Method and device for extracting fast ramping and switching scenario points of power system

Abstract

The application provides a kind of fast climbing and switching scene point extraction method and device of power system, belong to data processing field.The method comprises: obtaining the load data of target power system within preset time;Adopt adaptive periodic local feature anomaly detection algorithm APLAD, determine the daily periodicity characteristics and local difference of load data, and based on daily periodicity characteristics and local difference, remove the abnormal data in load data, obtain preprocessed load data;The preprocessed load data is reconstructed, and a plurality of groups of reconstruction data are obtained;And according to a plurality of groups of reconstruction data, a plurality of groups of fusion feature data are obtained;Based on a plurality of groups of fusion feature data, obtain the fast climbing and switching scene point of target power system.The application can overcome the problem of low accuracy of fast climbing and switching scene point extraction caused by poor data quality in traditional method.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a method and device for extracting fast ramping and switching scene points of a power system. BACKGROUND

[0002] As an important part of the new energy system, the energy internet integrates a large number of distributed renewable energy power generation, has the advantages of instant use, environmental protection and high efficiency, and is considered to be the development trend of future energy systems. However, with the high proportion of renewable energy access, the energy internet faces many challenges, especially the randomness, volatility and intermittency of renewable energy, which brings great pressure to the safe and stable operation of the energy system. In this context, effective measures need to be taken to face these challenges.

[0003] To solve this problem, the power system needs to have the ability to quickly ramp up, which requires accurate determination of fast ramping and switching scene points. However, due to the existence of low-quality and erroneous data, it often affects the accuracy of fast ramping and switching scene point extraction, so it is necessary to fine clean individual data to improve data quality. However, the inventors have found that the current method usually extracts wind power data by hour, ignoring long-term meteorological changes and system operation trends, and does not adequately consider the data processing process, resulting in poor data quality and affecting the accuracy of fast ramping and switching scene points. SUMMARY

[0004] The embodiments of the present application provide a method and device for extracting fast ramping and switching scene points of a power system to solve the problem of low accuracy of fast ramping and switching scene point extraction in the current power system.

[0005] In a first aspect, the embodiments of the present application provide a method for extracting fast ramping and switching scene points of a power system, comprising:

[0006] Obtaining load data of a target power system within a preset time;

[0007] Using an adaptive periodic local feature anomaly detection algorithm APLAD to determine the daily periodicity feature and local difference of the load data, and based on the daily periodicity feature and the local difference, removing the abnormal data in the load data to obtain preprocessed load data;

[0008] Reconstructing the preprocessed load data to obtain a plurality of groups of reconstructed data; and obtaining a plurality of groups of fusion feature data according to the plurality of groups of reconstructed data;

[0009] Based on the plurality of groups of fusion feature data, obtaining fast ramping and switching scene points of the target power system.

[0010] In a possible implementation, the removing of the abnormal data in the load data based on the daily periodicity feature and the local difference comprises:

[0011] For any one data in the load data, a comprehensive score of the load data is obtained according to the daily periodicity feature and the local difference corresponding to the load data.

[0012] When the comprehensive score of the load data is greater than a preset threshold of an abnormal point, the load data is removed as the abnormal data.

[0013] In a possible implementation, the calculation formula of the comprehensive score is:

[0014]

[0015] Wherein, ξ and δ are adjustment coefficients respectively; α is the local difference; β is the daily periodicity feature.

[0016] In a possible implementation, the obtaining of the multiple sets of fusion feature data according to the multiple sets of reconstructed data comprises:

[0017] For any one set of reconstructed data, a time domain feature, a statistical feature and a periodicity feature of the set of reconstructed data are calculated.

[0018] According to the time domain feature, the statistical feature and the periodicity feature of the set of reconstructed data, a short-time local feature and a long-time trend feature corresponding to the set of reconstructed data are extracted.

[0019] The short-time local feature and the long-time trend feature corresponding to the set of reconstructed data are weighted and fused to obtain fusion feature data corresponding to the set of reconstructed data.

[0020] In a possible implementation, the obtaining of the fast ramping and switching scene points of the target power system based on the multiple sets of fusion feature data comprises:

[0021] An improved density-based clustering algorithm DBSCAN is used to cluster the multiple sets of fusion features to obtain the fast ramping and switching scene points of the target power system; wherein the improved DBSCAN introduces a giant lotus optimization algorithm in the traditional DBSCAN, which is used to adaptively adjust the numerical difference and the change trend of the data in the calculation process.

[0022] In a possible implementation, the improved density-based clustering algorithm DBSCAN is used to cluster the multiple sets of fusion features to obtain the fast ramping and switching scene points of the target power system, comprising:

[0023] The improved density-based clustering algorithm DBSCAN is used to cluster the multiple sets of fusion features to obtain multiple clustering points.

[0024] For any one cluster point, determine all data points within the preset distance threshold range of the cluster point; if the number of data points is greater than the preset number, establish the cluster point and all data points within the preset distance threshold range of the cluster point as a new cluster label; if the number of data points is not greater than the preset number, establish the cluster point and all data points within the preset distance threshold range of the cluster point as a fast climbing and switching scene point label;

[0025] Traverse all cluster points and access all data points within the preset distance threshold range corresponding to the cluster points; if there is a data point without a label, determine whether the data point is a cluster point, if not, assign the data point a label corresponding to the cluster point; if yes, explore the density reachable object until it can no longer be explored, obtain a complete cluster, and return to the step of determining all data points within the preset distance threshold range of any one cluster point, until there is no data point without a label, and output the clustering result;

[0026] Based on the clustering result, the fast climbing and switching scene point of the target power system is obtained.

[0027] In a possible implementation, after obtaining the fast climbing and switching scene point of the target power system based on the fused feature data, the method further includes:

[0028] The fast climbing and switching scene point is reversely mapped into the corresponding time sequence or observation interval in the load data;

[0029] According to the index operation, the position of each fast climbing and switching scene point in the load data is located, and the multi-dimensional features and time sequence information of each fast climbing and switching scene point are restored.

[0030] In a possible implementation, after removing the abnormal data in the load data based on the daily periodicity feature and the local difference, the method further includes:

[0031] The data filling model is trained by using the load data after removing the abnormal data, so as to fill in the missing values in the load data, and obtain the preprocessed load data.

[0032] In a possible implementation, the calculation formula of the daily periodicity feature is:

[0033]

[0034] Wherein, n is the number of load data at the same time each day; x i (h) is the load data corresponding to the time point h; x t is the standardized value of the number load data at the time point t;

[0035] The calculation formula of the local difference is:

[0036]

[0037] wherein k is a local variation window length; x c (i) is the load data at time point i.

[0038] In a second aspect, an embodiment of the present application provides a device for extracting fast ramping and switching scene points of a power system, comprising:

[0039] A collection module is configured to acquire load data of a target power system within a preset time period.

[0040] A processing module is configured to determine daily periodic characteristics and local differences of the load data by using an adaptive periodic-local anomaly detection algorithm (APLAD), remove abnormal data in the load data based on the daily periodic characteristics and the local differences, and obtain preprocessed load data.

[0041] An extraction module is configured to reconstruct the preprocessed load data to obtain a plurality of groups of reconstructed data, and obtain a plurality of groups of fusion feature data based on the plurality of groups of reconstructed data.

[0042] A clustering module is configured to obtain fast ramping and switching scene points of the target power system based on the plurality of groups of fusion feature data.

[0043] The embodiment of the present application provides a method and a device for extracting fast ramping and switching scene points of a power system. The daily periodic characteristics and the local differences of load data are analyzed by using an adaptive periodic-local anomaly detection algorithm (APLAD), so that short-term volatility and periodic deviation in the load data can be analyzed, the fluctuation law of the load data can be accurately captured, abnormal data in the load data can be removed, and the accuracy of extracting fast ramping and switching scene points is preliminarily ensured. Considering that the collected load data at the present stage is mostly wind power data extracted by hours, the preprocessed load data is reconstructed, fusion feature data reflecting long-term weather changes and system operation trends is extracted from the reconstructed data, the fusion feature data is used to determine corresponding fast ramping and switching scene points, and the problem of low accuracy of extracting fast ramping and switching scene points caused by poor data quality in the traditional method can be overcome. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0045] Figure 1 is an implementation flowchart of the power system rapid ramping and switching scene point extraction method provided by the embodiment of the present application;

[0046] Figure 2 is a k-distance graph for searching for an eps parameter;

[0047] Figure 3 is a clustering result schematic diagram of the power system rapid ramping and switching scene point extraction method provided by the embodiment of the present application;

[0048] Figure 4 is an implementation flowchart of the power system rapid ramping and switching scene point extraction method provided by another embodiment of the present application;

[0049] Figure 5 is a structural schematic diagram of the power system rapid ramping and switching scene point extraction device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0050] In the following description, specific details are set forth in order to provide a thorough understanding of the embodiments of the present application. However, persons skilled in the art should understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary details.

[0051] In order to make the objects, technical solutions and advantages of the present application clearer, the following will be described by specific embodiments in conjunction with the drawings.

[0052] Figure 1 is an implementation flowchart of the power system rapid ramping and switching scene point extraction method provided by the embodiment of the present application. As shown in Figure 1 , the method comprises:

[0053] Step 110: Obtain the load data of the target power system within a preset time.

[0054] In this embodiment, the load data within the preset time period can be wind power load data under a medium-long time scale. The load data within the preset time can be time series data.

[0055] Step 120: Adopting the adaptive periodic local feature anomaly detection algorithm APLAD to determine the daily periodicity feature and local difference of the load data, and remove the abnormal data in the load data based on the daily periodicity feature and the local difference, to obtain the preprocessed load data.

[0056] In this embodiment, the APLAD can extract the daily periodicity feature of the load data to capture the periodic fluctuations in the load data. In order to more accurately capture the local difference, a sliding filter with a preset sliding window can be used to smooth the load data on the time series to calculate the difference between each load data and its neighborhood, and then obtain the local difference of each load data. Then, based on the analyzed daily periodicity feature and local difference, the regular fluctuations of the load data are analyzed, and the difference degree of the load data collected by each collection point is quantified.

[0057] Correspondingly, in an optional embodiment, for any one load data, its daily periodicity feature can be calculated by the following formula:

[0058]

[0059] Wherein, n is the number of load data at the same time each day; x i (h) is the load data corresponding to the time point h; x t is the standardized value of the load data at time point t;

[0060] For any one load data, its local difference can be calculated by the following formula:

[0061]

[0062] Wherein, k is the length of the local change window; x c (i) is the load data at time point i.

[0063] After determining the daily periodicity feature and the local difference corresponding to each load data, the abnormal data can be removed by the following steps:

[0064] For any one data in the load data, according to the daily periodicity feature and the local difference corresponding to the load data, the comprehensive score of the load data is obtained.

[0065] When the comprehensive score of the load data is greater than the threshold value of the preset abnormal point, the load data is removed as abnormal data.

[0066] In this embodiment, the comprehensive score of any one load data can be calculated by the following formula:

[0067]

[0068] wherein ξ and δ are adjustment coefficients, respectively, which can be dynamically adjusted according to the local difference and the diurnal periodicity feature; α is the local difference; and β is the diurnal periodicity feature.

[0069] The threshold of the preset abnormal point can be determined by the following formula:

[0070] θ = P (p, 100 x (1-p))

[0071] wherein p is the proportion of the abnormal point.

[0072] When the comprehensive score of any one load data is greater than the threshold of the preset abnormal point, it can be considered that the load data is abnormal data, and the load data is removed.

[0073] In this embodiment, after the abnormal data in the load data is removed according to the method provided in step 120, the load data after removing the abnormal data can be obtained. Considering that in addition to the abnormal data, there can also be missing data in the obtained data, after removing the abnormal data, the following can also be included:

[0074] The load data after removing the abnormal data is used to train a data filling model to fill in the missing values in the load data, to obtain the preprocessed load data.

[0075] In this embodiment, the data filling model can be a Light Gradient Boosting Machine (LightGBM). Through the LightGBM model, the missing values in the load data can be predicted, and the predicted data can be used to fill in the missing load data, to improve the quality of the load data.

[0076] Step 130: reconstructing the preprocessed load data to obtain a plurality of groups of reconstructed data; and obtaining a plurality of groups of fusion feature data according to the plurality of groups of reconstructed data.

[0077] In this embodiment, considering that the current method usually extracts wind power data by hour, ignoring long-term meteorological changes and system operation trends when extracting data, the preprocessed load data is reconstructed. Specifically, the preprocessed load data in the time sequence can be divided into a plurality of groups of reconstructed data by week.

[0078] Then, since the APLAD is used to preprocess the load data, it will bring short-term noise impact to a single load data. To eliminate this impact, the plurality of groups of data are further processed to obtain a plurality of groups of fusion feature data.

[0079] Correspondingly, in an optional embodiment, the step of obtaining a plurality of sets of fusion feature data according to the plurality of sets of reconstructed data in step 130 can include:

[0080] For any one set of reconstructed data, the time domain feature, the statistical feature, and the periodic feature of the set of reconstructed data are calculated.

[0081] According to the time domain feature, the statistical feature, and the periodic feature of the set of reconstructed data, the short-time local feature and the long-time trend feature corresponding to the set of reconstructed data are extracted.

[0082] The short-time local feature and the long-time trend feature corresponding to the set of reconstructed data are weighted and fused to obtain the fusion feature data corresponding to the set of reconstructed data.

[0083] In this embodiment, for each set of reconstructed data, the maximum value, the minimum value, the change amount, the fluctuation range, and the peak value can be calculated as the statistical feature; the amplitude, the frequency, the period, and the waveform can be calculated as the statistical feature; and the main frequency cost and the corresponding amplitude can be extracted through Fourier transform as the periodic feature. Based on these features, the structure of the load data can be simplified, the short-term fluctuations and noises are smoothed, important time sequence and frequency domain information are retained, the short-term noise influence on the load data caused by the APLAD algorithm is reduced, the long-term trend and the fast ramping and switching scene points are highlighted, and the characteristics of the weekly load data are comprehensively reflected.

[0084] In this embodiment, in order to obtain the short-time local feature and the long-time trend feature of each set of reconstructed data, the time domain feature, the statistical feature, and the periodic feature corresponding to each set of reconstructed data can be input into two convolution modules with different size convolution kernels for convolution to extract the short-time local feature and the long-time trend feature of each set of reconstructed data, and then different weights are given to the short-time local feature and the long-time trend feature of each set of reconstructed data for weighted fusion to obtain a plurality of sets of fusion features. The obtained fusion features can reveal the potential patterns and structures in the load data.

[0085] Step 140: obtaining the fast ramping and switching scene points of the target power system based on the plurality of sets of fusion feature data.

[0086] In this embodiment, the clustering method can be used to cluster the plurality of sets of fusion feature data to obtain the fast ramping and switching scene points of the target power system.

[0087] Correspondingly, in an optional embodiment, the improved density-based clustering algorithm (Density-Based Spatial Clustering of Applications with Noise, DBSCAN) can be selected for clustering.

[0088] In the embodiment, the improved DBSCAN algorithm is introduced into the traditional DBSCAN, and the numerical difference and change trend of data in the calculation process can be dynamically adjusted through the giant water lily optimization algorithm.

[0089] Correspondingly, in a specific embodiment, obtaining the fast ramping and switching scene points of the target power system based on the multiple sets of fusion feature data in step 130 can include:

[0090] The multiple sets of fusion features are clustered by using the improved density-based clustering algorithm DBSCAN to obtain multiple clustering points.

[0091] For any one clustering point, all data points within a preset distance threshold range of the clustering point are determined; if the number of data points is greater than a preset number, the clustering point and all data points within the preset distance threshold range are established as a new cluster label; if the number of data points is not greater than the preset number, the clustering point and all data points within the preset distance threshold range are established as a fast ramping and switching scene point label.

[0092] All clustering points are traversed, and all data points within the preset distance threshold range corresponding to the clustering points are accessed; if there is a data point without a label, it is determined whether the data point is a clustering point, if not, the data point is assigned a label corresponding to the clustering point; if yes, density-reachable objects are explored until no further exploration is possible, a complete cluster is obtained, and the step of determining all data points within the preset distance threshold range of any one clustering point is returned until there is no data point without a label, and the clustering result is output.

[0093] Based on the clustering result, the fast ramping and switching scene points of the target power system are obtained.

[0094] In the embodiment, the improved DBSCAN algorithm can be used to select a center point to divide the multiple sets of fusion feature data, obtain an active set and a candidate set, and only cluster the active set, thereby avoiding global neighborhood search on all fusion feature data and reducing the computational complexity. The preset distance threshold is eps, and the preset number is the MinPts parameter.

[0095] The active set can be represented as:

[0096] κ={q|q∈D,|q-p|≤n·eps}

[0097] Wherein, D represents any one set of fusion feature data, P represents the center point, q represents any one data in the set of fusion feature data; n represents the distance between the data and the center point, and eps represents the maximum distance.

[0098] The candidate set is a data point far from the current operation area and temporarily not required to be calculated, which can be expressed as:

[0099] τ = D - κ.

[0100] The giant water lily algorithm can balance global search and local optimization while searching widely, avoid falling into local optimum, and accelerate the convergence process, so that the model can adapt to data characteristics flexibly, thereby improving clustering effect and overall stability. Therefore, by combining the position difference of data points in space and their direction similarity, dynamically adjusting the relative importance of the position difference and the direction similarity of the data points in the candidate set in the giant water lily optimization algorithm, optimizing the parameters in the algorithm, and making it adaptive to data with different density distributions, the weight is determined by calculating the numerical difference and the trend of each data point. The greater the numerical difference, the more the clustering algorithm tends to emphasize the position difference of the data; and the more similar the change trend, the more the clustering algorithm focuses on the direction consistency of the data. On this basis, combined with the global exploration and local development ability of the giant water lily optimization algorithm, the process of this dynamic adjustment is enhanced.

[0101] Specifically, in the clustering process, the corresponding maximum distance parameter eps and threshold parameter MinPts can be obtained based on the characteristics after each clustering.

[0102] Wherein, Figure 2 is the k-distance graph of the eps parameter; as shown in FIG. 1, the abscissa represents the sample data amount (i.e. the number of data points), and the ordinate represents the corresponding distance, which increases with the increase of the sample amount. Figure 2

[0103] Through clustering, a plurality of clustering points are obtained, for any one clustering point, all data points within the eps range of the clustering point are determined, and the number of all found data points is compared with the MinPts parameter. If the number of data points is greater than the MinPts parameter, the clustering point and all data points within the eps range are established as a new cluster label; if the number of data points is not greater than the MinPts parameter, the clustering point and all data points within the eps range are established as a fast climbing and switching scene point label.

[0104] ​Traverse all the cluster points and visit all the data points within the eps range of the cluster points; determine whether the data points are assigned corresponding labels, if not, determine whether the data points without assigned labels are cluster points. If not, assign the label of the cluster point corresponding to the data point to the data point. If the point is a cluster point, recursively explore the density reachable objects until no further exploration is possible, and obtain a complete cluster.

[0105] Repeat the above steps to continue selecting data points without being visited until there is no data point without assigned labels, output the clustering result, and take the cluster obtained by clustering as the fast ramping and switching scenario point.

[0106] In an optional embodiment, after obtaining the fast ramping and switching scenario point of the target power system based on the fused feature data, the method can further include:

[0107] Reverse mapping the fast ramping and switching scenario point to the corresponding time series or observation interval in the load data.

[0108] According to the index operation, locate the position of each fast ramping and switching scenario point in the load data, and restore the multi-dimensional features and time series information of each fast ramping and switching scenario point.

[0109] In the embodiment, according to the fast ramping and switching scenario points obtained by clustering, the data corresponding to the points are reverse mapped to the corresponding time series or observation interval in the original data, that is, the load data obtained in step 110. Through accurate index operation, the specific position of each fast ramping and switching scenario point in the original data is located, and the original multi-dimensional features and time series information are restored, ensuring that the tracing of the fast ramping and switching scenario is not limited to the summary data level, but is traced back to complete fine-grained data, facilitating further analysis and verification.

[0110] In order to verify the effectiveness of the method provided by the embodiment of the application, the following experiment is performed in the embodiment: 78888 load data from January 1, 2006 to December 31, 2014 in the GEFCom2014 data set are used for simulation test. The load data is subjected to the operations in steps 120 to 140, and the clustering result is as shown in Figure 3

[0111] ​After clustering, 6 cluster clusters appear, and 48 fast climbing and switching scenarios away from the cluster cluster, finally the scene is extracted and analyzed, the fast climbing and switching scenario presents the time and seasonal regularity, and the data analysis finds that it is mainly concentrated in January, February, June, July and August, and the fast climbing and switching scenario extracted is obviously different from the normal situation load, which shows that the model achieves good effect, and the fast climbing and switching scene extraction of the medium and long time scale can be realized, and can be used for practical engineering application.

[0112] Figure 4 The implementation flowchart of the fast climbing and switching scene point extraction method of the power system provided by another embodiment of the present application is as shown in the figure, Figure 4 The method can include:

[0113] Collect and input the historical load data of the power system in a preset time period.

[0114] The daily periodicity characteristics and short-term trend deviation of the load data are analyzed by APLAD, the regularity fluctuation is dynamically modeled, and the abnormal degree of each time point is quantified. Through sensitivity control, the short-term fluctuation and periodic deviation in the load data are analyzed, the feature changes under different time scales are combined, the abnormal points are accurately captured and positioned, the adaptability and flexibility of the abnormal detection are ensured, and finally the abnormal data points are removed.

[0115] The missing values are predicted and filled using the LightGBM model, and the data cleaning is completed.

[0116] The preprocessed data is summarized by week, the data time domain features and statistical features of each week are extracted to describe the overall trend of load change, and the frequency domain analysis is performed on the data of each week to reveal the periodicity and implicit frequency characteristics of the data. These features are input into two convolution modules with different convolution kernel sizes for weighted fusion, local features and long-term trend features are extracted respectively, further clustering analysis is performed, and k-distance diagram is drawn to find the eps parameter of the DBSCAN model algorithm.

[0117] The improved DBSCAN is used to divide the data set into active set and candidate set, the calculation complexity is reduced, and the position difference of the data points in the space and their direction similarity are combined. The relative importance of the two data points in the candidate set is dynamically adjusted by the giant lotus optimization algorithm, and the parameters in the optimization algorithm are optimized, so that it is self-adaptive to data with different density distribution.

[0118] When the label of the clustered data point is-1, that is, all the corresponding labels of the data points are established, according to the label of the data point, the set of fast climbing and switching scene points is restored.

[0119] The contents not described in detail in the embodiment can be referred to the other related embodiments described above.

[0120] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the application.

[0121] The following is the device embodiment of the application. For details not described in detail, please refer to the corresponding method embodiments described above.

[0122] Figure 5 The structural schematic diagram of the device for extracting fast ramping and switching scene points of the power system provided by the embodiment of the application is shown. For ease of illustration, only the part related to the embodiment of the application is shown, and the details are as follows:

[0123] As shown in Figure 5 The device for extracting fast ramping and switching scene points of the power system 5 comprises:

[0124] The acquisition module 51 is configured to obtain load data of the target power system within a preset time;

[0125] The processing module 52 is configured to determine the daily periodicity feature and the local difference of the load data by using an adaptive periodic local feature anomaly detection algorithm (APLAD), and remove the abnormal data in the load data based on the daily periodicity feature and the local difference, to obtain preprocessed load data;

[0126] The extraction module 53 is configured to reconstruct the preprocessed load data to obtain a plurality of groups of reconstructed data, and obtain a plurality of groups of fusion feature data according to the plurality of groups of reconstructed data;

[0127] The clustering module 54 is configured to obtain the fast ramping and switching scene points of the target power system based on the plurality of groups of fusion feature data.

[0128] In a possible implementation manner, the processing module 52 is specifically configured to:

[0129] For any one data in the load data, the comprehensive score of the load data is obtained according to the daily periodicity feature and the local difference corresponding to the load data;

[0130] When the comprehensive score of the load data is greater than the threshold value of the preset abnormal point, the load data is removed as abnormal data.

[0131] In a possible implementation manner, the calculation formula of the comprehensive score is as follows:

[0132]

[0133] Wherein, ξ and δ are adjustment coefficients respectively; α is the local difference; and β is the daily periodicity feature.

[0134] In a possible implementation, the extraction module 53 is specifically configured to:

[0135] For any one group of reconstructed data, calculate the time domain feature, the statistical feature and the periodicity feature of the group of reconstructed data;

[0136] According to the time domain feature, the statistical feature and the periodicity feature of the group of reconstructed data, extract the short-time local feature and the long-time trend feature corresponding to the group of reconstructed data;

[0137] Perform weighted fusion on the short-time local feature and the long-time trend feature corresponding to the group of reconstructed data, to obtain the fusion feature data corresponding to the group of reconstructed data.

[0138] In a possible implementation, the clustering module 54 is specifically configured to:

[0139] Perform clustering on the multiple groups of fusion features by using an improved density-based clustering algorithm DBSCAN, to obtain the fast ramping and switching scenario points of the target power system; wherein the improved DBSCAN introduces a giant lotus optimization algorithm in the traditional DBSCAN, to adaptively adjust the numerical difference and the change trend of data in the calculation process.

[0140] In a possible implementation, the clustering module 54 is specifically configured to:

[0141] Perform clustering on the multiple groups of fusion features by using an improved density-based clustering algorithm DBSCAN, to obtain multiple clustering points;

[0142] For any one clustering point, determine all data points within a preset distance threshold range of the clustering point; if the number of data points is greater than a preset number, establish the clustering point and all data points within the preset distance threshold range of the clustering point as a new cluster label; if the number of data points is not greater than the preset number, establish the clustering point and all data points within the preset distance threshold range of the clustering point as a fast ramping and switching scenario point label;

[0143] Iterate through all the clustering points and access all the data points within the preset distance threshold range of the clustering points; if there is a data point without a label, determine whether the data point is a clustering point, if not, assign the data point with a label corresponding to the clustering point; if yes, explore the density-reachable objects until no further exploration is possible, to obtain a complete cluster, and return to the step of determining, for any one clustering point, all data points within a preset distance threshold range of the clustering point, until there is no data point without a label, and output the clustering result;

[0144] Based on the clustering result, obtain the fast ramping and switching scenario points of the target power system.

[0145] In a possible implementation, the clustering module 54 is further configured to:

[0146] map the fast ramping and switching scenario points reversely to corresponding time series or observation intervals in the load data;

[0147] According to the index operation, the position of each fast ramping and switching scenario point in the load data is located, and the multi-dimensional features and time series information of each fast ramping and switching scenario point are recovered.

[0148] In a possible implementation, the processing module 52 is further configured to:

[0149] The data filling model is trained by using the load data after removing the abnormal data, to fill in the missing values in the load data, and obtain the preprocessed load data.

[0150] In a possible implementation, the calculation formula of the daily periodicity feature is:

[0151]

[0152] wherein n is the number of load data at the same time each day; x i (h) is the load data corresponding to the time point h; x t is the standardized value of the load data at the time point t;

[0153] The calculation formula of the local difference is:

[0154]

[0155] wherein k is the length of the local change window; x c (i) is the load data at the time point i.

[0156] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in detail in a certain embodiment can be referred to the related description of other embodiments.

[0157] Those skilled in the art can realize that the templates, units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0158] The modules / units, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various power system fast ramping and switching scene point extraction method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0159] The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, rather than limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for extracting scene points for rapid ramping and switching in a power system, characterized in that, include: Obtain load data of the target power system within a preset time period; The Adaptive Periodic Local Feature Anomaly Detection Algorithm (APLAD) is used to determine the daily periodicity and local differences of the load data. Based on the daily periodicity and local differences, abnormal data in the load data are removed to obtain preprocessed load data. The preprocessed load data is reconstructed to obtain multiple sets of reconstructed data; and multiple sets of fused feature data are obtained based on the multiple sets of reconstructed data. Based on the multiple sets of fused feature data, the rapid ramp-up and switching scenario points of the target power system are obtained; The process of obtaining multiple sets of fused feature data based on the multiple sets of reconstructed data includes: For any set of reconstructed data, calculate the time-domain characteristics, statistical characteristics, and periodic characteristics of the reconstructed data. Based on the time domain characteristics, statistical characteristics, and periodic characteristics of this set of reconstructed data, the short-term local features and long-term trend features corresponding to this set of reconstructed data are extracted; The short-term local features and long-term trend features corresponding to the reconstructed data are weighted and fused to obtain the fused feature data corresponding to the reconstructed data. The maximum value, minimum value, change, fluctuation range, peak value, amplitude, frequency, period, and waveform of each set of reconstructed data are used as the statistical features; the main frequency cost and the amplitude corresponding to each set of reconstructed data are extracted by Fourier transform and used as the periodic features. The process of obtaining the rapid ramp-up and switching scenario points of the target power system based on the multiple sets of fused feature data includes: An improved density-based clustering algorithm, DBSCAN, is used to cluster the multiple sets of fused features to obtain the rapid ramp-up and switching scenario points of the target power system. The improved DBSCAN introduces the giant lily pad optimization algorithm into the traditional DBSCAN to adaptively adjust the numerical differences and changing trends of the data during the calculation process.

2. The method for extracting rapid ramp-up and switching scenario points in a power system according to claim 1, characterized in that, The process of removing outlier data from the load data based on the daily periodicity and local differences includes: For any one of the load data, a comprehensive score for that load data is obtained based on its daily periodicity characteristics and local differences. When the overall score of the load data exceeds the preset threshold for anomalies, the load data is removed as an anomaly.

3. The method for extracting rapid ramp-up and switching scenario points in a power system according to claim 2, characterized in that, The formula for calculating the comprehensive score is as follows: in, and Adjust the coefficients separately; For the aforementioned local differences; This refers to the daily periodicity characteristic.

4. The method for extracting rapid ramp-up and switching scenario points in a power system according to claim 1, characterized in that, The improved density-based clustering algorithm DBSCAN is used to cluster the multiple sets of fused features to obtain the rapid ramp-up and switching scenario points of the target power system, including: An improved density-based clustering algorithm, DBSCAN, is used to cluster the multiple sets of fused features to obtain multiple cluster points. For any given cluster point, determine all data points within a preset distance threshold range for that cluster point; if the number of data points is greater than a preset number, then establish the cluster point and all data points within the preset distance threshold range as a new cluster label; if the number of data points is not greater than the preset number, then establish the cluster point and all data points within the preset distance threshold range as a fast ramp and scene switching point label. Traverse all cluster points and access all data points within the preset distance threshold range corresponding to these cluster points; if there are unlabeled data points, determine whether the data point is a cluster point; if not, assign the label corresponding to its cluster point to the data point; if so, explore the objects that can be reached by its density until no further exploration is possible, obtain a complete cluster, and return to the step of determining all data points within the preset distance threshold range for any given cluster point, until there are no more unlabeled data points, and output the clustering result; Based on the clustering results, the rapid ramp-up and switching scenario points of the target power system are obtained.

5. The method for extracting rapid ramp-up and switching scenario points in a power system according to claim 1, characterized in that, After obtaining the rapid ramp-up and switching scenario points of the target power system based on the fused feature data, the method further includes: The rapid ramp-up and scene switching points are reverse-mapped to the corresponding time series or observation intervals in the load data; Based on the indexing operation, the position of each rapid ramp and switching scenario point in the load data is located, and the multidimensional features and time series information of each rapid ramp and switching scenario point are recovered.

6. The method for extracting rapid ramp-up and switching scenario points in a power system according to claim 1, characterized in that, After removing outlier data from the load data based on the daily periodicity and local differences, the method further includes: A data imputation model was trained using the load data after removing outliers to fill in missing values ​​in the load data, resulting in preprocessed load data.

7. The method for extracting rapid ramp-up and switching scenario points in a power system according to claim 1, characterized in that, The formula for calculating the diurnal periodicity is as follows: in, n This refers to the number of load data points at the same time each day. The load data corresponding to time point h; Number of load data at time points The standardized value; The formula for calculating the local differences is: in, k The length of the locally varying window; For at a certain point in time Load data at the location.

8. A device for extracting scene points for rapid ramping and switching in a power system, characterized in that, include: The data acquisition module is used to acquire load data of the target power system within a preset time period; The processing module is used to use the Adaptive Periodic Local Feature Anomaly Detection Algorithm (APLAD) to determine the daily periodicity and local differences of the load data, and based on the daily periodicity and local differences, remove abnormal data from the load data to obtain preprocessed load data. The extraction module is used to reconstruct the preprocessed load data to obtain multiple sets of reconstructed data; and to obtain multiple sets of fused feature data based on the multiple sets of reconstructed data. The clustering module is used to obtain the rapid ramp-up and switching scenario points of the target power system based on the multiple sets of fused feature data; the extraction module is specifically used for: For any set of reconstructed data, calculate the time-domain characteristics, statistical characteristics, and periodic characteristics of the reconstructed data. Based on the time domain characteristics, statistical characteristics, and periodic characteristics of this set of reconstructed data, the short-term local features and long-term trend features corresponding to this set of reconstructed data are extracted; The short-term local features and long-term trend features corresponding to the reconstructed data are weighted and fused to obtain the fused feature data corresponding to the reconstructed data. The maximum value, minimum value, change, fluctuation range, peak value, amplitude, frequency, period, and waveform of each set of reconstructed data are used as the statistical features; the main frequency cost and the amplitude corresponding to each set of reconstructed data are extracted by Fourier transform and used as the periodic features. The clustering module is specifically used for: An improved density-based clustering algorithm, DBSCAN, is used to cluster the multiple sets of fused features to obtain the rapid ramp-up and switching scenario points of the target power system. The improved DBSCAN introduces the giant lily pad optimization algorithm into the traditional DBSCAN to adaptively adjust the numerical differences and changing trends of the data during the calculation process.

Images