A method and system for power system data repair based on data completion and error correction
By using a missing data completion method based on DTW distance and dynamic weight configuration, combined with data simulation correction, the problems of missing and error data in power system are solved, achieving high-precision data completion and frequency optimization, and improving the quality of power system data.
Patent Information
- Application Number
- CN202511434493.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-09
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2045-10-09
AI Technical Summary
In existing technologies, power system data suffers from missing and error issues during the acquisition process. The accuracy of missing data completion is low, the error correction is not targeted enough, and there is a lack of dynamic acquisition frequency adjustment mechanism, resulting in insufficient data reliability and integrity.
A missing data completion method based on DTW distance is adopted, which combines dynamic weight configuration and data simulation completion to correct errors in real time, and adjust the acquisition frequency according to the running data of missing data completion.
It improves the accuracy of missing data completion and the pertinence of error correction, enables dynamic adjustment of data acquisition frequency, and enhances the integrity, accuracy, and reliability of power system data.
Smart Images

Figure CN120910035B_ABST
Abstract
Description
Technical Field
[0001] This invention proposes a method and system for power system data repair based on data completion and error correction, belonging to the field of power system data processing technology. Background Technology
[0002] Power system data is the core basis for power network monitoring, dispatch optimization, fault early warning, and equipment maintenance. Its completeness and accuracy directly affect the safe and stable operation of the power system. With the advancement of smart grid construction, the scale of power system data collection is constantly expanding, covering multi-dimensional information such as voltage, current, power, and equipment status. The collection frequency is also gradually increasing to meet the needs of real-time monitoring. However, in actual data acquisition, due to factors such as equipment failure, communication link interruption, electromagnetic interference, and insufficient sensor accuracy, power system data inevitably suffers from missing and error issues. For example, data transmission may be interrupted due to weak communication signals in remote distribution terminals, resulting in missing data; environmental factors such as high temperature and humidity may cause measuring instruments to drift, introducing erroneous data. In existing technologies, methods for completing missing data often employ linear interpolation, mean imputation, or single machine learning models (such as BP neural networks). These methods struggle to capture the temporal correlations and periodic characteristics of power system data, resulting in low completion accuracy, especially in scenarios with long missing sequences. For error correction, traditional methods rely heavily on fixed threshold judgments or simple filtering algorithms, failing to fully consider the inherent error characteristics of power measurement instruments (such as nonlinear errors and temperature drift errors), leading to unsatisfactory correction results. Furthermore, existing technologies typically lack dynamic evaluation mechanisms for data acquisition quality, failing to adaptively adjust the acquisition frequency based on the operational status of missing data completion (such as the percentage of missing data and completion time). This can result in resource waste due to excessively high acquisition frequencies or exacerbated data loss due to excessively low frequencies. Therefore, there is an urgent need for a power system data repair method that can comprehensively address the issues of low accuracy in data missing data completion, insufficient targeted error correction, and lack of dynamic adjustment of acquisition frequency, in order to improve the reliability and effectiveness of power system data. Summary of the Invention
[0003] This invention provides a power system data repair method and system based on data completion and error correction to solve the technical problems existing in the prior art. The technical solution adopted is as follows: A power system data repair method based on data completion and error correction, the power system data repair method includes: Real-time acquisition of power system data; The real-time power system data is detected to obtain missing data in the power system data summary, and the missing data is completed using a missing data completion method based on DTW distance to obtain the completed power system data. Error correction is performed on the completed power system data using a method based on data simulation completion and comparison with actual data, to obtain the corrected power system data; The system collects operational data in real time to complete missing data, detects the quality of power system data collection based on the completed operational data, and adjusts the power system data collection frequency.
[0004] Furthermore, real-time power system data is collected, including: The power system data is collected in real time using a power big data edge acquisition system architecture, wherein the power system data includes power process data, environmental data and field status data; The power system data is denoised to obtain real-time denoised power system data.
[0005] Furthermore, the power big data edge acquisition system architecture includes multiple edge nodes and an edge gateway module; wherein, the multiple edge nodes are connected to the edge gateway module; and the edge gateway module is connected to the missing data completion module. The edge node is used to collect power system data in real time and send the power system data to the edge gateway module; The edge gateway module is used to convert and encrypt the received power system data, and then send the converted and encrypted power system data to the missing data completion module.
[0006] Furthermore, missing data completion methods based on DTW distance include: The optimal set of similar segments S is determined by using a preset initial distance threshold D0 and the similarity between the candidate sequence set S1 and the target sequence corresponding to the missing data; The real-time operating condition adaptation coefficient is used to dynamically configure the weights of each historical sequence contained in the optimal similar segment set S. The missing data is categorized, and the missing data is filled in according to the categorization results.
[0007] Furthermore, the optimal similar segment set S is determined using a preset initial distance threshold D0 and the similarity between the candidate sequence set S1 and the target sequence corresponding to the missing data, including: A preset initial distance threshold D0 is retrieved, wherein the preset initial distance threshold D0 is 60% of the average DTW distance between all historical sequences and the target sequence corresponding to the missing data; The DTW distance between each historical sequence and the target sequence is compared with a preset initial distance threshold D0, and historical sequences with a DTW distance lower than the preset initial distance threshold D0 are selected as candidate sequence set S1; The cosine similarity T between each historical sequence and the target sequence in the candidate sequence set S1 is obtained using the first-order difference method. A preset similarity threshold is retrieved, wherein the preset similarity threshold is 0.7; The cosine similarity T between each historical sequence in the candidate sequence set S1 and the target sequence is compared with a preset similarity threshold, and historical sequences with a cosine similarity T not lower than the preset similarity threshold are selected to form a sequence set S2. The positions of abrupt change between the target sequence and the historical sequences contained in the sequence set S2 are extracted using wavelet transform, and the number of matching positions of abrupt change between the target sequence and the historical sequences contained in the sequence set S2 is obtained. The ratio of the number of matching mutation point locations to the total number of mutation points in the historical sequences contained in the sequence composition set S2 is used to obtain the mutation point matching rate between the target sequence and the historical sequences contained in the sequence composition set S2. Historical sequences with a mutation point matching rate exceeding a preset mutation point matching rate threshold are selected to generate an optimal set of similar segments S, wherein the preset mutation point matching rate threshold is 0.6.
[0008] Furthermore, the dynamic weights of each historical sequence contained in the optimal similar segment set S are configured using real-time operating condition adaptation coefficients, including: Each historical sequence contained in the optimal similar segment set S is retrieved as a reference historical sequence; Retrieve the DTW distance of each reference historical sequence, and use the reciprocal of the DTW distance of the reference historical sequence as the base weight value of each reference historical sequence; Retrieve the power system operating parameters at the corresponding time of each reference historical sequence and the power system operating parameters at the corresponding time of the target sequence, and obtain the Euclidean distance between the power system operating parameters at the corresponding time of each reference historical sequence and the power system operating parameters at the corresponding time of the target sequence. The dynamic weight of each reference historical sequence is obtained by combining the Euclidean distance between the power system operating parameters at the corresponding time of each reference historical sequence and the power system operating parameters of the target sequence with the basic weight value of each reference historical sequence.
[0009] Furthermore, the missing data is categorized, and missing data is filled in according to the categorization results, including: Retrieve the number of data collection points contained in the target sequence corresponding to the missing data; The number of data collection points contained in the target sequence is compared with a preset point count threshold. Missing data with more than the preset point count threshold is marked as a long missing segment, and missing data with less than the preset point count threshold is marked as a short missing segment; wherein, the preset point count threshold is 3. The missing data corresponding to the missing data in the long missing data segment and the short missing data segment are obtained by using the long missing data completion model and the short missing data completion model respectively, and the missing data in the long missing data segment and the short missing data segment are replaced by the completed data respectively.
[0010] Furthermore, error correction is performed on the completed power system data using a method based on data simulation completion and comparison with actual data to obtain corrected power system data, including: The time series Δt = (Δt1, Δt2, ..., Δt) is completed by real-time data retrieval. m ) Valid power system data of the same type as the missing data; where Δt1, Δt2, ..., Δt m These represent the 1 to m data collection times contained in the time series; The missing data completion method based on DTW distance is used to simulate and complete the data according to the data collection time included in the time series Δt, and the simulated and completed data corresponding to time Δt after data completion is obtained. Based on the data completion, the simulated completed data and time series Δt = (Δt1, Δt2, ..., Δt) corresponding to time Δt after data completion are given. m The corresponding error correction item for the actual collected valid power system data; The error correction term is used to correct the error in the completed data corresponding to the missing data, and the error-corrected completed data is obtained. The error-corrected completed data is the corrected power system data.
[0011] Furthermore, after the power system data has been corrected, operational data to fill in missing data is collected in real time, and the frequency of power system data collection is adjusted, including: After the power system data is corrected, the operation data during the missing data completion process is collected in real time. The operation data during the missing data completion process includes the proportion of missing data, the number of data collection points contained in each missing data segment, and the completion time of each missing data segment. The proportion of missing data, the number of data collection points contained in each missing data segment, and the time for completing each missing data segment are normalized to obtain the normalized proportion of missing data, the number of data collection points contained in each missing data segment, and the time for completing each missing data segment. The data collection frequency adjustment coefficient is obtained by using the proportion of missing data after normalization, the number of data collection points contained in each missing data segment, and the time for completing each missing data segment. The data acquisition frequency adjustment coefficient is used to adjust the acquisition frequency of the power system data to obtain the adjusted acquisition frequency.
[0012] Furthermore, the quality of power system data acquisition is tested based on the operational data supplemented by the missing data, including: The percentage of missing data and the number of data collection points contained in each segment of missing data are retrieved from the running data during the missing data completion process. The percentage of missing data is compared with a preset percentage threshold. When the proportion of missing data exceeds a preset proportion threshold, the number of data collection points contained in each segment of missing data is compared with a preset number of points threshold. Based on the comparison between the number of data collection points contained in each missing data segment and the preset number of points threshold, missing data segments with the number of data collection points exceeding the preset number of points threshold are selected as target missing data segments. Data quality parameters are obtained by combining the data volume corresponding to the target missing data segment, the total data volume of missing data, and the proportion of missing data. A power system data repair system based on data completion and error correction, the power system data repair system comprising: The data acquisition module is used to collect power system data in real time and obtain real-time power system data; The missing data completion module is used to detect the power system data, obtain the missing data in the power system data summary, and complete the missing data using the missing data completion method based on DTW distance to obtain the completed power system data. The error data correction module is used to perform error correction on the completed power system data by using a method based on data simulation completion and comparison with actual data, and to obtain the corrected power system data. The data acquisition quality monitoring module is used to collect operational data with missing data in real time after the power system data has been corrected, to detect the acquisition quality of the power system data based on the operational data with missing data, and to adjust the acquisition frequency of the power system data.
[0013] Beneficial effects of this invention: This invention proposes a power system data repair method and system based on data completion and error correction, improving the accuracy of missing data completion. By employing a DTW distance-based completion method, the temporal correlation and periodicity characteristics of power system data are effectively captured, especially in scenarios with long-sequence missing data, where the completion effect is superior to traditional methods. The method also enhances the targeting and accuracy of error correction. Utilizing the error model of power measurement instruments, the inherent error characteristics of the instruments are fully considered for correction, improving the unsatisfactory correction effect of traditional methods. Furthermore, the method enables dynamic adjustment of the data acquisition frequency. The acquisition frequency is adjusted in real time based on the operational data from missing data completion, avoiding resource waste caused by excessively high acquisition frequencies and the exacerbation of data loss due to excessively low frequencies, thus improving the rationality and efficiency of data acquisition. Overall, the method improves the integrity, accuracy, and reliability of power system data, providing high-quality data support for power network monitoring, dispatch optimization, fault early warning, and equipment maintenance. Attached Figure Description
[0014] Figure 1 This is a flowchart of the method described in this invention; Figure 2 This is a system block diagram of the system described in this invention. Detailed Implementation
[0015] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0016] This invention proposes a power system data repair method based on data completion and error correction, such as... Figure 1 As shown, the power system data repair method includes: Real-time acquisition of power system data; The real-time power system data is detected to obtain missing data in the power system data summary, and the missing data is completed using a missing data completion method based on DTW distance to obtain the completed power system data. Error correction is performed on the completed power system data using a method based on data simulation completion and comparison with actual data, to obtain the corrected power system data; The system collects operational data in real time to complete missing data, detects the quality of power system data collection based on the completed operational data, and adjusts the power system data collection frequency.
[0017] The working principle of the above technical solution is as follows: Real-time acquisition of power system data provides the raw data foundation for subsequent processing. The acquired power system data is inspected to identify missing data, which is then filled using a DTW distance-based missing data completion method. This method utilizes DTW distance to accurately filter historical data segments similar to the sequence containing the missing data. Combined with dynamic weight allocation and segmented completion strategies, it performs targeted completion on missing data of different lengths to obtain the completed power system data. The completed power system data is then corrected. Errors in the data are accurately identified and corrected to obtain corrected power system data. Real-time acquisition of operational data during the missing data completion process includes the proportion of missing data, the number of data collection points in each missing data segment, and the completion time for each missing data segment. Based on this data, a data acquisition frequency adjustment coefficient is obtained, thereby adjusting the power system data acquisition frequency to achieve dynamic optimization of the acquisition frequency.
[0018] The above technical solutions achieve the following effects: Improved accuracy in missing data completion. By employing a DTW distance-based completion method, the temporal correlation and periodicity characteristics of power system data are effectively captured. Especially in scenarios with long missing data sequences, the completion effect is superior to traditional methods, enhancing the targeting and accuracy of error correction. Utilizing the error model of power measurement instruments, the inherent error characteristics of the instruments are fully considered for correction, improving upon the unsatisfactory correction results of traditional methods. Dynamic adjustment of data acquisition frequency is achieved. The acquisition frequency is adjusted in real time based on the operational data from missing data completion, avoiding resource waste caused by excessively high acquisition frequencies and the exacerbation of data loss due to excessively low frequencies, thus improving the rationality and efficiency of data acquisition. Overall, the integrity, accuracy, and reliability of power system data are improved, providing high-quality data support for power network monitoring, dispatch optimization, fault early warning, and equipment maintenance.
[0019] One embodiment of the present invention involves real-time acquisition of power system data, including: The power system data is collected in real time using a power big data edge acquisition system architecture, wherein the power system data includes power process data, environmental data and field status data; The power system data is denoised to obtain real-time denoised power system data.
[0020] The power big data edge acquisition system architecture includes multiple edge nodes and an edge gateway module; the multiple edge nodes are connected to the edge gateway module; and the edge gateway module is connected to a missing data completion module. The edge node is used to collect power system data in real time and send the power system data to the edge gateway module; The edge gateway module is used to convert and encrypt the received power system data, and then send the converted and encrypted power system data to the missing data completion module.
[0021] The working principle of the above technical solution is as follows: Multiple edge nodes collect power system data in real time. This power system data includes power process data, environmental data, and field status data, and then sends the collected data to the edge gateway module. After receiving the data from the edge nodes, the edge gateway module performs format conversion and encryption to ensure data format uniformity and security. The processed power system data is then sent to the missing data completion module. Simultaneously, the power system data undergoes noise reduction processing to ultimately obtain real-time, denoised power system data.
[0022] The above technical solution achieves the following effects: This embodiment realizes real-time acquisition of multiple types of data from the power system. Utilizing multiple edge nodes, it comprehensively captures multi-dimensional data such as power processes, environment, and on-site conditions, ensuring the comprehensiveness and real-time nature of data acquisition. It improves the efficiency and security of data processing; the format conversion function of the edge gateway module unifies data from different sources, facilitating subsequent processing; encryption ensures the security of data transmission and prevents data leakage. Noise reduction processing effectively reduces noise interference in the data, providing more reliable foundational data for subsequent missing data completion and other data processing stages.
[0023] An embodiment of the present invention provides a missing data completion method based on DTW distance, comprising: The optimal set of similar segments S is determined by using a preset initial distance threshold D0 and the similarity between the candidate sequence set S1 and the target sequence corresponding to the missing data; The real-time operating condition adaptation coefficient is used to dynamically configure the weights of each historical sequence contained in the optimal similar segment set S. The missing data is categorized, and the missing data is filled in according to the categorization results.
[0024] The working principle of the above technical solution is as follows: First, using a preset initial distance threshold D0 as the screening condition, and combining the similarity between each sequence in the candidate sequence set S1 and the target sequence corresponding to the missing data, sequences that meet the requirements are selected from the candidate sequences to form the optimal similarity fragment set S, providing a reliable historical data foundation for subsequent completion. Then, for each historical sequence included in the optimal similarity fragment set S, a real-time operating condition adaptation coefficient is introduced. Based on the degree of matching between the current real-time operating conditions of the power system (such as system load, voltage deviation, etc.) and the corresponding operating conditions of the historical sequence, the weight of each historical sequence is dynamically adjusted to make the weight allocation more consistent with the actual operating state. Finally, the missing data is classified according to preset classification criteria (such as the length and distribution characteristics of the missing data). Then, based on different classification results, corresponding completion strategies are adopted (such as using a weighted average of adjacent points for short missing segments and interpolation followed by weighting for long missing segments) to complete the missing data, thereby improving the targeting and accuracy of the completion.
[0025] The above technical solution achieves the following effects: It improves the accuracy of similar segment selection. Through dual selection using an initial distance threshold and similarity, it ensures that the historical sequences in the optimal similar segment set S are highly similar to the target sequence, providing high-quality data references for completion. It enhances the adaptability of weight allocation. Dynamic weight configuration based on real-time operating condition adaptation coefficients allows the weights of each historical sequence to adjust with changes in operating conditions, avoiding the problem of completion results being out of sync with actual operating conditions caused by fixed weights. It improves the accuracy of missing data completion. By classifying missing data and adopting corresponding completion strategies, it achieves targeted processing of different types of missing data, effectively improving the performance of single completion methods in complex missing data scenarios. Overall, it enhances the reliability of missing data completion, providing strong support for the integrity of power system data and facilitating the smooth progress of subsequent data processing and analysis.
[0026] In one embodiment of the present invention, the optimal similar segment set S is determined by using a preset initial distance threshold D0 and the similarity between the candidate sequence set S1 and the target sequence corresponding to the missing data, including: A preset initial distance threshold D0 is retrieved, wherein the preset initial distance threshold D0 is 60% of the average DTW distance between all historical sequences and the target sequence corresponding to the missing data; The DTW distance between each historical sequence and the target sequence is compared with a preset initial distance threshold D0, and historical sequences with a DTW distance lower than the preset initial distance threshold D0 are selected as candidate sequence set S1; The cosine similarity T between each historical sequence and the target sequence in the candidate sequence set S1 is obtained using the first-order difference method. A preset similarity threshold is retrieved, wherein the preset similarity threshold is 0.7; The cosine similarity T between each historical sequence in the candidate sequence set S1 and the target sequence is compared with a preset similarity threshold, and historical sequences with a cosine similarity T not lower than the preset similarity threshold are selected to form a sequence set S2. The positions of abrupt change between the target sequence and the historical sequences contained in the sequence set S2 are extracted using wavelet transform, and the number of matching positions of abrupt change between the target sequence and the historical sequences contained in the sequence set S2 is obtained. The ratio of the number of matching mutation point locations to the total number of mutation points in the historical sequences contained in the sequence composition set S2 is used to obtain the mutation point matching rate between the target sequence and the historical sequences contained in the sequence composition set S2. Historical sequences with a mutation point matching rate exceeding a preset mutation point matching rate threshold are selected to generate an optimal set of similar segments S, wherein the preset mutation point matching rate threshold is 0.6.
[0027] The working principle of the above technical solution is as follows: A preset initial distance threshold D0 is retrieved (its value is 60% of the average DTW distance between all historical sequences and the target sequence). The DTW distance of each historical sequence and the target sequence is compared with D0. Historical sequences with a DTW distance lower than D0 are selected to form a candidate sequence set S. 1。 The cosine similarity T between each historical sequence in the candidate sequence set S1 and the target sequence is calculated using the first-order difference method. A preset similarity threshold (0.7) is used to filter historical sequences whose cosine similarity T is not lower than this threshold, forming a sequence composition set S2. Wavelet transform is used to extract the mutation point positions of the target sequence and the historical sequences in the sequence composition set S2. The number of matching mutation point positions between the two is calculated, and then the ratio is processed with the total number of mutation points of the historical sequence itself to obtain the mutation point matching rate. A preset mutation point matching rate threshold (0.6) is used to filter historical sequences whose mutation point matching rate exceeds this threshold, generating the optimal similar segment set S.
[0028] The effects of the above technical solution are as follows: This embodiment improves the hierarchy and accuracy of similar segment screening. Through multi-dimensional screening using DTW distance, cosine similarity, and mutation point matching rate, mismatched sequences are gradually eliminated, ensuring that sequences in the optimal similar segment set S highly match the target sequence in terms of overall distance, trend, and mutation characteristics. It enhances the objectivity and specificity of the screening criteria. Each threshold is set based on data characteristics (e.g., D0 is based on the average DTW distance, and the similarity threshold and mutation point matching rate threshold are based on experience and data characteristics), avoiding the arbitrariness of subjective settings and making the screening results more aligned with actual needs. It provides high-quality basic data for subsequent missing data completion. The accurate determination of the optimal similar segment set S reduces the interference of invalid historical sequences on the completion results, laying a reliable foundation for improving completion accuracy. Overall, it improves the efficiency of similar segment screening. Through multi-round progressive screening, while ensuring screening quality, the candidate range is gradually narrowed, reducing the computational load of subsequent processing.
[0029] Meanwhile, this embodiment employs a progressive application of multi-dimensional thresholds. First, an initial distance threshold based on historical data statistics (60% of the average DTW distance) is used to filter basic candidate sequences. Then, cosine similarity (threshold 0.7) is used for further refinement. Finally, mutation point matching rate (threshold 0.6) is combined to achieve deep filtering. This effectively avoids the limitations of single-indicator judgment, significantly improving the accuracy and targeting of similar segment selection. The optimal set of similar segments selected is more closely aligned with the target sequence in terms of overall trend, local features, and mutation patterns, reducing interference from irrelevant sequences. Furthermore, by integrating three indicators—DTW distance (measuring overall sequence similarity), first-order difference cosine similarity (reflecting the consistency of sequence change rates), and mutation point matching rate (capturing the degree of agreement at key turning points)—sequence similarity is comprehensively characterized from global trends to local details, and from stationary segments to mutation segments. This overcomes the one-sidedness caused by relying solely on a single similarity metric in existing technologies. It is particularly effective for sequences containing mutation features, enhancing the comprehensiveness and robustness of similarity judgment. Furthermore, the initial distance threshold is dynamically determined based on the average DTW distance between historical sequences and the target sequence, while the similarity threshold and mutation point matching rate threshold are set in conjunction with sequence feature analysis. This avoids the arbitrariness of setting thresholds based on subjective experience. This threshold system based on data statistics and feature patterns allows the screening criteria to better adapt to the sequence characteristics under different scenarios, improving the solution's adaptability to diverse sequence data. The high precision and comprehensiveness of the optimal similar fragment set provides a high-quality data foundation for subsequent tasks based on similar fragments (such as missing data completion and trend prediction), reducing processing errors caused by improper selection of similar fragments, thereby improving the accuracy and reliability of the overall data processing results.
[0030] One embodiment of the present invention utilizes real-time operating condition adaptation coefficients to dynamically configure the weights of each historical sequence contained in the optimal similar segment set S, including: Each historical sequence contained in the optimal similar segment set S is retrieved as a reference historical sequence; Retrieve the DTW distance of each reference historical sequence, and use the reciprocal of the DTW distance of the reference historical sequence as the base weight value of each reference historical sequence; Retrieve the power system operating parameters at the corresponding time of each reference historical sequence and the power system operating parameters at the corresponding time of the target sequence, and obtain the Euclidean distance between the power system operating parameters at the corresponding time of each reference historical sequence and the power system operating parameters at the corresponding time of the target sequence. The dynamic weight of each reference historical sequence is obtained by combining the Euclidean distance between the power system operating parameters at the corresponding time of each reference historical sequence and the power system operating parameters of the target sequence with the basic weight value of each reference historical sequence.
[0031] The dynamic weight of each reference historical sequence is obtained using the following formula: in, w represents the dynamic weight of the i-th reference history sequence; i L represents the base weight value of the i-th reference history sequence; i L represents the Euclidean distance between the power system operating parameters of the i-th reference historical sequence and the power system operating parameters of the target sequence; h This represents the sum of Euclidean distances corresponding to all reference history sequences.
[0032] The working principle of the above technical solution is as follows: First, select historical sequences from the optimal set of similar segments S as reference sequences. Use the reciprocal of the DTW distance as the base weight (since the DTW distance reflects sequence similarity, the smaller the distance, the more similar the sequences; taking the reciprocal gives higher base weights to sequences with high similarity) to initially distinguish sequence value. Extract the power condition parameters (such as voltage, current, load, etc.) at corresponding times for the reference and target sequences, calculate the Euclidean distance between them, and quantify the difference in power conditions—the smaller the Euclidean distance, the closer the power conditions. Then, use the formula... , where L h The basic weights are combined with the differences in operating conditions, using the sum of Euclidean distances across all reference sequences. Then, the basic weights are adjusted using the proportion of Euclidean distances to make the operating conditions closer to the reference sequences of the target sequence. The dynamic weights are strengthened, achieving a two-dimensional weighting of "similarity + operating condition adaptation".
[0033] The above technical solution achieves the following effects: It enhances the hierarchy and accuracy of similar segment selection. By using multi-dimensional screening based on DTW distance, cosine similarity, and mutation point matching rate, mismatched sequences are gradually eliminated, ensuring that sequences in the optimal similar segment set S highly match the target sequence in terms of overall distance, trend, and mutation characteristics. It strengthens the objectivity and specificity of the screening criteria; each threshold is set based on data characteristics, avoiding arbitrariness in subjective settings and making the screening results more aligned with actual needs. It provides high-quality foundational data for subsequent missing data completion; the accurate determination of the optimal similar segment set S reduces the interference of invalid historical sequences on the completion results, laying a reliable foundation for improving completion accuracy. Overall, it improves the efficiency of similar segment selection; through multi-round progressive screening, while ensuring screening quality, the candidate range is gradually narrowed, reducing the computational load of subsequent processing.
[0034] Meanwhile, traditional weighting methods often have limitations when determining the weights of historical sequences because they rely on only a single dimension of information. For example, while using Euclidean distance alone can measure the numerical differences in certain dimensions between sequences, it is easily affected by distortion along the time axis when processing time series, making it difficult to effectively identify potential patterns and failing to fully consider the overall shape of the sequence and temporal misalignment. On the other hand, using only the reciprocal of the DTW distance as the weight can better handle the scaling and translation of time series and achieve the matching of sequences with similar shapes but misaligned time sequences, but it ignores the impact of differences in real-time operating parameters on the weight.
[0035] This embodiment combines the two approaches, effectively solving the aforementioned problems. On one hand, it effectively improves the comprehensiveness and accuracy of weight determination. By considering the Euclidean distance between the reference historical sequence and the target sequence in terms of power system operating parameters, real-time operating conditions are incorporated into the weighting considerations. Different power system operating conditions have a significant impact on sequence similarity and subsequent applications. Furthermore, by combining the reciprocal of the DTW distance as the basic weight, the overall similarity of the sequences is taken into account. Weights are determined comprehensively from two important aspects: sequence morphology and real-time operating conditions. Compared to a single method, this significantly improves the accuracy of the weights.
[0036] On the other hand, it enhances the model's adaptability to dynamically changing environments. Power system operating conditions are constantly changing, and this solution can continuously adjust the weights of historical sequences based on real-time changes in operating parameters. When operating conditions change, changes in Euclidean distance cause corresponding changes in dynamic weights, ensuring that historical sequences can participate in subsequent data processing with appropriate weights under different operating conditions. For example, in different operating scenarios such as sudden changes in power system load and voltage fluctuations, tasks such as analysis and prediction based on dynamic weights can maintain high accuracy, overcoming the shortcomings of traditional fixed-weight methods that are difficult to adapt to dynamic changes in operating conditions. This improves the reliability and practicality of the entire technical solution in complex and ever-changing power system environments.
[0037] One embodiment of the present invention involves classifying missing data and completing the missing data according to the classification results, including: Retrieve the number of data collection points contained in the target sequence corresponding to the missing data; The number of data collection points contained in the target sequence is compared with a preset point count threshold. Missing data with more than the preset point count threshold is marked as a long missing segment, and missing data with less than the preset point count threshold is marked as a short missing segment; wherein, the preset point count threshold is 3. The missing data corresponding to the missing data in the long missing data segment and the short missing data segment are obtained by using the long missing data completion model and the short missing data completion model respectively, and the missing data in the long missing data segment and the short missing data segment are replaced by the completed data respectively.
[0038] The structure of the long missing data completion model is as follows: Among them, X ct This represents the completed data corresponding to long missing data; n represents the total number of reference historical sequences contained in the best similar segment set S; Indicates the dynamic weight of the i-th reference history sequence; interp(Y) i ,t) represents the data value obtained by linear interpolation of the t-th acquisition point corresponding to the short missing data in the i-th reference history sequence; The structure of the short missing data completion model is as follows: Among them, X dt This represents the completed data corresponding to the short missing data; n represents the total number of reference historical sequences contained in the best similar fragment set S; Y represents the dynamic weight of the i-th reference history sequence; it This represents the historical data value of the t-th acquisition point in the i-th reference historical sequence that corresponds to the short missing data.
[0039] The working principle of the above technical solution is as follows: The number of data collection points in the target sequence corresponding to the missing data is retrieved and used as the classification basis. The number of data collection points in the target sequence is compared with a preset point threshold (valued at 3). If the threshold is exceeded, the corresponding missing data is marked as a long missing segment; if not, it is marked as a short missing segment. For long and short missing segments, long and short missing data completion models are used respectively to obtain the corresponding completed data. The completed data replaces the original missing data, thus completing the missing data.
[0040] The above technical solution achieves the following effects: It clearly distinguishes between long and short missing segments by using the number of data collection points as a quantitative standard and combining preset thresholds, resulting in a clear classification logic and laying the foundation for subsequent targeted data completion. It improves the adaptability of missing data completion by employing dedicated completion models for missing segments of different lengths, avoiding the problem of a single model being insufficiently adaptable to different missing types, and making the completion method more suitable for the characteristics of various types of missing data. It enhances the accuracy of missing data completion; the differentiated design of the long and short missing segment completion models better addresses the completion needs of missing data of different lengths, reduces completion errors, and ensures the reliability of the completed data. Simultaneously, it enhances the logic and efficiency of the completion process; the categorized processing makes the completion process more organized, avoiding resource waste caused by indiscriminate processing and improving the overall efficiency of missing data completion.
[0041] Meanwhile, by classifying missing data into long and short missing segments, appropriate completion models can be adopted for missing segments of different lengths, improving the targeting of completion. Secondly, the long missing data completion model uses linear interpolation combined with dynamic weights to effectively address the completion problems caused by numerous missing data and incomplete information in long missing segments, making the completed data more closely match the sequence trend. The short missing data completion model is directly based on the historical data of the corresponding points in the reference historical sequence and combined with dynamic weights, ensuring completion efficiency while using the similarity of historical data to guarantee completion accuracy. Furthermore, the introduction of dynamic weights fully considers the differences between the reference historical sequence and the target sequence in terms of sequence similarity and real-time operating conditions, allowing the completion process to dynamically adjust the contribution of each reference sequence according to different situations, further improving the accuracy and reliability of missing data completion, and achieving efficient and accurate completion of different types of missing data overall.
[0042] One embodiment of the present invention utilizes a method based on data simulation completion and comparison with actual data to perform error correction on the completed power system data, thereby obtaining corrected power system data, including: The time series Δt = (Δt1, Δt2, ..., Δt) is completed by real-time data retrieval. m ) Valid power system data of the same type as the missing data; where Δt1, Δt2, ..., Δt m These represent the 1 to m data collection times contained in the time series; The missing data completion method based on DTW distance is used to simulate and complete the data according to the data collection time included in the time series Δt, and the simulated and completed data corresponding to time Δt after data completion is obtained. Based on the data completion, the simulated completed data and time series Δt = (Δt1, Δt2, ..., Δt) corresponding to time Δt after data completion are given. mThe corresponding error correction item for the actual collected valid power system data; The error correction term is used to correct the error in the completed data corresponding to the missing data, and the error-corrected completed data is obtained. The error-corrected completed data is the corrected power system data.
[0043] The structure of the error correction term is as follows: Where S represents the error correction term; m represents the total number of data acquisition times contained within the time series Δt; X si and X mi This represents the actual and valid power system data and simulation completion data collected at the i-th data acquisition time within the time series Δt; Meanwhile, the error-corrected completed data is obtained using the following formula: Among them, X wi X represents the error-corrected completed data corresponding to the i-th missing data; ti represents the completed data corresponding to the i-th missing data; S represents the error correction term.
[0044] The working principle of the above technical solution is as follows: real-time retrieval and data completion of the time series Δt = (Δt1, Δt2, ..., Δt... m The system lacks valid power system data of the same type, from Δt1 to Δt. m These represent data acquisition times 1 to m in the time series. A missing data completion method based on DTW distance is used to simulate and complete the data at each acquisition time in the time series Δt, obtaining the simulated completed data corresponding to time Δt. Based on the simulated completed data at time Δt and the actual valid power system data acquired in the time series Δt, an error correction term is calculated. This error correction term reflects the deviation between the simulated completed data and the actual valid data. The obtained error correction term is used to correct the errors in the completed data corresponding to the missing data. The completed data obtained after correction is the corrected power system data.
[0045] The above technical solution achieves the following effects: By comparing simulated completed data with actual valid data to obtain error correction terms, the correction process has a reliable basis and can specifically correct deviations in the completed data. This improves the reliability of the completed data; adjusting the completed data with error correction terms effectively reduces errors that may be introduced during the completion process, making the final corrected data closer to the real power system data. It enhances the rationality of data correction; since the retrieved valid power system data and the missing data are of the same type and are simulated using the same completion method, the applicability of the error correction terms is ensured, making the correction logic more closely match the characteristics of actual data. It provides high-quality data support for the subsequent application of power system data; the corrected power system data is more accurate and can better meet the data quality requirements of power system monitoring and dispatching.
[0046] Meanwhile, this embodiment improves the accuracy of the completed power system data by comparing simulated data completion with actual data for error correction. Real-time retrieval of similar valid data and simulation completion provides a reliable reference for error correction. By calculating the error correction term, the deviation between the simulated completed data and the actual valid data is precisely quantified. This error correction term is then used to adjust the completion results of missing data, specifically correcting deviations caused by model assumptions, data similarity, and other factors during the completion process. This makes the final corrected power system data closer to reality, enhancing data reliability and usability. Furthermore, as the real-time operating conditions of the power system change, the comparison between simulated completion and actual data dynamically updates the error correction term, allowing the correction process to adapt to dynamic changes in the system without frequent manual adjustments to the correction rules, enhancing the applicability of the solution in complex and variable power scenarios. Through multiple interactive verifications between simulated completion and actual data, the system effectively resists noise interference and occasional abnormal data points. Even with local data fluctuations, the error correction mechanism ensures the overall stability and reliability of the data.
[0047] In one embodiment of the present invention, after the power system data has been corrected, operational data with missing data supplemented is collected in real time, and the collection frequency of the power system data is adjusted, including: After the power system data is corrected, the operation data during the missing data completion process is collected in real time. The operation data during the missing data completion process includes the proportion of missing data, the number of data collection points contained in each missing data segment, and the completion time of each missing data segment. The proportion of missing data, the number of data collection points contained in each missing data segment, and the time for completing each missing data segment are normalized to obtain the normalized proportion of missing data, the number of data collection points contained in each missing data segment, and the time for completing each missing data segment. The data collection frequency adjustment coefficient is obtained by using the proportion of missing data after normalization, the number of data collection points contained in each missing data segment, and the time for completing each missing data segment. The data acquisition frequency adjustment coefficient is used to adjust the acquisition frequency of the power system data to obtain the adjusted acquisition frequency.
[0048] The adjusted sampling frequency is obtained using the following formula: Among them, f t f0 represents the sampling frequency after adjustment; B represents the sampling frequency before adjustment; k represents the percentage of missing data after normalization; C represents the total number of missing data segments that have been completed. i T represents the number of data collection points contained in the i-th segment of missing data after normalization; ei This represents the time required to complete the missing data for the i-th segment after normalization.
[0049] The working principle of the above technical solution is as follows: Real-time acquisition of operational data during the missing data completion process. This data includes the proportion of missing data, the number of data collection points in each missing data segment, and the completion time for each missing data segment, providing raw data for subsequent analysis. The acquired proportion of missing data, the number of data collection points in each missing data segment, and the completion time for each missing data segment are normalized to eliminate the influence of different data units, ensuring that all data are on the same order of magnitude, facilitating comprehensive analysis. Based on the normalized proportion of missing data, the number of data collection points in each missing data segment, and the completion time for each missing data segment, a data acquisition frequency adjustment coefficient is calculated. This coefficient comprehensively reflects the degree of influence of operational data on the acquisition frequency. The current acquisition frequency of the power system data is adjusted using the data acquisition frequency adjustment coefficient to obtain the adjusted acquisition frequency, achieving dynamic optimization of the acquisition frequency.
[0050] The above technical solution achieves the following effects: It enables effective detection of power system data acquisition quality. By collecting and analyzing operational data with missing data completion in real time, it allows for timely understanding of the status of missing data and the difficulty of completion during the data acquisition process, providing a quantitative basis for evaluating acquisition quality. It enables dynamic adjustment of the power system data acquisition frequency. Based on the adjustment coefficient obtained from operational data, the acquisition frequency can be flexibly changed according to the actual data acquisition situation, avoiding the drawbacks of a fixed frequency. It improves the adaptability and rationality of data acquisition. Normalization processing ensures fairness in the analysis of different operational data, making the calculation of adjustment coefficients more scientific, and thus making the adjustment of the acquisition frequency more closely aligned with actual needs. It helps improve the overall efficiency of power system data acquisition. By rationally adjusting the acquisition frequency, while ensuring data quality, it can reduce unnecessary resource consumption and improve the economy and efficiency of data acquisition.
[0051] One embodiment of the present invention involves detecting the acquisition quality of power system data based on the operational data completed with the missing data, including: The percentage of missing data and the number of data collection points contained in each segment of missing data are retrieved from the running data during the missing data completion process. The percentage of missing data is compared with a preset percentage threshold. When the proportion of missing data exceeds a preset proportion threshold, the number of data collection points contained in each segment of missing data is compared with a preset number of points threshold. Based on the comparison between the number of data collection points contained in each missing data segment and the preset number of points threshold, missing data segments with the number of data collection points exceeding the preset number of points threshold are selected as target missing data segments. Data quality parameters are obtained by combining the data volume corresponding to the target missing data segment, the total data volume of missing data, and the proportion of missing data. The data quality parameters are compared with preset data quality parameter thresholds. When the data quality parameters exceed the preset data quality parameter thresholds, the data acquisition quality is determined to be abnormal, and an alarm for abnormal acquisition quality is triggered.
[0052] The data quality parameters are obtained using the following formula: Where U represents the data quality parameter; g represents the number of target missing data segments; G j G represents the amount of data corresponding to the j-th missing data segment; C represents the total amount of missing data; j B represents the number of data collection points contained in the j-th missing data segment after normalization; B represents the proportion of missing data after normalization.
[0053] The working principle of the above technical solution is as follows: Two key indicators are extracted from the missing data completion process: the proportion of missing data (reflecting the overall degree of missing data) and the number of data collection points contained in each segment of missing data (reflecting the scale of a single missing segment). A progressive filtering logic is adopted. First, the proportion of missing data is compared with a preset threshold. Only when the proportion exceeds the limit is the number of collection points for each segment of missing data compared with the threshold, filtering out segments with excessive collection points as "target missing data segments" (focusing on concentrated, large-scale missing data that has a more significant impact on data quality). Based on the number of target missing segments, the data volume corresponding to each segment, and the total amount of missing data, combined with the normalized number of collection points for a single missing segment and the proportion of missing data, a data quality parameter U (comprehensively reflecting the severity, distribution characteristics, and overall impact of missing data) is calculated using a specific formula. The data quality parameter U is compared with a preset threshold. If U exceeds the limit, the data collection quality is determined to be abnormal, and an alarm mechanism is triggered.
[0054] The above technical solution achieves the following results: By progressively filtering through "missing percentage → number of collection points per segment," it prioritizes large-scale, high-impact missing segments, avoiding misjudgments due to minor or scattered missing data, thus focusing the detection more intently on core issues. The data quality parameter U integrates the number of missing segments, data volume, normalized collection points, and missing percentage, comprehensively reflecting the "scale-distribution-overall impact" of the missing data. Compared to a single indicator, it more objectively reflects the true state of data collection quality. A threshold trigger mechanism enables automatic judgment and alarm for abnormal data collection quality, quickly identifying problems and initiating intervention, reducing the impact of data quality issues on power system analysis and decision-making. Detection is conducted based on operational data from the missing data completion process, fully utilizing existing data resources without adding extra collection burden, balancing detection accuracy and implementation efficiency.
[0055] This invention proposes a power system data repair system based on data completion and error correction, such as... Figure 2 As shown, the power system data repair system includes: The data acquisition module is used to collect power system data in real time and obtain real-time power system data; The missing data completion module is used to detect the power system data, obtain the missing data in the power system data summary, and complete the missing data using the missing data completion method based on DTW distance to obtain the completed power system data. The error data correction module is used to perform error correction on the completed power system data by using a method based on data simulation completion and comparison with actual data, and to obtain the corrected power system data. The data acquisition quality monitoring module is used to collect operational data with missing data in real time after the power system data has been corrected, to detect the acquisition quality of the power system data based on the operational data with missing data, and to adjust the acquisition frequency of the power system data.
[0056] The working principle of the above technical solution is as follows: Real-time acquisition of power system data provides the raw data foundation for subsequent processing. The acquired power system data is inspected to identify missing data, which is then filled using a DTW distance-based missing data completion method. This method utilizes DTW distance to accurately filter historical data segments similar to the sequence containing the missing data. Combined with dynamic weight allocation and segmented completion strategies, it performs targeted completion on missing data of different lengths to obtain the completed power system data. The completed power system data is then corrected. Errors in the data are accurately identified and corrected to obtain corrected power system data. Real-time acquisition of operational data during the missing data completion process includes the proportion of missing data, the number of data collection points in each missing data segment, and the completion time for each missing data segment. Based on this data, a data acquisition frequency adjustment coefficient is obtained, thereby adjusting the power system data acquisition frequency to achieve dynamic optimization of the acquisition frequency.
[0057] The above technical solutions achieve the following effects: Improved accuracy of missing data completion. The DTW distance-based completion method effectively captures the temporal correlation and periodicity characteristics of power system data, especially in scenarios with long missing sequences, where the completion effect is superior to traditional methods. Enhanced the targeting and accuracy of error correction. Utilizing the error model of power measurement instruments and fully combining the inherent error characteristics of the instruments for correction, the unsatisfactory correction effect of traditional methods is improved. Dynamic adjustment of data acquisition frequency is achieved. The acquisition frequency is adjusted in real time based on the operational data from missing data completion, avoiding resource waste caused by excessively high acquisition frequencies and the data loss problem exacerbated by excessively low frequencies, thus improving the rationality and efficiency of data acquisition. Overall, the integrity, accuracy, and reliability of power system data are improved, providing high-quality data support for power network monitoring, dispatch optimization, fault early warning, and equipment maintenance.
[0058] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A power system data repair method based on data completion and error correction, characterized in that, The power system data repair method includes: Real-time acquisition of power system data; The real-time power system data is detected to obtain missing data in the power system data summary, and the missing data is completed using a missing data completion method based on DTW distance to obtain the completed power system data. Error correction is performed on the completed power system data using a method based on data simulation completion and comparison with actual data, to obtain the corrected power system data; After the power system data is corrected, the operation data with missing data is collected in real time. The quality of the power system data collection is detected based on the operation data with missing data, and the collection frequency of the power system data is adjusted. Among them, the missing data completion method based on DTW distance includes: The optimal set of similar segments S is determined by using a preset initial distance threshold D0 and the similarity between the candidate sequence set S1 and the target sequence corresponding to the missing data; The real-time operating condition adaptation coefficient is used to dynamically configure the weights of each historical sequence contained in the optimal similar segment set S. Classify the missing data and complete the missing data according to the classification results; This includes classifying missing data and completing the missing data according to the classification results, including: Retrieve the number of data collection points contained in the target sequence corresponding to the missing data; compare the number of data collection points contained in the target sequence with a preset point count threshold, mark the missing data with the number of data collection points exceeding the preset point count threshold as long missing segments, and mark the missing data with the number of data collection points not exceeding the preset point count threshold as short missing segments; wherein, the preset point count threshold is set to 3; The missing data corresponding to the missing data of the long missing data segment and the short missing data segment are obtained by using the long missing data completion model and the short missing data completion model respectively, and the missing data of the long missing data segment and the short missing data segment are replaced by the completed data respectively. The structure of the long missing data completion model is as follows: Among them, X ct This represents the completed data corresponding to long missing data; n represents the total number of reference historical sequences contained in the best similar segment set S; Indicates the dynamic weight of the i-th reference history sequence; interp(Y) i ,t) represents the data value obtained by linear interpolation of the t-th acquisition point corresponding to the short missing data in the i-th reference history sequence; The structure of the short missing data completion model is as follows: Among them, X dt This represents the completed data corresponding to the short missing data; n represents the total number of reference historical sequences contained in the best similar fragment set S; Y represents the dynamic weight of the i-th reference history sequence; it This represents the historical data value of the t-th acquisition point in the i-th reference historical sequence that corresponds to the short missing data.
2. The power system data repair method based on data completion and error correction according to claim 1, characterized in that, Real-time acquisition of power system data, including: The power system data is collected in real time using a power big data edge acquisition system architecture, wherein the power system data includes power process data, environmental data and field status data; The power system data is denoised to obtain real-time denoised power system data; The power big data edge acquisition system architecture includes multiple edge nodes and an edge gateway module; the multiple edge nodes are connected to the edge gateway module; and the edge gateway module is connected to a missing data completion module. The multiple edge nodes are used to collect power system data in real time and send the power system data to the edge gateway module; The edge gateway module is used to convert and encrypt the received power system data, and then send the converted and encrypted power system data to the missing data completion module.
3. The power system data repair method based on data completion and error correction according to claim 1, characterized in that, The optimal set of similar segments S is determined using a preset initial distance threshold D0 and the similarity between the candidate sequence set S1 and the target sequence corresponding to the missing data, including: A preset initial distance threshold D0 is retrieved, wherein the preset initial distance threshold D0 is 60% of the average DTW distance between all historical sequences and the target sequence corresponding to the missing data; The DTW distance between each historical sequence and the target sequence is compared with a preset initial distance threshold D0, and historical sequences with a DTW distance lower than the preset initial distance threshold D0 are selected as candidate sequence set S1; The cosine similarity T between each historical sequence and the target sequence in the candidate sequence set S1 is obtained using the first-order difference method. A preset similarity threshold is retrieved, wherein the preset similarity threshold is 0.7; The cosine similarity T between each historical sequence in the candidate sequence set S1 and the target sequence is compared with a preset similarity threshold, and historical sequences with a cosine similarity T not lower than the preset similarity threshold are selected to form a sequence set S2. The positions of abrupt change between the target sequence and the historical sequences contained in the sequence set S2 are extracted using wavelet transform, and the number of matching positions of abrupt change between the target sequence and the historical sequences contained in the sequence set S2 is obtained. The ratio of the number of matching mutation point locations to the total number of mutation points in the historical sequences contained in the sequence composition set S2 is used to obtain the mutation point matching rate between the target sequence and the historical sequences contained in the sequence composition set S2. Historical sequences with a mutation point matching rate exceeding a preset mutation point matching rate threshold are selected to generate an optimal set of similar segments S, wherein the preset mutation point matching rate threshold is 0.
6.
4. The power system data repair method based on data completion and error correction according to claim 1, characterized in that, The dynamic weighting of each historical sequence contained in the optimal similar segment set S is configured using real-time operating condition adaptation coefficients, including: Each historical sequence contained in the optimal similar segment set S is retrieved as a reference historical sequence; Retrieve the DTW distance of each reference historical sequence, and use the reciprocal of the DTW distance of the reference historical sequence as the base weight value of each reference historical sequence; Retrieve the power system operating parameters at the corresponding time of each reference historical sequence and the power system operating parameters at the corresponding time of the target sequence, and obtain the Euclidean distance between the power system operating parameters at the corresponding time of each reference historical sequence and the power system operating parameters at the corresponding time of the target sequence. The dynamic weight of each reference historical sequence is obtained by combining the Euclidean distance between the power system operating parameters at the corresponding time of each reference historical sequence and the power system operating parameters of the target sequence with the basic weight value of each reference historical sequence.
5. The power system data repair method based on data completion and error correction according to claim 1, characterized in that, Error correction is performed on the completed power system data using a method based on data simulation completion and comparison with actual data to obtain corrected power system data, including: The real-time data call the time series Δt= (Δt1, Δt2, …, Δt m ) after the data completion, and the valid power system data corresponding to the missing data of the same type; wherein Δt1, Δt2, …, Δt m respectively represent 1 to m data collection time points contained in the time series; The missing data completion method based on DTW distance is used to simulate and complete the data according to the data collection time included in the time series Δt, and the simulated and completed data corresponding to time Δt after data completion is obtained. Based on the data completion, the simulated completed data and time series Δt = (Δt1, Δt2, ..., Δt) corresponding to time Δt after data completion are given. m The corresponding error correction item for the actual collected valid power system data; The error correction term is used to correct the error in the completed data corresponding to the missing data, and the error-corrected completed data is obtained. The error-corrected completed data is the corrected power system data.
6. The power system data repair method based on data completion and error correction according to claim 1, characterized in that, After the power system data is corrected, operational data with missing data is collected in real time to fill in the gaps, and the frequency of power system data collection is adjusted, including: After the power system data is corrected, the operation data during the missing data completion process is collected in real time. The operation data during the missing data completion process includes the proportion of missing data, the number of data collection points contained in each missing data segment, and the completion time of each missing data segment. The proportion of missing data, the number of data collection points contained in each missing data segment, and the time for completing each missing data segment are normalized to obtain the normalized proportion of missing data, the number of data collection points contained in each missing data segment, and the time for completing each missing data segment. The data collection frequency adjustment coefficient is obtained by using the proportion of missing data after normalization, the number of data collection points contained in each missing data segment, and the time for completing each missing data segment. The data acquisition frequency adjustment coefficient is used to adjust the acquisition frequency of the power system data to obtain the adjusted acquisition frequency.
7. The power system data repair method based on data completion and error correction according to claim 1, characterized in that, The quality of power system data acquisition is tested based on the operational data completed from the missing data, including: The percentage of missing data and the number of data collection points contained in each segment of missing data are retrieved from the running data during the missing data completion process. The percentage of missing data is compared with a preset percentage threshold. When the proportion of missing data exceeds a preset proportion threshold, the number of data collection points contained in each segment of missing data is compared with a preset number of points threshold. Based on the comparison between the number of data collection points contained in each missing data segment and the preset number of points threshold, missing data segments with the number of data collection points exceeding the preset number of points threshold are selected as target missing data segments. Data quality parameters are obtained by combining the data volume corresponding to the target missing data segment, the total data volume of missing data, and the proportion of missing data. The data quality parameters are compared with preset data quality parameter thresholds. When the data quality parameters exceed the preset data quality parameter thresholds, the data acquisition quality is determined to be abnormal, and an alarm for abnormal acquisition quality is triggered.
8. A power system data repair system based on data completion and error correction, characterized in that, The power system data repair system includes: The data acquisition module is used to collect power system data in real time and obtain real-time power system data; The missing data completion module is used to detect the power system data, obtain the missing data in the power system data summary, and complete the missing data using the missing data completion method based on DTW distance to obtain the completed power system data. The error data correction module is used to perform error correction on the completed power system data by using a method based on data simulation completion and comparison with actual data, and to obtain the corrected power system data. The data acquisition quality monitoring module is used to acquire operational data with missing data completion in real time, detect the acquisition quality of power system data based on the operational data with missing data completion, and adjust the acquisition frequency of power system data. Among them, the missing data completion method based on DTW distance includes: The optimal set of similar segments S is determined by using a preset initial distance threshold D0 and the similarity between the candidate sequence set S1 and the target sequence corresponding to the missing data; The real-time operating condition adaptation coefficient is used to dynamically configure the weights of each historical sequence contained in the optimal similar segment set S. Classify the missing data and complete the missing data according to the classification results; This includes classifying missing data and completing the missing data according to the classification results, including: Retrieve the number of data collection points contained in the target sequence corresponding to the missing data; compare the number of data collection points contained in the target sequence with a preset point count threshold, mark the missing data with the number of data collection points exceeding the preset point count threshold as long missing segments, and mark the missing data with the number of data collection points not exceeding the preset point count threshold as short missing segments; wherein, the preset point count threshold is set to 3; The missing data corresponding to the missing data of the long missing data segment and the short missing data segment are obtained by using the long missing data completion model and the short missing data completion model respectively, and the missing data of the long missing data segment and the short missing data segment are replaced by the completed data respectively. The structure of the long missing data completion model is as follows: Among them, X ct This represents the completed data corresponding to long missing data; n represents the total number of reference historical sequences contained in the best similar segment set S; Indicates the dynamic weight of the i-th reference history sequence; interp(Y) i ,t) represents the data value obtained by linear interpolation of the t-th acquisition point corresponding to the short missing data in the i-th reference history sequence; The structure of the short missing data completion model is as follows: Among them, X dt This represents the completed data corresponding to the short missing data; n represents the total number of reference historical sequences contained in the best similar fragment set S; Y represents the dynamic weight of the i-th reference history sequence; it This represents the historical data value of the t-th acquisition point in the i-th reference historical sequence that corresponds to the short missing data.
Citation Information
Patent Citations
Power missing data complementation method
CN115511002A
