Power distribution network operation data quality improvement method and system considering spatial-temporal characteristics

Abstract

The invention relates to the technical field of novel power distribution network operation data detection, and particularly discloses a power distribution network operation data quality improvement method and system considering spatial-temporal characteristics, and the method comprises the steps: carrying out the abnormal data detection of operation data through employing an improved DBSCAN clustering algorithm; obtaining and preprocessing operation data of the power distribution network in a historical time period, and inputting the obtained two-dimensional feature map into the trained time prediction model to obtain a time feature prediction value of the to-be-detected time period; acquiring spatial positions of all devices in the data source measurement device area, and predicting to obtain a spatial feature prediction value of the to-be-detected time period based on the data value of the adjacent measurement device; calculating a space-time predicted value of the to-be-detected time period based on the time feature predicted value and the space feature predicted value; and replacing the abnormal data points in the to-be-detected time period by using the electric quantity prediction data at the corresponding moment. Information in the data can be captured more comprehensively, and the method is more suitable for processing long-sequence data.

Description

Technical Field

[0001] The present invention relates to the field of novel distribution network operation data detection technology, and in particular to a distribution network operation data quality improvement method and system taking into account spatiotemporal characteristics. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the advancement of the construction of new power systems, the distribution network is evolving rapidly. A large number of decentralized distributed power sources, electric vehicles, energy storage and other new sources and loads are connected to the distribution network on a large scale, and the demand for intelligent interaction between new distribution networks and users is gradually emerging.

[0004] To ensure the transparency of the new distribution network, three-phase voltage, current, power, frequency and other operating data are used in state estimation, fault location and other links. Obtaining accurate and effective operating data is the key to ensuring the reliable operation of the distribution network, and it is also the data basis for improving distribution network monitoring or fault prediction.

[0005] The quality of distribution network operation data includes three aspects: data accuracy, real-time and reliability, which are used to characterize the ability of measured data to reflect the actual working status of the power grid. Poor data quality will lead to incorrect judgments on the operating status of the power grid and incorrect decision-making. The quality of distribution network operation data is affected by many factors, such as: new energy equipment in the new distribution network (such as photovoltaic power generation and wind power generation) may be affected by natural factors such as sunshine and weather; the load of some electrical appliances will also fluctuate due to seasonal changes. In addition, the operation data of the new distribution network may also experience abnormal phenomena such as mutation, missing or long-term fluctuations during the collection and transportation process. All of the above situations may lead to the randomness and abnormality of the distribution network operation data.

[0006] Therefore, accurately detecting abnormal operation data of the new distribution network and effectively replacing it is the key to ensuring the quality of the operation data of the new distribution network. Abnormal data refers to data that deviates from the normal measurement range, which may be caused by various reasons such as measurement equipment failure or data transmission problems.

[0007] The prior art discloses a method for detecting abnormal data of a photovoltaic power station, which selects photovoltaic residual data whose corresponding density value and distance value meet preset conditions and uses them as the cluster center point of at least one photovoltaic residual data; clusters the photovoltaic residual data based on the cluster center point, and determines the photovoltaic residual data that does not belong to any cluster in the clustering result as abnormal data. The prior art discloses a method for analyzing electricity consumption based on machine learning, which constructs a time series decomposition model to model the time series relationship of the input electricity consumption data sequence, extracts the characteristics of the electricity consumption data for prediction, obtains the electricity consumption prediction value corresponding to each moment as the comparison value for abnormal identification; calculates the residual of the comparison value at each moment and the actual value of the electricity consumption data, and uses the improved K-means clustering algorithm to cluster the residual and identify abnormal data.

[0008] The above-mentioned method for detecting abnormal data only considers the temporal characteristics of the data, but does not consider the spatial characteristics of different measuring devices, which has certain limitations. Summary of the invention

[0009] In order to solve the above problems, the present invention proposes a method and system for improving the quality of distribution network operation data taking into account the temporal and spatial characteristics, by adopting an improved density-based clustering algorithm (DBSCAN) to detect abnormal data on the operation data; an ultra-short-term prediction method based on variational mode decomposition (VMD) feature extraction and hybrid neural network is adopted to obtain the temporal characteristic prediction value of the data; the spatial characteristic prediction value of the data is obtained by using the spatial position information of the measuring device, and finally the detection and completion of abnormal and missing values ​​are realized.

[0010] In some embodiments, the following technical solutions are adopted:

[0011] A method for improving the quality of distribution network operation data considering spatiotemporal characteristics, comprising:

[0012] Obtaining the distribution network operation data for the time period to be detected and performing preprocessing;

[0013] The improved DBSCAN clustering algorithm is used to detect abnormal data in the operating data;

[0014] The distribution network operation data of the historical time period is obtained and preprocessed, and multiple intrinsic mode functions (IMFs) are extracted using the VMD algorithm, and the multiple IMFs are converted into two-dimensional feature graphs; the obtained two-dimensional feature graphs are input into the trained time prediction model to obtain the time feature prediction value of the time period to be detected;

[0015] Obtain the spatial positions of all devices in the data source measurement device area, combine the line length and number of nodes between the measurement devices, and predict the spatial feature prediction value of the time period to be detected based on the data values ​​of the adjacent measurement devices;

[0016] Based on the time feature prediction value and the space feature prediction value, the time and space prediction value of the time period to be detected is calculated;

[0017] The abnormal data points in the time period to be detected are replaced with the power forecast data at the corresponding moment.

[0018] As an optional solution, an improved DBSCAN clustering algorithm is used to detect abnormal data in the running data, specifically:

[0019] Select a sample point p from an unvisited running data set p(t) i , check its setting neighborhood range E ps If the neighborhood contains at least a set number of sample points, then the sample point p i is the core point, and it forms a cluster with all points in the neighborhood;

[0020] The above method is used to determine whether each point meets the core point conditions. If so, all sample points in its neighborhood are added to the cluster of the center point with each core point as the center; if it does not meet the core point conditions, but the sample point is in the neighborhood of other clusters, it is added to the cluster; if it neither meets the core point conditions nor is in the neighborhood of a cluster, it is marked as an outlier.

[0021] As an optional solution, multiple IMFs are converted into two-dimensional feature maps, specifically:

[0022] For each set of feature vectors a i After VMD decomposition, we can get t subsequences of length τ, named imf1 to imf t , spliced ​​into a two-dimensional feature map F of size t*τ in the order of numbers i , replacing the original eigenvector a i .

[0023] As an optional scheme, the time prediction model includes a convolutional neural network layer, a bidirectional long short-term memory network layer, an attention mechanism layer and a fully connected output layer arranged in sequence; the input data is subjected to local feature extraction through the convolutional neural network layer, and the extracted local features enter the bidirectional long short-term memory network layer, and the information of the entire time series is output; the output of the bidirectional long short-term memory network layer is weighted and summed through the attention mechanism layer, and the output of the attention mechanism layer passes through the fully connected output layer to generate the final time feature prediction result.

[0024] As an optional solution, the output of the bidirectional long short-term memory network layer is weighted summed through the attention mechanism layer, specifically:

[0025] The output matrix Y of the bidirectional long short-term memory network layer outScore and get the correlation score matrix:

[0026]

[0027] The correlation score matrix is ​​converted into a specific weight value through the softmax function to obtain the weight of each element:

[0028]

[0029] According to the weight of each element, perform weighted summation on each element of the output matrix;

[0030] Among them, V a 、M a and U a are the learnable weight matrices, Q is the query matrix, E is the query matrix Q and the output matrix Y out The correlation score matrix between them, tanh is the activation function; e n is the element in the correlation score matrix E, y tn The output matrix Y out The elements in .

[0031] As an optional solution, based on the data value of the adjacent measurement device, the predicted value of the spatial feature of the time period to be detected is predicted, specifically:

[0032]

[0033] L = l1 + l2 + l3 + l4;

[0034] Wherein, l1, l2, l3 and l4 represent the line distance between the current measuring device and each adjacent measuring device respectively; A (t), p B (t), p D (t) and p E (t) represents the measurement values ​​of the four measuring devices A, B, D, and E at time t respectively.

[0035] As an optional solution, a weighted sum of the time feature prediction value and the space feature prediction value is performed to obtain the space-time prediction value of the time period to be detected.

[0036] In other embodiments, the following technical solutions are adopted:

[0037] A distribution network operation data quality improvement system considering spatiotemporal characteristics, comprising:

[0038] A data acquisition module is used to acquire the distribution network operation data of the time period to be detected and perform preprocessing;

[0039] An abnormal data detection module is used to detect abnormal data on running data using an improved DBSCAN clustering algorithm;

[0040] The time feature prediction module is used to obtain and pre-process the distribution network operation data in the historical time period, extract multiple intrinsic mode functions IMFs using the VMD algorithm, and convert the multiple IMFs into two-dimensional feature graphs; input the obtained two-dimensional feature graphs into the trained time prediction model to obtain the time feature prediction value of the time period to be detected;

[0041] The spatial feature prediction module is used to obtain the spatial positions of all devices in the data source measurement device area, and predict the spatial feature prediction value of the time period to be detected based on the data values ​​of the adjacent measurement devices in combination with the line length and the number of nodes between the measurement devices;

[0042] A spatiotemporal prediction module, used to calculate the spatiotemporal prediction value of the time period to be detected based on the time feature prediction value and the spatial feature prediction value;

[0043] The abnormal data replacement module is used to replace the abnormal data points in the time period to be detected with the power forecast data at the corresponding moment.

[0044] In other embodiments, the following technical solutions are adopted:

[0045] A terminal device includes a processor and a memory, the processor is used to implement instructions; the memory is used to store multiple instructions, and the instructions are suitable for the processor to load and execute the above-mentioned distribution network operation data quality improvement method considering time and space characteristics.

[0046] In other embodiments, the following technical solutions are adopted:

[0047] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the above-mentioned method for improving the quality of distribution network operation data taking into account spatiotemporal characteristics.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) The present invention adopts an improved DBSCAN clustering algorithm to detect abnormal data on the running data. By adjusting the coordinates of the input data, it can avoid the failure of the density-based clustering algorithm caused by small data fluctuations.

[0050] (2) The temporal prediction model of the present invention combines the convolutional neural network (CNN) with the bidirectional long short-term memory (LSTM) network layer, and introduces the attention mechanism, combining the local feature extraction capability of CNN with the long-term feature extraction capability of LSTM, and dynamically focusing on important features through the attention mechanism, thereby improving the accuracy of the prediction and the robustness of the model. This method can capture the information in the data more comprehensively and is more suitable for processing long sequence data.

[0051] (3) The present invention calculates the spatiotemporal prediction value based on the time feature prediction value and the space feature prediction value, and fully considers the time feature and space feature of the data when predicting the data. The time feature can reflect the change of the data in a short period of time and capture the information of the previous monitoring values. If it only relies on the time feature, the accuracy of the data within a certain time range before the monitoring point must be guaranteed, which will lead to higher requirements for time series data. Therefore, the present invention introduces the spatial feature, which can reflect the change trend of the operating data within a certain area at the same time.

[0052] (4) The present invention uses VMD to extract time series data features, and calculates the time feature prediction value through the time prediction model; combined with the line length and the number of nodes between each measuring device, the data values ​​of the corresponding time and type of the adjacent measuring devices are used as the basis for weighted prediction to obtain the spatial feature prediction value. Based on the time feature prediction value and the spatial feature prediction value, the spatiotemporal prediction value is calculated; when predicting data, the time and spatial features of the data are fully considered. Compared with other feature extraction algorithms, the VMD algorithm used in the present invention can avoid phenomena such as modal aliasing, and the hybrid neural network can accurately extract the time series features of the data; the spatial feature value weighted by the distance of the measuring device can capture the features of the adjacent device at the same time. Weighting the time prediction value and the spatial prediction value can take into account the spatiotemporal features at the same time, and adjust the proportion of the two according to data requirements.

[0053] Other features and advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a flow chart of a method for improving the quality of distribution network operation data taking into account spatiotemporal characteristics in an embodiment of the present invention;

[0055] Figure 2 A schematic diagram of a DBSCAN clustering algorithm in an embodiment of the present invention;

[0056] FIG3(a) and FIG3(b) respectively show the original data before coordinate adjustment and the data schematic diagram after coordinate adjustment;

[0057] Figure 4 This is a schematic diagram of VMD decomposition results in an embodiment of the present invention;

[0058] Figure 5 Schematic diagram of the spatial position of the measuring device in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0060] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0061] Embodiment 1

[0062] In one or more embodiments, a method for improving the quality of distribution network operation data considering spatiotemporal characteristics is disclosed, combined with Figure 1 , specifically including the following process:

[0063] S101: Acquire the distribution network operation data of the time period to be detected and perform preprocessing;

[0064] In this embodiment, operation data of the time period to be measured is obtained from a distribution network area, such as voltage, current, active power, etc., to form an overall data sample. In this embodiment, active power is used as an example for explanation, and the processing process of other types of operation data is the same.

[0065] S102: Use the improved DBSCAN clustering algorithm to detect abnormal data on the operating data.

[0066] This embodiment uses an improved DBSCAN clustering algorithm to detect abnormal data on the operating data for active power data within a time period to be tested, such as active power data within three months.

[0067] Assume that the power data set to be processed is p(t)=[p1,p2,…,p D ] represents a set of active power time series data of size 1*D. The DBSCAN clustering algorithm is used to perform cluster analysis on the data, dividing the data into several clusters and outliers. These clusters are aggregated from normal data, and the outliers are the abnormal data identified by the DBSCAN clustering algorithm.

[0068] Combination Figure 2 The specific implementation process of the DBSCAN clustering algorithm in this embodiment is as follows:

[0069] S1021: Select a sample point p in p(t) that has not been visited i , check its setting neighborhood range E ps All sample points in the neighborhood are core points if they contain at least Minpts sample points (Minpts is a set value), and they form a cluster with all points in the neighborhood.

[0070] S1022: Repeat step S1021 for all sample points to determine whether each point meets the core point condition. If so, take each core point as the center and add all sample points in its neighborhood to the cluster of the center point.

[0071] S1023: If the sample point does not meet the core point condition, but is located in the neighborhood of another cluster, it is treated as a boundary point and added to the cluster. If the sample point does not meet the core point condition and is not in the neighborhood of a cluster, it is marked as an outlier.

[0072] S103: Obtain and pre-process the distribution network operation data in the historical time period, extract multiple intrinsic mode functions IMFs using the VMD algorithm, and convert the multiple IMFs into two-dimensional feature graphs.

[0073] In this embodiment, p(t) is time series data, and a two-dimensional graph is drawn for the data of p(t) with the monitoring point number as the horizontal coordinate and the value as the vertical coordinate. However, if the vertical coordinate interval is not selected properly, the abnormal values ​​between the data cannot be distinguished, and the DBSCAN algorithm fails.

[0074] Therefore, this embodiment draws a two-dimensional graph as input for the data p(t); in this embodiment, p(t) is a 1*D row vector, and a vertical coordinate Y is set for each data d :

[0075]

[0076] Among them, d is the horizontal coordinate number in p(t), and the outlier value P is screened out in p(t) o .P max and P min are the maximum and minimum values ​​in the data set p(t), respectively; D represents the number of horizontal axes.

[0077] By adjusting the range of the vertical axis, the data can be better dispersed to avoid the problem of density concentration caused by small data fluctuations, which ultimately leads to the failure of the density-based clustering algorithm. Figure 3 (a) and Figure 3 (b) respectively show the original data before adjustment and the schematic diagram of the data after adjustment.

[0078] In order to establish the prediction value of the time feature, it is necessary to build a time prediction model. Therefore, this embodiment selects the time series active data x(t) within three months as the training set for building the time prediction model. Since there are outliers in the time series data set x(t), Lagrange interpolation is used to replace the outliers to obtain the original data sequence.

[0079] The specific process of using Lagrange interpolation to replace outliers is as follows:

[0080] First, find the Lagrange interpolation basis function:

[0081]

[0082] On this basis, the Lagrange interpolation polynomial is obtained as follows:

[0083]

[0084] For outliers x in time series data x(t) i , select three data before and after it, a total of six data [x i-3 ,x i-2 ,x i-1 ,x i+1 ,x i+2 ,x i+3 ]. Assume that the seven data containing the outlier are the function values ​​of function f on the horizontal coordinates 1 to 7. Based on these six data, construct function f, satisfying f(l) = x i-3 , f(2) = x i-2 , f(3) = x i-1 , f(5) = x i+1 , f(6) = x i+2 , f(7) = x i+3 .

[0085] Lagrange interpolation basis function l k (x) can be expressed as:

[0086]

[0087] The Lagrange interpolation polynomial function f can be expressed as:

[0088] f(x)=l1x i-3 +l2x i-2 +l3x i-1 +l5xi+1 +l6x i+2 +l7x i+3 (5)

[0089] After obtaining the approximate expression of f, f(4) can be regarded as x i After filling the outliers in the PMU data x(t), the original data sequence x for time series prediction is obtained. pre (t).

[0090] After the above processing, the original data sequence x pre (t) There are no outliers anymore, which can be used as the basis for training the prediction model. pre (t) Reconstruct the predicted input set A and output set B, and extract time domain features based on VMD. Reconstruct the original data sequence into n prediction samples.

[0091]

[0092] Among them, x pre.i Represents the original data sequence x pre (t) the i-th data, a1 represents the first set of input feature vectors, which are used to predict the first output value b1; n represents the number of samples in the prediction model, τ represents the data window length used for training each sample, that is, the number of features. Using the data in the range of (t-τ, t-1), the data at time t can be predicted.

[0093] Combination Figure 4 After reconstructing the input data set A and the output data set B, for each set of feature vectors a i Perform VMD decomposition. Each set of features can be decomposed into t intrinsic mode functions and residual mode components.

[0094] For each set of feature vectors a i After VMD decomposition, we can get t subsequences of length τ, named imf1 to imf t , spliced ​​into a two-dimensional feature map F of size t*τ in the order of numbers i , replacing the original eigenvector a i .

[0095] S104: Input the obtained two-dimensional feature map into the trained time prediction model to obtain the time feature prediction value of the time period to be detected.

[0096] In this embodiment, each two-dimensional feature map F i As input, corresponding to b i As output, combined with the attention mechanism, the time prediction model is trained to realize the time series prediction function.

[0097] The architecture of the time prediction model specifically includes:

[0098] (1) CNN layer: First, the input data set F (F is all F i The convolutional layer is also called a filter, which can divide the input data into specific parts and further extract features, namely:

[0099]

[0100] Where l is the serial number of the network layer. l is the output of each layer, W l and b l Characterizes weights and biases. g is an activation function, and this embodiment uses ReLU as the activation function.

[0101]

[0102] The output of CNN is a feature map after convolution operation. In the conventional CNN architecture, a pooling layer needs to be inserted to compress the feature dimension to avoid overfitting. However, the application of pooling layer in time series is prone to information loss. Therefore, this embodiment directly adds a bidirectional long short-term memory network (BiLSTM) layer after the convolution layer.

[0103] (2) BiLSTM layer: The BiLSTM layer is used to capture the long-term dependencies of the time series. By processing the time series forward and backward, BiLSTM can consider both past and future information. The output of the CNN enters the BiLSTM layer and outputs the information of the entire time series.

[0104] In the BiLSTM network structure, the conversion equation is as follows:

[0105]

[0106] Among them, i(t) represents the input gate; f(t) represents the forget gate; o(t) represents the output gate; c(t) represents the storage unit; h(t) represents the hidden state; σ represents the logical sigmoid function; ⊙ represents the element-wise multiplication operation; the forget gate controls the amount of storage erased from each storage unit, the input gate controls the amount of update of each unit, and the output gate controls the exposure of the internal storage state.

[0107] The forward hidden state matrix is ​​obtained as Where N m is the length of the feature sequence output by CNN. Similarly, the inverse hidden state matrix H is obtained bw , the final output matrix of the BiLSTM layer is Y out =[H fw;H bw ].

[0108] (3) Attention mechanism layer: Weight the output of BiLSTM to emphasize the most important time points for prediction. First, the matrix Y out Score and obtain the correlation score matrix.

[0109]

[0110] Where E is the query matrix Q and Y out The correlation score matrix between a 、M a and U a is a learnable weight matrix; Q is the currently set query vector, and tanh is the activation function.

[0111] The score matrix is ​​converted into a specific weight value through the softmax function. The weight of each element is:

[0112]

[0113] Among them, e n Represents the elements in E. The output is calculated based on the weight distribution:

[0114]

[0115] Among them, y tn is the matrix Y out The elements in .

[0116] (4) Fully connected output layer: takes the weighted sum of the output of the attention mechanism layer as input and generates the final regression prediction based on the output of the attention layer.

[0117] After the time prediction model is constructed as described above, the distribution network operation data of the historical time period is used as input to obtain the time characteristic prediction value P of the time period to be detected. time .

[0118] S105: Acquire the spatial positions of all devices in the data source measurement device area, combine the line lengths and the number of nodes between the measurement devices, and predict the spatial feature prediction value of the time period to be detected based on the data values ​​of the adjacent measurement devices.

[0119] In this embodiment, the corresponding data of the neighboring devices are used as the features of the spatial prediction value, and the distance between the selected neighboring devices and the target device is no more than two devices. Figure 5 , assuming that the active power data p(t) comes from the measurement device C, the selected neighboring devices include A, B, D and E.

[0120] The spatial characteristic prediction value P of active power in device C space It can be calculated as:

[0121]

[0122] Wherein, l1, l2, l3 and l4 represent the line distance between the current measuring device C and each adjacent measuring device A, B, D and E respectively. A (t), p B (t), p D (t) and p E (t) represents the measurement values ​​of the four measuring devices A, B, D, and E at time t respectively.

[0123] S106: Predicting value P based on time characteristics time and spatial feature prediction value P space , calculate the spatiotemporal prediction value of the time period to be detected; the details are as follows:

[0124] P predict =ξP space +(1-ξ)P time (13)

[0125] Among them, ξ is a proportional factor, which is used to adjust the proportion between the time and space feature prediction values ​​and is preset according to different scenarios.

[0126] S107: Replace the abnormal data points in the time period to be detected with the power forecast data at the corresponding time.

[0127] This embodiment uses an improved DBSCAN clustering algorithm to detect abnormal data on the operating data. An ultra-short-term prediction method based on VMD feature extraction and CNN-BiLSTM hybrid neural network is used to obtain the time feature prediction value of the data; the spatial feature prediction value of the data is obtained by using the spatial position information of the measuring device, and finally the abnormal and missing values ​​are supplemented to improve the data quality.

[0128] Embodiment 2

[0129] In one or more embodiments, a distribution network operation data quality improvement system considering spatiotemporal characteristics is disclosed, comprising:

[0130] A data acquisition module is used to acquire the distribution network operation data of the time period to be detected and perform preprocessing;

[0131] An abnormal data detection module is used to detect abnormal data on running data using an improved DBSCAN clustering algorithm;

[0132] The time feature prediction module is used to obtain and pre-process the distribution network operation data in the historical time period, extract multiple intrinsic mode functions IMFs using the VMD algorithm, and convert the multiple IMFs into two-dimensional feature graphs; input the obtained two-dimensional feature graphs into the trained time prediction model to obtain the time feature prediction value of the time period to be detected;

[0133] The spatial feature prediction module is used to obtain the spatial positions of all devices in the data source measurement device area, and predict the spatial feature prediction value of the time period to be detected based on the data values ​​of the adjacent measurement devices in combination with the line length and the number of nodes between the measurement devices;

[0134] A spatiotemporal prediction module, used to calculate the spatiotemporal prediction value of the time period to be detected based on the time feature prediction value and the spatial feature prediction value;

[0135] The abnormal data replacement module is used to replace the abnormal data points in the time period to be detected with the power forecast data at the corresponding moment.

[0136] The specific implementation of the above modules is the same as that in Example 1 and will not be described in detail.

[0137] Embodiment 3

[0138] In one or more embodiments, a terminal device is disclosed, including a server, the server including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method for improving the quality of distribution network operation data considering spatiotemporal characteristics in embodiment 1 when executing the program. For the sake of brevity, it will not be described in detail here.

[0139] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0140] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0141] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software.

[0142] Embodiment 4

[0143] In one or more embodiments, a computer-readable storage medium is disclosed, in which a plurality of instructions are stored, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the method for improving the quality of distribution network operation data considering spatiotemporal characteristics described in Example 1.

[0144] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for improving the quality of distribution network operation data considering spatiotemporal characteristics, characterized in that: include: Obtaining the distribution network operation data for the time period to be detected and performing preprocessing; The improved DBSCAN clustering algorithm is used to detect abnormal data in the operating data; The distribution network operation data of the historical time period is obtained and preprocessed, and multiple intrinsic mode functions (IMFs) are extracted using the VMD algorithm, and the multiple IMFs are converted into two-dimensional feature graphs; the obtained two-dimensional feature graphs are input into the trained time prediction model to obtain the time feature prediction value of the time period to be detected; Obtain the spatial positions of all devices in the data source measurement device area, combine the line length and number of nodes between the measurement devices, and predict the spatial feature prediction value of the time period to be detected based on the data values ​​of the adjacent measurement devices; Based on the time feature prediction value and the space feature prediction value, the time and space prediction value of the time period to be detected is calculated; The abnormal data points in the time period to be detected are replaced with the power forecast data at the corresponding moment.

2. A method for improving the quality of distribution network operation data considering spatiotemporal characteristics as claimed in claim 1, characterized in that: The improved DBSCAN clustering algorithm is used to detect abnormal data in the running data, specifically: Select a sample point p from an unvisited running data set p(t) i , check its setting neighborhood range E ps If the neighborhood contains at least a set number of sample points, then the sample point p i is the core point, and it forms a cluster with all points in the neighborhood; The above method is used to determine whether each point meets the core point condition. If it does, all sample points in its neighborhood are added to the cluster of the center point with each core point as the center. If it does not meet the core point condition, but the sample point is in the neighborhood of other clusters, it is added to the cluster. If a point does not meet the requirements of a core point and is not in the neighborhood of a cluster, it is marked as an outlier.

3. A method for improving the quality of distribution network operation data considering spatiotemporal characteristics as claimed in claim 1, characterized in that: Convert multiple IMFs into two-dimensional feature maps, specifically: For each set of feature vectors a i After VMD decomposition, we can get t subsequences of length τ, named imf1 to imf t , spliced ​​into a two-dimensional feature map F of size t*τ in the order of numbers i , replacing the original eigenvector a i .

4. A method for improving the quality of distribution network operation data considering spatiotemporal characteristics as claimed in claim 1, characterized in that: The time prediction model includes a convolutional neural network layer, a bidirectional long short-term memory network layer, an attention mechanism layer and a fully connected output layer which are arranged in sequence; the input data is subjected to local feature extraction through the convolutional neural network layer, the extracted local features enter the bidirectional long short-term memory network layer, and the information of the entire time series is output; the output of the bidirectional long short-term memory network layer is weighted and summed through the attention mechanism layer, the output of the attention mechanism layer passes through the fully connected output layer, and the final time feature prediction result is generated.

5. A method for improving the quality of distribution network operation data considering spatiotemporal characteristics as claimed in claim 4, characterized in that: The output of the bidirectional long short-term memory network layer is weighted and summed through the attention mechanism layer, specifically: The output matrix Y of the bidirectional long short-term memory network layer out Score and get the correlation score matrix: The correlation score matrix is ​​converted into a specific weight value through the softmax function to obtain the weight of each element: According to the weight of each element, perform weighted summation on each element of the output matrix; Among them, V a 、M a and U a are the learnable weight matrices, Q is the query matrix, E is the query matrix Q and the output matrix Y out The correlation score matrix between them, tanh is the activation function; e n is the element in the correlation score matrix E, y tn The output matrix Y out The elements in .

6. A method for improving the quality of distribution network operation data considering spatiotemporal characteristics as claimed in claim 1, characterized in that: Based on the data values ​​of the nearby measuring devices, the predicted spatial feature values ​​of the time period to be detected are predicted as follows: L＝l1+l2+l3+l4； Wherein, l1, l2, l3 and l4 represent the line distance between the current measuring device and each adjacent measuring device respectively; A (t), p B (t), p D (t) and p E (t) represents the measurement values ​​of the four measuring devices A, B, D, and E at time t respectively.

7. A method for improving the quality of distribution network operation data considering spatiotemporal characteristics as claimed in claim 1, characterized in that: The time feature prediction value and the space feature prediction value are weighted and summed to obtain the space-time prediction value of the time period to be detected.

8. A distribution network operation data quality improvement system considering spatiotemporal characteristics, characterized in that: include: A data acquisition module is used to acquire the distribution network operation data of the time period to be detected and perform preprocessing; An abnormal data detection module is used to detect abnormal data on running data using an improved DBSCAN clustering algorithm; The time feature prediction module is used to obtain and pre-process the distribution network operation data in the historical time period, extract multiple intrinsic mode functions IMFs using the VMD algorithm, and convert the multiple IMFs into two-dimensional feature graphs; input the obtained two-dimensional feature graphs into the trained time prediction model to obtain the time feature prediction value of the time period to be detected; The spatial feature prediction module is used to obtain the spatial positions of all devices in the data source measurement device area, and predict the spatial feature prediction value of the time period to be detected based on the data values ​​of the adjacent measurement devices in combination with the line length and the number of nodes between the measurement devices; A spatiotemporal prediction module, used to calculate the spatiotemporal prediction value of the time period to be detected based on the time feature prediction value and the spatial feature prediction value; The abnormal data replacement module is used to replace the abnormal data points in the time period to be detected with the power forecast data at the corresponding moment.

9. A terminal device, comprising a processor and a memory, wherein the processor is used to implement instructions; and the memory is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the method for improving the quality of distribution network operation data taking into account spatiotemporal characteristics as described in any one of claims 1-7.

10. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the method for improving the quality of distribution network operation data taking into account spatiotemporal characteristics as described in any one of claims 1-7.

Images