in the reconstructed information data matrix C , find the largest high-frequency coefficient and solve the minimum binary digit corresponding to the high-frequency coefficient k ,in k =log2 C ; Step 2, dividing the sub-band where the high frequency coefficients are located into diagonal blocks, horizontal blocks, and vertical blocks, and determining the thresholds corresponding to the diagonal blocks, horizontal blocks, and vertical blocks; Step 3: high frequency coefficients K The sub-band in front of the band adopts full scanning mode. K Subband scanning horizontal block after generation; compare the current threshold of high frequency coefficients with K The thresholds of all blocks in the previous subband are compared. If the total threshold is greater than the block threshold of the subband, the block is not scanned. K It is one third of the number of wavelet coefficient decomposition levels; Step 4: further scan the high-frequency coefficients, and output the first k important positions, among which k =log2 C ; Step 5: Update the current total threshold. If C < C min , C min If it is the lower limit of the set threshold, the current encoding is stopped; If the current total threshold C ≥ C min , then the current total threshold is changed to 1 / 2 of its original value, and the process returns to step 3 for the next sequential scan and fine scan.

5. The method for compressing and fusing power distribution data according to claim 1, characterized in that: The method for classifying and fusing the compressed main component distribution power consumption information data is completed using a multi-input-single output network structure, and the specific steps include: Step 1: Select user power distribution and utilization information data samples that need to be classified and fused; Step 2: Set the expected output of the user's power distribution information data sample that meets the current object to 1, and set the expected output that does not belong to the object to 0, and learn the approximation by selecting a neural network model with a multi-input-single-output mode - the Hermite orthogonal basis neural network; Step 2: Use the user power distribution information data samples that need to be classified and fused as input variables of the trained neural network, and map the unknown user power distribution information data samples to corresponding categories according to the output results of each neural network; Step 3: Statistically analyze the final classification results and classify and identify the electricity consumption information samples of unknown users.

6. A power distribution data compression and fusion device based on the power distribution data compression and fusion method according to any one of claims 1 to 5, characterized in that: It includes time series acquisition module, symbolization processing module, cleaning processing module, compression processing module and classification fusion module. The time series acquisition module is used to decompose the original power distribution and consumption information data to obtain the time series of the main component power distribution and consumption information data; The symbolic processing module is used to symbolize the time series of the decomposed main component distribution power consumption information data; The cleaning processing module is used to reconstruct the distribution and use information data of the main component distribution and use information time series after symbolization processing, and complete the cleaning of the collected original distribution and use information data; The compression processing module is used to compress the cleaned power distribution information data; The classification and fusion module is used to classify and fuse the compressed main component distribution power consumption collection information data.

7. The device for compressing and fusing power data according to claim 6, characterized in that: The time series acquisition module is used to decompose the original power distribution information data, and the decomposition method adopts the parallelized EEMD algorithm.

8. The device for compressing and fusing power data according to claim 6, characterized in that: The method for symbolizing the time series of the decomposed main component distribution power consumption information data by the symbolic processing module uses feature-based symbol clustering approximation to symbolize the time series. Perform symbolic clustering approximation of features.

9. The device for compressing and fusing power data according to claim 6, characterized in that: The compression processing module is used to compress the cleaned power distribution information data using the SPIHT compression algorithm.

Images