A method for formulating loss reduction measures based on synchronous line loss anomaly identification

Abstract

The present invention discloses a method for formulating loss reduction measures based on the identification of abnormal line loss in the same period, including identifying abnormal line loss data in the same period and taking corresponding loss reduction measures based on the identification results. The fluctuation characteristics of the abnormal line loss data in the same period are extracted through statistical analysis; the low-rank characteristics of the abnormal line loss data in the same period are extracted by constructing a line loss power matrix in the same period and performing singular value decomposition on the matrix; the same-direction correlation characteristics of the allocated line loss abnormal data in the same period are extracted by performing grey correlation analysis on the line loss allocation data in the same period; and the "rank sum is approximately equal" characteristics of the abnormal line loss data in the same period deviating from the normal data are obtained by theoretically calculating the line loss and analyzing the differences between the line losses in the same period. The present invention can effectively extract the characteristics of the abnormal line loss data in the same period and construct a corresponding neural network for training, and can realize the effective identification of the abnormal line loss data in the same period. Corresponding loss reduction measures are taken according to different identification results, which can effectively improve the loss reduction efficiency.

Description

Technical Field

[0001] The present invention relates to the problem of formulating loss reduction measures for power system transmission lines, and in particular to a method for formulating loss reduction measures based on identification of synchronous line loss anomalies, belonging to the technical field of power grid management and control. Background Art

[0002] With the large-scale integration of renewable energy into the power system, inter-provincial, regional, and national interconnections have become an inevitable development trend in order to improve the operational efficiency of the entire power industry and optimize resource allocation across society. It has become the norm to export renewable energy from the northwest to the central and eastern regions with high electricity demand. This large-scale power / electricity exchange has increased grid losses. At the same time, the drawbacks of traditional grid management models and improper line loss statistical assessment methods have also led to large line loss electricity levels during the same period.

[0003] Targeted loss reduction measures can effectively address the issue of high line losses during the same period. However, existing methods for analyzing loss reduction measures fail to directly address the underlying causes of high line losses. Most analyze only technical line losses, failing to consider subjective factors such as actual mismanagement. This can lead to inconsistent results. To address this issue, it's necessary to perform anomaly diagnosis on the line loss data during the same period to determine whether any abnormalities are due to mismanagement or component failure, allowing for the implementation of targeted loss reduction measures.

[0004] Theoretical research on anomaly diagnosis has a solid theoretical foundation both domestically and internationally. Artificial neural networks (ANNs) were among the first to be applied to multivariate process monitoring and anomaly diagnosis, becoming an important method for solving process diagnosis problems. While ANNs offer certain advantages in anomaly identification, their poor generalization and the need for expert experience in network construction present challenges for their application. Kernel functions and support vector machines (SVMs) are intelligent methods for nonlinear classification in anomaly diagnosis that have been developed in recent years. While kernel methods have been studied in process diagnosis, their practical application is directly related to the specific form of the kernel function and the values of the corresponding parameters. Different parameter values often significantly affect the effectiveness of the model. Therefore, optimizing the parameters of kernel methods in process diagnosis models to achieve model optimization is a topic worthy of further study. Currently, models for diagnosing abnormal line losses at home and abroad are primarily based on low-voltage substations. Some literature proposes a front-end and back-end service separation mechanism for power supply companies, which manages the distribution of power load characteristics based on classification. It also implements hierarchical management of online line losses, monitors daily changes in line losses at each substation level, and promptly issues alarms for substations with excessive fluctuations. Other scholars have proposed a low-voltage substation line loss diagnosis model based on support vector machines. Building on traditional substation line loss management, this model, combined with the application of various information systems and the advancement of line loss refinement, establishes a scientific diagnosis model for abnormal substation line loss rates through the application of support vector machines. However, due to the limitations of substation samples and the complexity and diversity of field sites, the prediction accuracy of this model is still far from sufficient to achieve professional analysis and judgment of the causes of abnormal substation line loss rates.

[0005] Therefore, although there is considerable research on data anomaly diagnosis methods at home and abroad, there are few reports on using them to diagnose line loss data anomalies in combination with the inherent characteristics of line loss in the same period. Summary of the Invention

[0006] The main technical problem solved by the present invention is the mismatch between the loss reduction measures for the line loss of the same period and the influencing factors of the actual higher line loss. A method for formulating loss reduction measures based on the identification of abnormalities in the line loss of the same period is provided, which can identify abnormalities in the line loss data of the same period and formulate corresponding technical or management loss reduction measures based on the identification results, and can effectively solve the current unreasonable loss reduction measures.

[0007] In order to solve the above technical problems, a technical solution adopted by the present invention is:

[0008] Step 1: Determine the area where loss reduction is required, and select training samples for neural network training from the massive historical line loss electricity data of the same period in this area. The training samples include two types: those with known anomalies and those with unknown anomalies. The time scale of the samples is consistent, and daily line loss electricity values are usually selected.

[0009] Step 2: Calculate the characteristic index values of the selected line loss sample data of the same period. First, calculate the volatility characteristic index of the sample data. Volatility characteristic indicator calculation formula is the line loss at the next moment, is the line loss at the previous moment; then calculate the singular value characteristic index R l , we need to construct the line loss power matrix P first, Among them, P ij (i=1,2,…,m;j=1,2,…,T) is the daily line power loss of the jth day of the i-th month. The singular value matrix is obtained by performing singular value decomposition on the matrix The set of singular values that are much larger than the rest of the singular values is D h ,N(D h ) is the set capacity of the singular values that are greater than the rest of the singular values. The calculation formula of the singular value characteristic index is: That is, the proportion of high-rank singular values; then we can calculate the same-direction correlation index γ based on the apportioned line loss data i ,According to the grey correlation analysis of the apportioned line loss data, the grey correlation between the apportioned line loss data can be obtained. If there is an anomaly in the line loss data of the same period, the apportioned line loss data will show the same change characteristics. The calculation formula is ε i (k) is the grey correlation degree of the apportioned data; the rank and approximate equality characteristic index D(R + ), by comparing the line loss data of the same period with the theoretical line loss data, a difference sequence can be constructed. According to the rank sum approximate equality characteristic, the difference between the line loss power of the same period and the theoretical line loss power should be randomly and dispersedly distributed in the neighborhood of zero, so the rank value of the difference sequence should be evenly distributed between [-n, +n]. The positive "rank sum" R + With negative "rank sum" |R - The absolute value of | is approximately equal. The formula for calculating the rank sum characteristic is D(R + )=1 / |R + -n(n+1) / 4|;S j is the jth singular value of the singular value matrix, S i is the i-th singular value of the singular value matrix, i≠j.

[0010] Step 3: Construct a radial basis function neural network based on semi-supervised learning. In the traditional radial basis function neural network, the idea of semi-supervised learning is added to establish a semi-supervised radial basis function (SSL-RBF) neural network. That is, a large number of unlabeled training samples are used to improve the training effect of the radial basis function neural network and improve its accuracy. The network uses four characteristic index values as input variables and whether the line loss data is abnormal as two output variables; clustering iteration calculates the network cluster center c j , use the negative gradient descent method to calculate the width σ i and weight ω ji .

[0011] Step 4: Use training samples, that is, the line loss sample data of the same period in the area to be reduced, to train the network. The training data includes a small amount of normal data with known characteristic values (1 model sample), a small amount of abnormal data with known characteristic values (2 model sample), and a large amount of unknown abnormal data with known characteristic values (3 model sample). After training, a neural network for identifying abnormal line loss data of the same period can be obtained.

[0012] Step 5: Calculate the characteristic index values of the suspected abnormal line loss data of the same period in the area to be reduced, and use the neural network for identification and diagnosis to determine whether there is indeed abnormal line loss data caused by faults or improper management.

[0013] Step 6: Based on the identification results, if there are abnormal data for line losses during the same period, it indicates a fault or incorrect statistical management, and appropriate management and loss reduction measures need to be taken based on the location of the abnormal data. If there are no abnormal data, it means that the high line loss is due to inherent factors of the grid lines or transformers, and appropriate technical loss reduction measures need to be taken at the location with high line losses.

[0014] Step 7: Based on actual conditions, technical line loss reduction measures include but are not limited to reasonable configuration of reactive power compensation to reduce line reactive power, adjustment of the actual operating voltage of the power grid, addition of parallel lines to 330kV heavy-load lines, optimization of the power grid structure, and line diameter expansion and reconstruction; management line loss reduction measures include increasing the penetration rate and accuracy of metering devices, regularly inspecting metering devices, promptly replacing damaged equipment, increasing data repair methods, promptly repairing abnormal data, further strengthening business census work, improving the anti-electricity theft mechanism, basically eliminating the occurrence of electricity theft, and improving the line loss assessment and performance evaluation management system.

[0015] The present invention proposes a method for formulating loss reduction measures based on the identification of concurrent line loss anomalies, and adopts corresponding loss reduction measures according to the diagnosis results of concurrent line loss anomaly data, which has guiding significance for management decision-making departments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a specific flow chart for formulating loss reduction measures for identifying line loss anomalies provided by the present invention;

[0017] Figure 2 This is a schematic diagram of the semi-supervised radial basis function neural network training of the present invention. DETAILED DESCRIPTION

[0018] The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more precise definition of the protection scope of the present invention.

[0019] This paper proposes a method for formulating loss reduction measures based on the identification of concurrent line loss anomalies. A high line loss area in Gansu was selected for simulation verification. The specific technical implementation plan is as follows:

[0020] 1. Select historical sample data for the region. In this example, 200 sample data sets are selected, including 10 sets of normal line loss data, 10 sets of abnormal line loss data, and 180 sets of unknown abnormal data.

[0021] 2. According to the calculation method of step 2 of the technical solution proposed in the invention content, the characteristic index values of the selected samples are calculated respectively, and the normal data with known characteristic values (1 model sample), the abnormal data with known characteristic values (2 model samples), and the unknown abnormal data with known characteristic values (3 model samples) are obtained as shown in the following table:

[0022] (1) 1 model sample data

[0023] Table 11 Input values and marking values of model samples

[0024]

[0025] (2) 2 Model Sample Data

[0026] Table 22 Input values and marking values of model samples

[0027]

[0028]

[0029] (3) 3 Model Sample Data

[0030] The method for obtaining the 3-type sample is simple. Calculate the line loss anomaly characteristic index values of 180 samples in the same period, but do not verify whether these samples are in the line loss anomaly state in the same period. These are the 3-type samples.

[0031] 3. As attached Figure 2As shown in the figure, a radial basis function neural network for semi-supervised learning is constructed. 20 labeled samples (including 10 samples of model 1 and 10 samples of model 2) and 180 unlabeled samples are used. The learning efficiency of width and weight is set to η = β = 0.1. The semi-supervised radial basis function neural network for distinguishing line loss anomalies in the same period is trained.

[0032] Table 3. Overall situation of training samples

[0033]

[0034] The cluster center, width and weight parameter results obtained through training are shown in Table 4. There are two types of line loss abnormality and non-abnormality in the output layer, corresponding to weights ω i1 and weight ω i2 .

[0035] Table 4 Cluster center, width and weight parameter results

[0036]

[0037]

[0038] 4. After training is complete, the system can quickly identify abnormalities in line loss during the same period. The characteristic indicator values of the line loss data to be identified in the high-line-loss area are calculated. The results are: a fluctuation eigenvalue of 0.347, a singular value eigenvalue of 0.725, a positive "rank sum" eigenvalue of 13.287, and a unidirectional correlation eigenvalue of 0.256. These four eigenvalues are used as input to the trained neural network, and the output indicates that the data is normal, indicating that the line loss and electricity data are normal.

[0039] 6. By identifying anomalies in line loss data for the same period, we concluded that no abnormal data existed. Therefore, the high line loss in this area is due to technical reasons, not data anomalies caused by mismanagement or faults. Therefore, specific technical loss reduction measures can be implemented based on the actual situation. Analysis of actual lines in areas with high line losses reveals that due to the long construction history of these lines and the recent increase in power load, overload issues may occur during peak operating conditions. Line expansion and reconstruction can be implemented to reduce the high line losses caused by overload.

[0040] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structures or equivalent process changes made using the contents of the present invention's description and drawings, or directly or indirectly applied to other related technical fields, are also included in the patent protection of the present invention.

Claims

1. A method for formulating loss reduction measures based on identification of concurrent line loss anomalies, characterized in that: The method comprises the following steps: Step 1: Select training samples from massive contemporaneous line loss data, including two types: those with known abnormalities and those with unknown abnormalities; Step 2: Calculate the characteristic index values of the selected samples, including the volatility characteristic index Singular value characteristic index R l , same-direction correlation index γ i , rank and approximate equality characteristic index D(R + ); Step 3: Construct a radial basis function neural network with four characteristic index values as input variables and abnormality of line loss data as two output variables; calculate the network cluster center c j , width σ i and weight ω ji ; Step 4: Use the samples to train the network and obtain a neural network for identifying abnormal line loss data during the same period; Step 5: Calculate the characteristic index value of the line loss data to be identified during the same period and use the neural network for identification and diagnosis; Step 6: Based on the identification results, if there are abnormal data for line losses during the same period, appropriate management loss reduction measures are implemented based on the location of the abnormal data. If there are no abnormal data, appropriate technical loss reduction measures are implemented at locations with higher line losses. The known or unknown abnormal samples in step 1 are obtained through state estimation calculation; the volatility characteristic index in step 2 is the line loss at the next moment, is the line loss at the previous moment; singular value characteristic index Same-direction correlation index ε i (k) is the grey correlation degree of line loss allocation data; the rank sum approximate equality characteristic index D(R + )=1 / |R + -n(n+1) / 4|,R + is the positive rank sum, n is the sample size; the set of singular values larger than the rest of the singular values is D h ,N(D h ) is the set capacity of the singular values that are larger than the rest of the singular values; S j is the jth singular value of the line loss singular value matrix, S i is the i-th singular value of the line loss power singular value matrix.

2. The method for formulating loss reduction measures based on identification of concurrent line loss anomalies according to claim 1, characterized in that: The network training using samples described in step 4 is a semi-supervised training method, that is, the training samples include some normal labeled samples with known characteristic indicators, some abnormal labeled samples with known characteristic indicators, and a large number of unlabeled samples with only known characteristic indicators.

3. The method for formulating loss reduction measures based on identification of concurrent line loss anomalies according to claim 1, characterized in that: The line loss data to be identified for the same period in step 5 is data indicating high line loss and requiring loss reduction measures, but it is unknown whether there are any abnormalities. The technical loss reduction measures in step 6 include reasonably configuring reactive power compensation to reduce line reactive power, adjusting the actual operating voltage of the power grid, adding parallel lines, optimizing the power grid structure, and expanding and reconstructing lines. The management line loss reduction measures include increasing the penetration rate and accuracy of metering devices, regularly inspecting metering devices, promptly replacing damaged equipment, improving the anti-electricity theft mechanism, improving the line loss assessment indicator system, and improving the line loss assessment performance assessment management system.

Images