CN105184661A - Grid monitoring signal analysis method based on weighted Mahalanobis distance discrimination - Google Patents

Abstract

The present invention discloses a grid monitoring signal analysis method based on weighted Mahalanobis distance discrimination. Through the classification and statistics of grid dispatching automation system monitoring signals such as accident trip, alarm signal and measuring imbalance data, system to area monitoring signals can be divided into the accident, abnormality, out of limit, displacement and notification. Then different weight values are given to monitoring indexes, the emergency category to which each area monitoring signal belongs is judged through a weighted Mahalanobis distance, and classification processing is adopted for the emergency categories in different areas. The method is accurate, convenient and effective.

Description

Translated fromChinese

基于加权马氏距离判别的电网监控信号分析方法Analysis Method of Power Grid Monitoring Signal Based on Weighted Mahalanobis Distance Discrimination

技术领域technical field

本发明涉及电网调度自动化安全领域，具体涉及一个基于加权马氏距离判别的电网监控信号分析方法。The invention relates to the field of power grid scheduling automation security, in particular to a power grid monitoring signal analysis method based on weighted Mahalanobis distance discrimination.

背景技术Background technique

随着电网规模的不断扩大，电网结构日趋复杂,系统运行方式快速多变,运行管理面临着更大的挑战。而电网监控信息种类繁多、数据量庞大，通过科学的方法监测电网监控信息，归类分析发现电网运行规律以快速做出决策则显得十分必要。With the continuous expansion of the scale of the power grid, the structure of the power grid is becoming more and more complex, the operation mode of the system is changing rapidly, and the operation management is facing greater challenges. However, there are many types of power grid monitoring information and a huge amount of data. It is very necessary to monitor power grid monitoring information through scientific methods, classify and analyze to find out the operation rules of the power grid to make decisions quickly.

监控业务是在“三集五大”调整之后并入调控中心业务中的，之前各地区关于监控信号数据的处理与分析还处于探索阶段。对监控数据的处理分析也主要基于对收集上来的信息进行汇总统计，缺乏全面、系统的分析手段对信息进行归类整理。目前电网调度自动化系统的监控信号数据种类繁多、数据量庞大，数据层次低，缺乏有效的分析手段，使得管理电网的工作消耗很大的人力物力，效率却很低。The monitoring business was merged into the control center business after the adjustment of the "Three Sets and Five Majors". Previously, the processing and analysis of monitoring signal data in various regions was still in the exploratory stage. The processing and analysis of monitoring data is also mainly based on the summary statistics of the collected information, and there is a lack of comprehensive and systematic analysis methods to classify the information. At present, the monitoring signal data of the power grid dispatching automation system is various, the data volume is huge, the data level is low, and there is no effective analysis method, which makes the work of managing the power grid consume a lot of manpower and material resources, but the efficiency is very low.

电网监控信号分析方法是从繁杂的监控跳闸、告警以及不平衡性等数据中归纳、抽取关键监测指标数据，并进行分析、归类、总结，以帮助监控人员总结发现历史规律，辅助决策，从而减轻其繁重的统计分析工作，有效提高电网监控部门的监测分析能力。The power grid monitoring signal analysis method is to summarize and extract key monitoring index data from the complicated data such as monitoring trips, alarms, and imbalances, and analyze, classify, and summarize to help monitoring personnel summarize and discover historical laws and assist decision-making. Reduce its heavy statistical analysis work and effectively improve the monitoring and analysis capabilities of the power grid monitoring department.

发明内容Contents of the invention

本发明的目的在于提供一种方便、有效的基于加权马氏距离判别的电网监控信号分析方法。The purpose of the present invention is to provide a convenient and effective grid monitoring signal analysis method based on weighted Mahalanobis distance discrimination.

本发明的技术解决方案是：Technical solution of the present invention is:

一种基于加权马氏距离判别的电网监控信号分析方法，其特征是：包括下列步骤：A power grid monitoring signal analysis method based on weighted Mahalanobis distance discrimination is characterized in that it includes the following steps:

(1)数据结构设计(1) Data structure design

将每个地区的监控信息抽象成一个m维的元组，表示为X＝｛x₁，x₂，...，x_m｝^T，m为监控信息的维数，本例中m＝5，分别代表类别为事故、异常、越限、变位和告知；Abstract the monitoring information of each region into an m-dimensional tuple, expressed as X={x₁ , x₂ ,...,x_m }^T , m is the dimension of the monitoring information, in this example m=5 , which represent the categories of accident, abnormality, limit violation, displacement and notification respectively;

(2)加权马氏距离(2) Weighted Mahalanobis distance

设G＝｛X₁，X₂，...，X_m}^T为m元总体(考察m个监控信号指标)，样本X＝｛x₁，x₂，...，x_m}^T；令μ_i＝E(X_i)(i＝1，2，...，m)，则总体均值向量μ＝{μ₁，μ₂，...，μ_m}^T；总体G的协方差矩阵为Let G={X₁ , X₂ ,...,X_m }^T be an m-element population (inspecting m monitoring signal indicators), sample X={x₁ , x₂ ,..., x_m }^T ; Let μ_i =E(X_i )(i=1, 2, ..., m), then the population mean vector μ = {μ₁ , μ₂ , ..., μ_m }^T ; the covariance of the population G The matrix is

Σ＝cov(G)＝E［(G－μ)(G－μ)^T]Σ＝cov(G)＝E[(G－μ)(G－μ)^T ]

样本X与总体G的马氏距离平方为The square of the Mahalanobis distance between the sample X and the population G is

α²(X，G)＝(X－μ)^TΣ^-1(X－μ)α² (X, G)＝(X－μ)^T Σ^-1 (X－μ)

然而各指标在判定样本X属于总体G所起的作用是不尽相同的，即其重要程度存在差异；因此在马氏距离的基础上加入指标的权重，以区分各指标的重要性，即权值越大，指标越重要；得到加权马氏距离平方为：However, each indicator plays a different role in judging that the sample X belongs to the overall G, that is, its importance is different; therefore, the weight of the indicator is added on the basis of the Mahalanobis distance to distinguish the importance of each indicator, that is, the weight The larger the value, the more important the indicator; the square of the weighted Mahalanobis distance is obtained as:

α²(X，G)＝(X－μ)^ＴWΣ^－1W(X-μ)α² (X, G)＝(X－μ)^T WΣ^－1 W(X-μ)

其中W＝diag(w₁，w₂，...，w_m)，式中：W为对角矩阵；w_i∈[0，1](i＝1，2，...，m)，对应于各指标在距离函数中的权重，Σw_i＝1；Wherein W=diag(w₁ ,w₂ ,...,w_m ), where: W is a diagonal matrix; w_i ∈[0,1](i=1,2,...,m), Corresponding to the weight of each index in the distance function, Σw_i =1;

(3)利用加权马氏距离判别法对各地区的监控信息进行分类，具体步骤如下：(3) Use the weighted Mahalanobis distance discriminant method to classify the monitoring information in each region. The specific steps are as follows:

步骤1：将所有地区信息分为k个m元总体，即G₁，G₂，...，G_k（k＞2）；Step 1: Divide all regional information into k m-element populations, namely G₁ , G₂ , ..., G_k (k>2);

步骤2：计算每个总体的均值向量和协方差矩阵分别μ_i，Σ_i(i＝1，2，...，k)；Step 2: Calculate the mean vector and covariance matrix of each population respectively μ_i , Σ_i (i=1, 2, ..., k);

步骤3：给m个指标赋以权值，得到W＝diag(w₁，ｗ₂，...，w_m)；Step 3: assign weights to m indicators, and get W=diag(w₁ , w₂ ,..., w_m );

步骤4：分别计算样本X到k个总体的加权马氏距离平方为α²(X，G_i)(i＝1，2，...，k)；Step 4: Calculate the squared weighted Mahalanobis distance from sample X to k populations as α² (X, G_i ) (i=1, 2, ..., k);

步骤5：利用判别准则：若i＝l，Step 5: Utilize the discriminant criterion: if i=l,

α²(X，G₁)＝min{α²(X，G_i)}(i＝1，2，...，k)，则样本X∈G₁，将样本X归入相应的总体；α² (X, G₁ )=min{α² (X, G_i )}(i=1, 2, ..., k), then sample X∈G₁ , classify sample X into the corresponding population;

(4)通过上述方法将各地区监控信号分别归入相应的总体，然后结合历史情况和专业经验对每个总体的监控信号给出不同的处理方案；k＝4时，将监控信息分为4类，针对不同的类别给出如下解决方案：(4) Classify the monitoring signals of each region into corresponding populations by the above method, and then provide different processing schemes for the monitoring signals of each population in combination with historical conditions and professional experience; when k=4, divide the monitoring information into 4 Classes, the following solutions are given for different categories:

a)G₄总体：延时处理，对相应地区的监控信号观察一段时间，若恢复正常，则返回继续监测；否则调出声像信息，做进一步的分析；a)_G4 overall: Delay processing, observe the monitoring signal in the corresponding area for a period of time, if it returns to normal, then return to continue monitoring; otherwise call out the audio and video information for further analysis;

b)G₃总体：调出声像信息，分析监控信号是否会对电网的正常运行造成影响；若会造成影响，则派出人员对相应地区的线路及设备进行检修；否则返回继续监测；b) G₃ Overall: call out the audio-visual information, and analyze whether the monitoring signal will affect the normal operation of the power grid; if it will affect, send personnel to overhaul the lines and equipment in the corresponding area; otherwise, return to continue monitoring;

c)G₂总体：立即派出人员对相应地区的线路及设备进行检修，及时排除隐患c) G₂ Overall: Immediately send personnel to overhaul the lines and equipment in the corresponding area, and eliminate hidden dangers in time

d)G₁总体：立即派出人员进行检修的同时，必须向上级汇报，调集更多的技术骨干，群策群力，尽快解决问题。d) G₁ Overall: Immediately send personnel to carry out maintenance, and must report to the superior, mobilize more technical backbones, and work together to solve the problem as soon as possible.

本发明方便、有效、准确。通过对电网监控信号如事故跳闸、告警信号以及量测不平衡等数据进行归类和统计，得到的监控信号主要可以分为事故、异常、越限、变位和告知。然后给各监控指标赋以不同的权值，通过加权马氏距离判别各地区监控信号属于的紧急类别，针对不同地区的紧急类别，采取分类处理。The invention is convenient, effective and accurate. By classifying and counting the data of power grid monitoring signals such as accident trips, alarm signals, and measurement imbalances, the obtained monitoring signals can be mainly divided into accidents, abnormalities, limit violations, displacements, and notifications. Then assign different weights to each monitoring index, and use the weighted Mahalanobis distance to determine the emergency category of the monitoring signal in each region, and adopt classification processing for the emergency category in different regions.

附图说明Description of drawings

下面结合附图和实施例对本发明作进一步说明。The present invention will be further described below in conjunction with drawings and embodiments.

图1是监控信号分析流程图。Figure 1 is a flow chart of monitoring signal analysis.

具体实施方式Detailed ways

如图1所示，通过对事故、异常、越限、变位和告知5类监控信息的收集，然后将每个地区的监控信息传至输入层，继而通过输入层将信息交给判别层进行处理，最后判别层运用加权马氏距离判别法对样本进行判定，并通过输出层将分类结果输出。As shown in Figure 1, through the collection of five types of monitoring information: accident, abnormality, limit violation, displacement and notification, the monitoring information of each region is transmitted to the input layer, and then the information is passed to the discrimination layer through the input layer for further processing. Finally, the discriminant layer uses the weighted Mahalanobis distance discriminant method to judge the samples, and outputs the classification results through the output layer.

对电网调度自动化系统的事故跳闸、告警信号以及量测不平衡数据进行归类和统计，为该方法提供有效的数据支持。具体内容为：Classify and count the accident trips, alarm signals and measurement unbalance data of the power grid dispatching automation system to provide effective data support for this method. The specific content is:

(1)事故跳闸统计,例如按地区、电压等级统计线路跳闸次数。(1) Statistics on accident trips, such as counting the number of line trips by region and voltage level.

(2)告警分析，例如按维护责任区统计告警信息情况。(2) Alarm analysis, such as statistics of alarm information by maintenance responsibility area.

(3)量测不平衡性统计，例如按厂站统计，点击可查看某厂站每天的不平衡情况。(3) Measurement imbalance statistics, such as statistics by plant, click to view the daily imbalance of a certain plant.

3.加权马氏距离判别方法3. Weighted Mahalanobis distance discriminant method

(1)数据结构设计(1) Data structure design

将每个地区的监控信息抽象成一个m维的元组，表示为X＝{x₁，x₂，...，x_m}^T，m为监控信息的维数，本例中m＝5，分别代表监控信号类别为事故、异常、越限、变位和告知。Abstract the monitoring information of each region into an m-dimensional tuple, expressed as X={x₁ , x₂ ,...,x_m }^T , m is the dimension of the monitoring information, in this example m=5 , which respectively represent the monitoring signal categories as accident, abnormality, limit violation, displacement and notification.

(2)加权马氏距离(2) Weighted Mahalanobis distance

设G＝{X₁，X₂，...，X_m}^T为m元总体(考察m个监控指标)，样本X＝{x₁，x₂，...，x_m}^T。令μ_i＝E(X_i)(i＝1，2，...，m)，则总体均值向量μ＝{μ₁，μ₂，...，μ_m}^T。总体G的协方差矩阵为Let G={X₁ , X₂ ,...,X_m }^T be an m-element population (investigate m monitoring indicators), sample X={x₁ , x₂ ,..., x_m }^T . Let μ_i =E(X_i )(i=1, 2, . . . , m), then the population mean vector μ={μ₁ , μ₂ , . . . , μ_m }^T . The covariance matrix of the population G is

Σ＝cov(G)＝E[(G-μ)(G-μ)^T]Σ=cov(G)=E[(G-μ)(G-μ)^T ]

样本X与总体G的马氏距离平方为The square of the Mahalanobis distance between the sample X and the population G is

α²(X，G)＝(X-μ)^TΣ^-1(X-μ)α² (X, G) = (X-μ)^T Σ^-1 (X-μ)

然而各指标在判定样本X属于总体G所起的作用是不尽相同的，即其重要程度存在差异。因此在马氏距离的基础上加入指标的权重，以区分各指标的重要性，即权值越大，指标越重要。得到加权马氏距离平方为：However, each indicator plays a different role in judging that the sample X belongs to the population G, that is, there are differences in their importance. Therefore, the weight of the index is added on the basis of the Mahalanobis distance to distinguish the importance of each index, that is, the larger the weight, the more important the index. The squared weighted Mahalanobis distance is obtained as:

α²(X，G)＝(X-μ)^TWΣ^-1W(X-μ)α² (X, G)＝(X-μ)^T WΣ^-1 W(X-μ)

其中W＝diag(w₁，w₂，...，w_m)，式中：W为对角矩阵；w_i∈[0，1](i＝1，2，...，m)，对应于各指标在距离函数中的权重，Σw_i＝1。Wherein W=diag(w₁ ,w₂ ,...,w_m ), where: W is a diagonal matrix; w_i ∈[0,1](i=1,2,...,m), Corresponding to the weight of each indicator in the distance function, Σw_i =1.

例如根据各属性指标对判别结果影响程度，给事故、异常、越限、变位、告知5个属性指标赋以权值w_i(i＝1，2，3，4，5)，其中w₁＝0.35，w₂＝0.25，w₃＝0.2，w₄＝0.15，w₅＝0.05，则有For example, according to the degree of influence of each attribute index on the judgment result, assign weights w_i (i=1, 2, 3, 4, 5) to the five attribute indexes of accident, abnormality, overrun, displacement, and notification, where w₁ =0.35, w₂ =0.25, w₃ =0.2, w₄ =0.15, w₅ =0.05, then we have

${<span><span>\&Sigma;</span>\&Sigma;</span>}_{\mathrm{<span><span>i</span>i</span>}<span><span>=</span>=</span>\mathrm{<span><span>1</span>1</span>}}^{\mathrm{<span><span>5</span>5</span>}}{\mathrm{<span><span>w</span>w</span>}}_{\mathrm{<span><span>i</span>i</span>}}<span><span>=</span>=</span>\mathrm{<span><span>1</span>1</span>}<span><span>(</span>(</span>\mathrm{<span><span>0</span>0</span>}<span><span>\&le;</span>\&le;</span>{\mathrm{<span><span>w</span>w</span>}}_{\mathrm{<span><span>i</span>i</span>}}<span><span>\&le;</span>\&le;</span>\mathrm{<span><span>1</span>1</span>}<span><span>)</span>)</span><span><span>,</span>,</span>\mathrm{<span><span>W</span>W</span>}<span><span>=</span>=</span>\mathrm{<span><span>d</span>d</span>}\mathrm{<span><span>i</span>i</span>}\mathrm{<span><span>a</span>a</span>}\mathrm{<span><span>g</span>g</span>}<span><span>(</span>(</span>\mathrm{<span><span>0.35</span>0.35</span>}<span><span>,</span>,</span>\mathrm{<span><span>0.25</span>0.25</span>}<span><span>,</span>,</span>\mathrm{<span><span>0.2</span>0.2</span>}<span><span>,</span>,</span>\mathrm{<span><span>0.15</span>0.15</span>}<span><span>,</span>,</span>\mathrm{<span><span>0.05</span>0.05</span>}<span><span>)</span>)</span><span><span>.</span>.</span>$

(3)利用加权马氏距离判别法对各地区的监控信息进行分类，具体步骤如下：(3) Use the weighted Mahalanobis distance discriminant method to classify the monitoring information in each region. The specific steps are as follows:

步骤1：将所有地区信息分为k个m元总体，即G₁，G₂，...，G_k(k＞2)。Step 1: Divide all regional information into k m-ary populations, namely G₁ , G₂ , . . . , G_k (k>2).

步骤2：计算每个总体的均值向量和协方差矩阵分别μ_i，Σ_i(i＝1，2，...，k)。Step 2: Calculate the mean vector and covariance matrix of each population respectively μ_i , Σ_i (i=1, 2, . . . , k).

步骤3:给m个指标赋以权值，得到W＝diag(w₁，w₂，...，w_m)。Step 3: Assign weights to the m indicators to obtain W=diag(w₁ , w₂ , . . . , w_m ).

步骤4：分别计算样本X到k个总体的加权马氏距离的平方为α²（X，G_i）（i＝1，2，...，k)Step 4: Calculate the square of the weighted Mahalanobis distance from sample X to k populations as α² (X, G_i ) (i=1, 2, ..., k)

步骤5：利用判别准则：若ｉ＝l，Step 5: Use the discriminant criterion: if i=l,

α²(X，G_i)＝min{α²(X，G_i)｝(i＝1，2，...，k)，则样本X∈G_i，将样本X归入相应的总体。α² (X, G_i )=min{α² (X, G_i )}(i=1, 2, ..., k), then sample X∈G_i , classify sample X into the corresponding population.

(4)通过上述方法将各地区监控信号分别归入相应的总体，然后结合历史情况和专业经验对每个总体的监控信号给出不同的处理方案。例如k＝4时，即可将监控信息分为4类，针对不同的类别给出如下解决方案：(4) Classify the monitoring signals of each region into corresponding populations through the above method, and then provide different processing schemes for the monitoring signals of each population in combination with historical conditions and professional experience. For example, when k=4, the monitoring information can be divided into 4 categories, and the following solutions are given for different categories:

a)G₄总体：延时处理，对相应地区的监控信号观察一段时间，若恢复正常，则返回继续监测；否则调出声像信息，做进一步的分析。a) G₄ Overall: Delay processing, observe the monitoring signal in the corresponding area for a period of time, if it returns to normal, return to continue monitoring; otherwise, call out the audio and video information for further analysis.

b)G₃总体：调出声像信息，分析该监控信号是否会对电网的正常运行造成影响。若会造成影响，则派出人员对相应地区的线路及设备进行检修；否则返回继续监测。b) G₃ Overall: call out the audio and video information, and analyze whether the monitoring signal will affect the normal operation of the power grid. If it will cause an impact, send personnel to overhaul the lines and equipment in the corresponding area; otherwise, return to continue monitoring.

c)G₂总体：立即派出人员对相应地区的线路及设备进行检修，及时排除隐患。c) G₂ Overall: Immediately send personnel to overhaul the lines and equipment in the corresponding area, and eliminate hidden dangers in time.

d)G₁总体：立即派出人员进行检修的同时，必须向上级汇报，调集更多的技术骨干，群策群力，尽快解决问题。d) G₁ Overall: Immediately send personnel to carry out maintenance, and must report to the superior, mobilize more technical backbones, and work together to solve the problem as soon as possible.

Claims (1)

1., based on the power system monitor signal analysis method that weighted Mahalanobis distance method differentiates, it is characterized in that: comprise the following steps:

(1) Data Structure Design

The monitor message in each area is abstracted into the tuple of a m dimension, is expressed as X={x₁, x₂..., x_m}^t, m is the dimension of monitor message, m=5 in this example, represent respectively classification be accident, abnormal, out-of-limit, conjugate and inform;

(2) weighted Mahalanobis distance method

If G={X₁, X₂..., X_m}^tfor m unit overall (investigating m pilot signal index), sample X={x₁, x₂..., x_m}^t; Make μ_i=E (X_i) (i=1,2 ..., m), then population mean vector μ=(μ₁, μ₂..., μ_m}^t; The covariance matrix of overall G is

∑＝cov(G)＝E[(G-μ)(G-μ)^T]

The mahalanobis distance square of sample X and overall G is

a²(X，G)＝(X-μ)^T∑^-1(X-μ)

But each index is judging that sample X belongs to overall G role and is not quite similar, namely its significance level there are differences; Therefore on the basis of mahalanobis distance, add the weight of index, to distinguish the importance of each index, namely weights are larger, and index is more important; Obtaining weighted Mahalanobis distance method square is:

a²(X，G)＝(X-μ)^TW∑^-1W(X-μ)

Wherein W=diag (w₁, w₂..., w_m), in formula: W is diagonal matrix; w₁∈ [0,1] (i=1,2 ..., m), corresponding to the weight of each index in distance function, ∑ w₁=1;

(3) utilize the monitor message of weighted Mahalanobis distance method diagnostic method to each department to classify, concrete steps are as follows:

Step 1: all regions information be divided into k m unit overall, i.e. G₁, G₂..., G_k(k > 2);

Step 2: calculate each overall mean vector and covariance matrix μ respectively_i, ∑_i(i=1,2 ..., k);

Step 3: compose with weights to m index, obtain W=diag (w₁, w₂..., w_m);

Step 4: calculating sample X to a k overall weighted Mahalanobis distance method square is respectively a²(X, G_i) (i=1,2 ..., k);

Step 5: utilize criterion: if i=l,

A²(X, G_i)=min{a²(X, G_i) (i=1,2 ..., k), then sample X ∈ G_i, sample X is included into corresponding overall;

(4) each department pilot signal is included into respectively corresponding overall by said method, then in conjunction with account of the history and professional experiences, different processing schemes is provided to each overall pilot signal; During k=4, monitor message is divided into 4 classes, provides following solution for different classifications:

AG₄overall: delay process, a period of time is observed to the pilot signal of corresponding area, if recover normal, then returns and continue monitoring; Otherwise recall audio and video information, be further analyzed;

B) G₃overall: to recall audio and video information, whether analysis monitoring signal can impact the normal operation of electrical network; If can impact, then expatriate personnel overhaul the circuit of corresponding area and equipment; Otherwise return and continue monitoring;

C) G₂overall: expatriate personnel overhaul the circuit of corresponding area and equipment immediately, get rid of hidden danger in time

D) G₁overall: while expatriate personnel carry out overhauling immediately, must report by superior, assemble more technical backbone, pool our ideas and make concerted efforts, deal with problems as early as possible.