Urban power grid voltage correction method based on data fusion and topology inspection

Abstract

The invention discloses an urban power grid voltage correction method based on data fusion and topology inspection, and belongs to the technical field of intelligent management and control of a power system. According to the method, data cleaning is carried out, a basic data set is constructed, then AMI measurement data is fused, and finally, a high-frequency data set is formed and data uploading is carried out. According to the urban power grid voltage correction method based on data fusion and topology inspection provided by the invention, the problem of asynchronous sampling of actual urban power grid measurement data is solved by utilizing fusion of multi-source measurement data; the problem that traditional urban power grid state estimation seriously depends on measurement of an intelligent electric meter is solved; the data quality of voltage measurement is improved, and a prerequisite is provided for realizing accurate state evaluation of an urban power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of electric power systems, and in particular to a method for correcting urban power grid voltage based on data fusion and topology verification. Background Art

[0002] Voltage data is the core of urban power grid state estimation. By measuring and analyzing the voltage, phase angle and other parameters of each node in the urban power grid, the operating status of the entire distribution system can be inferred. However, the structural parameters of the urban power grid are often difficult to accurately identify. Various measurements in urban power grids, such as PMU measurement, SCADA measurement, AMI measurement, etc., all have interference, missing or bad data, insufficient measurement information, difficulty in covering the global network, and different measurement time scales, which seriously affect the accuracy of urban power grid state estimation.

[0003] With the improvement of the information level of urban power grids and the development of machine learning technology, urban power grid status assessment has put forward higher requirements for measurement data. There are multi-source measurement data with different time scales and measurement accuracy in urban power grids. The asynchrony of data brings difficulties to improving the accuracy of urban power grid status estimation; at the same time, traditional single measurement has its own limitations. For example, AMI measurement is difficult to adapt to the needs of new power systems due to its large time scale. Fully integrating multi-source measurement data to obtain reliable and complete voltage data is the prerequisite for achieving fast and accurate urban power grid status estimation; therefore, it is necessary to design an urban power grid voltage correction method based on data fusion and topology verification to solve the above problems. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for urban power grid voltage correction based on data fusion and topology verification. The method aims to solve the problems of large differences in the time scale, measurement accuracy and synchronization of actual urban power grid measurement sampling, interference or even bad data in the voltage measurement data of the input model, and poor tolerance of ordinary parameterization methods due to the lack of physical information embedding.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for correcting urban power grid voltage based on data fusion and topology verification, comprising the following steps:

[0006] S1, the multi-source measurement and acquisition terminal collects the original data of each measurement node of the urban power grid distribution line, cleans the original data and sends it to the acquisition master station;

[0007] S2, select a certain sampling time of AMI measurement as the starting time, and set the sampling period T of AMI measurement l The basic dataset X bThe time scale is determined to determine each sampling time node; extract the data of three measurements corresponding to each sampling time node, establish basic data labels, and form the basic data set X b ;

[0008] S3, perform fusion processing on the AMI measurement data, using the basic data set X of the three measurement data in step S2 b , use the corresponding algorithm to fit the functional relationship and obtain the fused function;

[0009] S4, construct a high-frequency data set with the sampling period T in step S3 s As a new time scale, select a certain sampling time of AMI measurement as the starting time, determine each sampling time node, extract the data of three measurements corresponding to each sampling time node, and form a time scale of T s A high-frequency dataset X h ;

[0010] S5, data upload, based on the basic data set X in step S2 b , pre-build an urban power grid topology identification model based on voltage clustering sorting, check whether the AMI measurement fusion data obtained in step S3 is qualified, and upload the qualified data to the big data platform.

[0011] Preferably, in step S1, the original data of each measurement node includes line flow, bus feed-in power, bus voltage, type and quantity of measurement devices and measurement error data.

[0012] Preferably, the data cleaning of the basic data in step S1 includes outlier removal; outlier removal includes one or more of a Z-Score method, a box plot method, a sliding window method, and a trend decomposition algorithm.

[0013] Furthermore, considering that the measurement data collected by urban power grids are basically time series data, machine learning algorithms are also used to eliminate outliers, such as isolation forest and clustering-based methods. These methods can learn the normal patterns of data and thus identify data points that do not conform to the normal patterns. The learning method makes full use of the structural information of the measurement data, and the characteristics derived from the bus, such as voltage amplitude, voltage phase, injected power, load power, etc. are represented by node information; branch current, line impedance, etc. are represented by edge information to form a graph topology.

[0014] Preferably, the three types of measurement data in step S2 are AMI measurement, PMU measurement and SCADA measurement.

[0015] Preferably, in step S3, a basic data set X of AMI measurement, PMU measurement and SCADA measurement is obtained. bThen, the corresponding algorithm is used to fit the functional relationship between AMI measurement, PMU measurement and SCADA measurement. After fitting, AMI measurement is expressed as a function obtained by integrating PMU measurement and SCADA measurement. At the same time, a shorter sampling period T is selected. s to obtain the data.

[0016] Furthermore, because the data characteristics of multi-source measurement devices are different, the data collected by the main acquisition station are highly asynchronous in terms of time scale, frequency alignment and accuracy, the SCADA data is not accurate enough, the AMI data update cycle is long, and the overall PMU coverage is low. When the three types of data coexist, fully integrating multi-source measurement data to quickly and accurately estimate the state of the urban power grid has great innovation and research significance.

[0017] Preferably, in step S5, the method for constructing the urban power grid topology identification model is as follows:

[0018] S51, the urban power grid feeder terminal collects the instantaneous voltage amplitude at the feeder switch in real time, and forms an input vector of the clustering algorithm through normalization preprocessing;

[0019] S52, clustering the voltage time series collected by the nodes using a mean shift algorithm, clustering nodes with high node voltage similarity into one category, to form a branch category;

[0020] S53, sorting the nodes in the branch class according to the average voltage to form an ordered branch class, and determining the upstream and downstream relationship;

[0021] S54. Select the node with the largest average voltage in the ordered branch class as the branch head node, compare the average voltage between the branch head node and the nodes in the trunk, determine the node in the trunk with an average voltage greater than this point, connect the node with the smallest average voltage among the nodes to the branch head node, and identify the urban power grid topology.

[0022] Furthermore, the transformation of node measurement variables always has topological properties, and its characteristic propagation always has the physical topological information of the urban power grid.

[0023] Preferably, in step S5, the high frequency data set X obtained in step S4 is h Input the city power grid topology identification model, observe the model output topology identification results, judge whether the topology branches can be accurately clustered, and whether the upstream and downstream structures of the branches are accurately judged; if the topology structure obtained by the model is consistent with the actual topology, it means that the measurement fusion data is qualified, otherwise it means that the measurement fusion data is unqualified, and it is necessary to adjust the fusion strategy parameters and repeat the data fusion process.

[0024] The present invention has the following beneficial effects:

[0025] The present invention can fully integrate multi-source measurement data through the integration of multi-source measurement data to obtain reliable and complete voltage data, solve the problem of asynchronous sampling of measurement data in actual urban power grids, and further solve the problem that traditional urban power grid state estimation is heavily dependent on the measurement of smart meters; it improves the data quality of voltage measurement and provides a prerequisite for accurate state assessment of urban power grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0027] Figure 1 It is a schematic diagram of the model flow of the present invention. DETAILED DESCRIPTION

[0028] Embodiment 1:

[0029] like Figure 1 As shown, a method for urban power grid voltage correction based on data fusion and topology verification includes the following steps:

[0030] S1, the multi-source measurement and acquisition terminal collects the original data of each measurement node of the urban power grid distribution line, cleans the original data and sends it to the acquisition master station;

[0031] S2, select a certain sampling time of AMI measurement as the starting time, and set the sampling period T of AMI measurement l The basic dataset X b The time scale is determined to determine each sampling time node; extract the data of three measurements corresponding to each sampling time node, establish basic data labels, and form the basic data set X b ;

[0032] S3, perform fusion processing on the AMI measurement data, using the basic data set X of the three measurement data in step S2 b , use the corresponding algorithm to fit the functional relationship and obtain the fused function;

[0033] S4, construct a high-frequency data set with the sampling period T in step S3 s As a new time scale, select a certain sampling time of AMI measurement as the starting time, determine each sampling time node, extract the data of three measurements corresponding to each sampling time node, and form a time scale of T s A high-frequency dataset X h ;

[0034] S5, data upload, based on the basic data set X in step S2 b , pre-build an urban power grid topology identification model based on voltage clustering sorting, check whether the AMI measurement fusion data obtained in step S3 is qualified, and upload the qualified data to the big data platform.

[0035] Preferably, in step S1, the original data of each measurement node includes line flow, bus feed-in power, bus voltage, type and quantity of measurement devices and measurement error data.

[0036] Preferably, the data cleaning of the basic data in step S1 includes outlier removal; outlier removal includes one or more of a Z-Score method, a box plot method, a sliding window method, and a trend decomposition algorithm.

[0037] Furthermore, considering that the measurement data collected by urban power grids are basically time series data, machine learning algorithms are also used to eliminate outliers, such as isolation forest and clustering-based methods. These methods can learn the normal patterns of data and thus identify data points that do not conform to the normal patterns. The learning method makes full use of the structural information of the measurement data, and the characteristics derived from the bus, such as voltage amplitude, voltage phase, injected power, load power, etc. are represented by node information; branch current, line impedance, etc. are represented by edge information to form a graph topology.

[0038] Preferably, the three types of measurement data in step S2 are AMI measurement, PMU measurement and SCADA measurement.

[0039] Preferably, in step S3, a basic data set X of AMI measurement, PMU measurement and SCADA measurement is obtained. b Then, the corresponding algorithm is used to fit the functional relationship between AMI measurement, PMU measurement and SCADA measurement. After fitting, AMI measurement is expressed as a function obtained by integrating PMU measurement and SCADA measurement. At the same time, a shorter sampling period T is selected. s to obtain the data.

[0040] Furthermore, because the data characteristics of multi-source measurement devices are different, the data collected by the main acquisition station are highly asynchronous in terms of time scale, frequency alignment and accuracy, the SCADA data is not accurate enough, the AMI data update cycle is long, and the overall PMU coverage is low. When the three types of data coexist, fully integrating multi-source measurement data to quickly and accurately estimate the state of the urban power grid has great innovation and research significance.

[0041] Preferably, in step S5, the method for constructing the urban power grid topology identification model is as follows:

[0042] S51, the urban power grid feeder terminal collects the instantaneous voltage amplitude at the feeder switch in real time, and forms an input vector of the clustering algorithm through normalization preprocessing;

[0043] S52, clustering the voltage time series collected by the nodes using a mean shift algorithm, clustering nodes with high node voltage similarity into one category, to form a branch category;

[0044] S53, sorting the nodes in the branch class according to the average voltage to form an ordered branch class, and determining the upstream and downstream relationship;

[0045] S54. Select the node with the largest average voltage in the ordered branch class as the branch head node, compare the average voltage between the branch head node and the nodes in the trunk, determine the node in the trunk with an average voltage greater than this point, connect the node with the smallest average voltage among the nodes to the branch head node, and identify the urban power grid topology.

[0046] Furthermore, the transformation of node measurement variables always has topological properties, and its characteristic propagation always has the physical topological information of the urban power grid.

[0047] Preferably, in step S5, the high frequency data set X obtained in step S4 is h Input the city power grid topology identification model, observe the model output topology identification results, judge whether the topology branches can be accurately clustered, and whether the upstream and downstream structures of the branches are accurately judged; if the topology structure obtained by the model is consistent with the actual topology, it means that the measurement fusion data is qualified, otherwise it means that the measurement fusion data is unqualified, and it is necessary to adjust the fusion strategy parameters and repeat the data fusion process.

[0048] Embodiment 2:

[0049] The present invention provides a method for correcting urban power grid voltage based on data fusion and topology verification, comprising the following steps:

[0050] S1. Establish a multi-source measurement and acquisition terminal to collect basic data of each measurement node in the urban power grid, including the three-phase voltage module value Injection current magnitude Injected active power Injected reactive power

[0051] According to injection The current injection equation can be established as follows:

[0052]

[0053] Expanding the above formula, we can get the real part and imaginary part equations respectively:

[0054]

[0055]

[0056] In the formula is the power injected into endpoint i; is the active power injected into endpoint i; Reactive power for injection endpoint i; The real and imaginary parts of the voltage at point i on phase d; The mutual conductance and admittance of the endpoints.

[0057] According to the voltage modulus value Uid, the voltage modulus value equation can be established as follows:

[0058]

[0059] According to the injection current modulus The current modulus equation can be established as follows:

[0060]

[0061] Use the Z-Score method to screen for erroneous data, calculate the mean and standard deviation, set the error threshold, and eliminate abnormal data;

[0062] S2. Since the sampling periods of AMI measurement, SCADA and PMU measurement are quite different, they are normalized:

[0063] S21, since PMU samples approximately every 40ms, a large amount of PMU data has been collected and transmitted to the data center between two AMI data. Therefore, the optimal buffer length of PMU is calculated so that the equivalent section data and AMI data can be combined for application. 30s before the arrival of AMI, PMU sampling has approximately 750 measurement data. The method for calculating the optimal buffer length is to assume that the PMU sampling frequency is n r samples / s. The system updates the status every time the next AMI data arrives. 30s before the AMI arrives:

[0064] (1) The number of PMU sampling measurements is N = 30*n r , n r The sampling measurements are divided into n subset There are measurement subsets, usually n subset =30, so the number of each measurement subset is n meas =n r , n subset The subsets are represented as: Z(1), Z(2), ..., Z(n subset );

[0065] (2) For the latest PMU measurement subset Z(n subset ) to perform change point detection:

[0066] Calculate Z(n subset) and compare the calculated standard deviation with the threshold value (the amplitude threshold value is generally set to 10-4pu, and the phase angle threshold value is generally set to 10-3°). If the standard deviation is smaller than the threshold value, it means that there is no change point in Z(nsubset), and Z(n subset ) is completely contained in the buffer, and step (3) is performed;

[0067] If the standard deviation is greater than the threshold, Z(n subset ) has a change point, then take Z(n subset ) The latest PMU in the equivalent section is the measurement value, and the calculation is ended;

[0068] (3) Based on the internal function ttest2 in MATLAB, Z(n subset ) and Z(n subset -1), ..., Z(3) and Z(2), Z(2) and Z(1) are tested with a default probability of 95%. The results are expressed as h i (i=1,2,…n subset -1) is expressed as:

[0069] Where: N op_bf is the optimal buffer length of PMU; the overline is the "not" operator;

[0070] Therefore, based on the above step S2, the optimal buffer length can be calculated for each PMU measurement point, and the PMU statistical data and variance in the buffer zone can be calculated, thereby realizing cross-section normalization;

[0071] S22, similarly, for SCADA measurement, since it samples approximately every 2 seconds, more SCADA data has been collected and transmitted to the data center in the gap between two AMI data, so it is also necessary to calculate the optimal SCADA buffer length so that the equivalent section data and AMI data can be combined for application. Within 15 minutes, the SCADA sampling data series has approximately 450 measurement data, and the method for calculating the optimal buffer length is similar to that of PMU measurement;

[0072] S23, after the above process, the PMU and SCADA are normalized, and all processed data and the incoming AMI data are regarded as data in the same section, and a certain sampling time of AMI measurement is selected as the starting time, and the AMI sampling period T is set. l For data set X b The time scale is T, and each sampling time node is determined. The three voltage measurement data corresponding to each sampling time node are extracted, and the basic data labels are established to form a time scale of T. l Basic data set X b ;

[0073] S3. The main acquisition station performs data fusion processing on the AMI measurement. The AMI measurement sampling period is 15 minutes, and the sampling frequency is relatively low. In order to increase the data frequency, data fusion is performed. The PMU measurement data, SCADA measurement data, and AMI measurement data are aligned in time, and their respective data time series are extracted and counted as In this case, a long short-term memory neural network is used to analyze and process it, and the functional relationship between AMI measurement, PMU measurement, and SCADA measurement is fitted.

[0074] Using the data obtained in step S2, the PMU measurement data, SCADA measurement data, and AMI measurement data are divided into training sets and test sets. PMU measurements, SCADA measurements, and AMI measurements are input into the network as multiple input sequences, and the LSTM transfer process is as follows:

[0075]

[0076]

[0077]

[0078]

[0079]

[0080] h t =o t tanh(C t );

[0081]

[0082] Where: f t Represents the output signal of the forget gate, whose value determines the forgetting ratio of the memory unit c; i t Represents the output signal of the output gate, whose value determines how much of the current input information is input into the memory unit c; Represents the preliminary information to be input into the memory unit c, and its value is the same as i t Point product to get the information in memory unit c; o t represents the output signal of the output gate, whose value determines the proportion of the memory unit c output to the current state s; b f 、b i 、b c 、b o are the bias terms; W f , W i , W c , W o are the weight matrices; h tRepresents the memory cell C at time t t After being filtered by the forget gate and the input gate, the hidden state is obtained by filtering by the output gate; represents the predicted AMI measurement fusion data;

[0083] After fitting, AMI measurement is expressed as a function obtained by integrating PMU measurement and SCADA measurement. According to the actual situation, a short sampling period T is selected. s Data collection, that is, AMI measurement, through the integration of PMU measurement and SCADA measurement data, speeds up the sampling frequency and solves the problem of long AMI sampling cycle, making it meet the needs of new power systems;

[0084] S4, with the sampling period T in step S3 s As a new time scale, select a certain sampling time of AMI measurement as the starting time, determine each sampling time node, extract the voltage data of three measurements corresponding to each sampling time node, and form a time scale of T s A high-frequency dataset X h ;

[0085] S5, based on the basic data set X obtained in step S2 b , build a city power grid topology identification model based on voltage clustering sorting, set the topology identification model to have excellent judgment performance, and check whether the AMI measurement fusion data is qualified. In this model, input the high-frequency data set X obtained in step S4 h Observe the topological identification results output by the model to determine whether the topological branches can be accurately clustered and whether the branch structure is accurate. Assume that the topological identification success rate is The identified error information includes branch clustering error events α and branch structure errors β, σ is its weight, the number of errors is n, and the topology identification success rate is The formula is:

[0086]

[0087] If the topology obtained by the model is consistent with the actual topology, If it is greater than 98%, it means that the measurement fusion data is qualified. Otherwise, it means that the measurement fusion data is unqualified. It is necessary to adjust the fusion strategy parameters and repeat the data fusion process. h , that is, accurate and reliable voltage measurement data can be obtained.

[0088] Therefore, the present invention adopts the above-mentioned urban power grid voltage correction method based on data fusion and topology verification, and utilizes the fusion of multi-source measurement data to solve the problem of asynchronous sampling of actual urban power grid measurement data; solves the problem that traditional urban power grid state estimation is heavily dependent on the measurement of smart meters; improves the data quality of voltage measurement, and provides a prerequisite for accurate state assessment of urban power grids.

[0089] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for urban power grid voltage correction based on data fusion and topology verification, characterized in that: The following steps are involved: S1, the multi-source measurement and acquisition terminal collects the original data of each measurement node of the urban power grid distribution line, cleans the original data and sends it to the acquisition master station; S2, select a certain sampling time of AMI measurement as the starting time, and set the sampling period T of AMI measurement l The basic dataset X b The time scale is determined to determine each sampling time node; extract the data of three measurements corresponding to each sampling time node, establish basic data labels, and form the basic data set X b ; S3, perform fusion processing on the AMI measurement data, using the basic data set X of the three measurement data in step S2 b , use the corresponding algorithm to fit the functional relationship and obtain the fused function; S4, construct a high-frequency data set with the sampling period T in step S3 s As a new time scale, select a certain sampling time of AMI measurement as the starting time, determine each sampling time node, extract the data of three measurements corresponding to each sampling time node, and form a time scale of T s A high-frequency dataset X h ; S5, data upload, based on the basic data set X in step S2 b , pre-build an urban power grid topology identification model based on voltage clustering sorting, check whether the AMI measurement fusion data obtained in step S3 is qualified, and upload the qualified data to the big data platform.

2. The urban power grid voltage correction method based on data fusion and topology verification according to claim 1 is characterized in that: In step S1 , the original data of each measurement node includes line flow, bus feed-in power, bus voltage, type and quantity of measurement devices and measurement error data.

3. The urban power grid voltage correction method based on data fusion and topology verification according to claim 1 is characterized in that: In step S1, data cleaning of basic data includes outlier removal; outlier removal includes one or more of a Z-Score method, a box plot method, a sliding window method, and a trend decomposition algorithm.

4. The urban power grid voltage correction method based on data fusion and topology verification according to claim 1 is characterized in that: The three types of measurement data in step S2 are AMI measurement, PMU measurement and SCADA measurement.

5. The urban power grid voltage correction method based on data fusion and topology verification according to claim 1 is characterized in that: In step S3, the basic data set X of AMI measurement, PMU measurement and SCADA measurement is obtained. b Then, the corresponding algorithm is used to fit the functional relationship between AMI measurement, PMU measurement and SCADA measurement; After fitting, AMI measurement is expressed as a function obtained by integrating PMU measurement and SCADA measurement, and a shorter sampling period T is selected. s to obtain the data.

6. The urban power grid voltage correction method based on data fusion and topology verification according to claim 1 is characterized in that: In step S5, the construction method of the urban power grid topology identification model is as follows: S51, the urban power grid feeder terminal collects the instantaneous voltage amplitude at the feeder switch in real time, and forms an input vector of the clustering algorithm through normalization preprocessing; S52, clustering the voltage time series collected by the nodes using a mean shift algorithm, clustering nodes with high node voltage similarity into one category, to form a branch category; S53, sorting the nodes in the branch class according to the average voltage to form an ordered branch class, and determining the upstream and downstream relationship; S54. Select the node with the largest average voltage in the ordered branch class as the branch head node, compare the average voltage between the branch head node and the nodes in the trunk, determine the node in the trunk with an average voltage greater than this point, connect the node with the smallest average voltage among the nodes to the branch head node, and identify the urban power grid topology.

7. The urban power grid voltage correction method based on data fusion and topology verification according to claim 6 is characterized in that: In step S5, the high-frequency data set X obtained in step S4 is h Input the city power grid topology identification model, observe the model output topology identification results, judge whether the topology branches can be accurately clustered, and whether the upstream and downstream structures of the branches are accurately judged; if the topology structure obtained by the model is consistent with the actual topology, it means that the measurement fusion data is qualified, otherwise it means that the measurement fusion data is unqualified, and it is necessary to adjust the fusion strategy parameters and repeat the data fusion process.

Images