Power load system fault detection method based on sparse dynamic canonical correlation analysis

Abstract

The invention discloses a power load system fault detection method based on sparse dynamic canonical correlation analysis. The method comprises the following steps: firstly, acquiring historical operation data of a power load system in a normal state and normalizing the historical operation data as training data; establishing a sparse dynamic canonical correlation analysis model by using the training data, calculating a normal Hotelling statistic T2 according to the established model, and calculating a fault threshold J by using T2; during operation, daily operation data of the power load system are collected and normalized as daily data, and daily Hotelling statistics Tt2 are calculated by using the daily data; when Tt2gt; when J, judging that the power load system has a fault; when Tt2lt; and J, the power load system operates normally. Wherein the fault threshold value J is calculated through a kernel density estimation method. According to the method, fault analysis is performed on the power load system by using the sparse dynamic canonical correlation analysis method, so that the purposes of reducing the labor cost and improving the operation efficiency of the power load system are achieved.

Description

Technical Field

[0001] The invention relates to the field of power load analysis, and in particular to a power load system fault detection method based on sparse dynamic canonical correlation analysis. Background Art

[0002] The power load system is one of the infrastructures of modern society, which provides necessary electricity for residential, commercial and industrial use. With the development of the economy and the growth of population, the demand for electricity continues to rise, which makes the stability and reliability of the power load system essential. However, the power load system may encounter various faults during operation, such as equipment aging, natural disasters, operational errors, etc., which may lead to power supply interruptions and affect people's normal life and work.

[0003] In the power load system, fault detection technologies can ensure the stable operation of the power system. These technologies can monitor the operating status of the power load system in real time and detect and diagnose potential faults in a timely manner. By using advanced sensors, data analysis and machine learning algorithms, fault detection systems can predict and identify various fault modes, so as to take rapid response and reduce the impact of faults on the power load system.

[0004] Therefore, an accurate and stable method is needed to detect in real time whether a fault occurs in the power load system, so that the operator can take effective measures on the power load system in a timely manner. Summary of the invention

[0005] The present invention mainly provides a power load system fault detection method based on sparse dynamic canonical correlation analysis with high accuracy, good real-time performance, safety and reliability.

[0006] The present invention aims to solve the above technical problems mainly through the following technical solutions: a method for detecting faults in a power load system based on sparse dynamic canonical correlation analysis, which is based on a sparse dynamic canonical correlation analysis algorithm, and automatically detects abnormal conditions in a power load system by mining the hidden intrinsic relationships between power load data (such as power generation, real-time electricity prices, temperature, humidity, power consumption, voltage, current, frequency, etc.), and comprises the following steps:

[0007] Step 1: Collect historical operating data of the power load system under normal conditions and normalize them as training data;

[0008] Step 2: Use the training data to establish a sparse dynamic canonical correlation analysis model, and calculate the normal Hotelling statistic T based on the established sparse dynamic canonical correlation analysis model. 2 ;

[0009] Step 3: Use T 2 Calculate the fault threshold J;

[0010] Step 4: Collect the daily operation data of the power load system and normalize it as daily data, and use the daily data to calculate the daily Hotelling statistic T t 2 ;

[0011] Step 5: Determine whether the power load system has a fault: When T t 2 > J, the power load system has a fault; when T t 2 < J, the power load system is operating normally.

[0012] Preferably, in step 2, the establishment of the sparse dynamic canonical correlation analysis model is specifically as follows:

[0013] u(t) ∈ R l is the normalized input variable, t is the serial number directly corresponding to time, y(t) ∈ R m is the normalized output variable, l is the number of input variables, m is the number of output variables, and the past input-output vector z p (t) is defined as:

[0014] z p (t) = [y T (t - p) … y T (t - 1) u T (t - p) … u T (t - 1)] T ,

[0015] Define the future input vector u f (t) as:

[0016] u f (t) = [u T (t) … u T (t + f)] T ;

[0017] Define the future output vector y f (t) as:

[0018] y f (t) = [y T (t) … y T (t + f)] T

[0019] p is the past time lag, and f is the future time lag;

[0020] Construct the input covariance matrix C zz as:

[0021]

[0022] Construct the output covariance matrix C yy for:

[0023]

[0024] Construct the cross-covariance matrix C zy for:

[0025]

[0026] Among them, Z P , U f and Y f is a Hankel matrix, defined as follows:

[0027] Z p =[z p (1) … z p (N)]

[0028] U f =[u f (1) … u f (N)]

[0029] Y f =[y f (1) … y f (N)]

[0030] Where N is the number of power load data samples; the T in the upper right corner represents matrix transposition;

[0031] The optimization goal of dynamic canonical correlation analysis is:

[0032]

[0033]

[0034] Among them, W z =[w z (1) w z (2) … w z (s) ],W y =[w y (1) w y (2) … w y (s) ]; w Z and w yis the weight vector obtained after calculating the optimization objective function, I is the unit matrix, and the number in the bracket in the upper right corner of the weight term indicates the number of the most important weight vectors retained, such as w z (1) Indicates retaining 1 most important weight vector;

[0035] Use singular value decomposition as follows:

[0036]

[0037]

[0038] The s1 and s2 in the lower right corner of the matrix are used to represent the matrix decomposed by column, for example, Φ s1 and Φ s2 is Φ s The two matrices decomposed are s is a diagonal matrix containing the maximum s canonical correlation coefficients;

[0039] The sparse weight matrix W is obtained by the following formula sz and W sy ,

[0040]

[0041] Among them, rank(C zy )=s is the matrix C zy rank(C zy )=s,

[0042] According to the relevant sparse weight matrix W sy and W sz , and then the residual generator r of sparse dynamic canonical correlation analysis is obtained s (t) and Hotelling statistic T 2 (t) is as follows,

[0043]

[0044]

[0045] Preferably, in step 3, the fault threshold J is determined by the following formula:

[0046]

[0047] Where P is the probability value of Hotelling statistic, q(T 2 ) is T 2The confidence level can be selected between 0.90-0.99 according to the actual situation, and is generally set to 0.95.

[0048] Preferably, the operation data (including historical operation data and daily operation data) includes power generation, real-time electricity price, temperature, humidity, power consumption, voltage, current and frequency. Among them, power generation, real-time electricity price, temperature and humidity are inputs, and power consumption, voltage, current and frequency are outputs.

[0049] The substantial effect brought about by the present invention is that it can automatically detect abnormal conditions in the power load system, reduce labor costs, improve the operating efficiency of the power load system, ensure that the core components of the power system are away from potential risks, and enhance the safety, rapidity and stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of a method for detecting faults in a power load system based on sparse dynamic canonical correlation analysis of the present invention;

[0051] Figure 2 It is a fault detection schematic diagram of the present invention. DETAILED DESCRIPTION

[0052] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0053] Embodiment: A power load system fault detection method based on sparse dynamic canonical correlation analysis in this embodiment is based on a sparse dynamic canonical correlation analysis algorithm, which automatically detects abnormal conditions in the power load system by mining the hidden intrinsic relationships between power load data (such as power generation, real-time electricity prices, temperature, humidity, power consumption, voltage, current, frequency, etc.). Figure 1 As shown, the method comprises the following steps:

[0054] Step 1: Collect data of the power load system under normal conditions (such as power generation, real-time electricity price, temperature, humidity, power consumption, voltage, current, frequency, etc.) and normalize them as training data;

[0055] Step 2: Use the training data to build a sparse dynamic canonical correlation analysis model, and calculate the Hotelling statistic T based on the established model. 2 , T 2 As an indicator to determine whether a fault has occurred in the power load system;

[0056] Step 3: Use T 2 Calculate the fault threshold J;

[0057] Step 4: Collect the daily operation data of the power load system (such as power generation, real-time electricity price, temperature, humidity, electricity consumption, voltage, current, frequency, etc.) and normalize it as test data, and use this data to calculate the Hotelling statistic T of the test data t 2 ;

[0058] Step 5: The method for judging whether the power load system has a fault is as follows: When T t 2 > J, the power load system has a fault; when T t 2 < J, the power load system is operating normally.

[0059] Specifically, Step 2 is: Construct the past input and output vectors z l and y(t) ∈ R m to describe the dynamic characteristics of the power load system by p (t), the future input vector u f (t) and the future output vector y f (t).

[0060] z p (t) = [y T (t - p) … y T (t - 1) u T (t - p) … u T (t - 1)] T

[0061] u f (t) = [u T (t) … u T (t + f)] T

[0062] y f (t) = [y T (t) … y T (t + f)] T

[0063] Among them, since the power load system is a dynamic process, p and f are defined as the past time delay and the future time delay respectively, so as to establish a dynamic model. l is the number of input variables, and m is the number of output variables. The Hankel matrix is used to calculate the covariance in the next step and is defined as follows:

[0064] Z p = [z p (1) … z p (N)]

[0065] U f = [uf (1) … u f (N)]

[0066] Y f =[y f (1) … y f (N)]

[0067] Where N is the number of power load data samples.

[0068] According to the power load data (such as power generation, real-time electricity price, temperature, humidity, power consumption, voltage, current, frequency, etc.), the covariance matrix C is constructed zz ,C yy and the cross-covariance matrix C zy The calculation method is as follows:

[0069]

[0070]

[0071] The optimization objectives of dynamic canonical correlation analysis are as follows:

[0072]

[0073]

[0074] Among them, C zz and C yy are the covariance matrices of the input and output, respectively, C zy is the covariance matrix between the input and output, w Z and w y It is the weight vector obtained after calculating the optimization objective function, and T represents transpose.

[0075] The optimization problem in the previous step can be restated as:

[0076]

[0077]

[0078] Among them, W z =[w z (1) w z (2) … w z (s) ],W y =[w y (1) w y (2) … w y (s) ].

[0079] Use singular value decomposition as follows:

[0080]

[0081]

[0082] The sparse weight matrix W of sparse dynamic canonical correlation analysis is obtained by solving the following optimization problem sz and W sy ,

[0083]

[0084] Among them, rank(C zy )=s,

[0085] According to the relevant sparse weight matrix W sy and W sz , and then the residual generator r of sparse dynamic canonical correlation analysis is obtained s and Hotelling statistic T 2 as follows,

[0086]

[0087]

[0088] J calculated using kernel density estimation is shown below:

[0089]

[0090] Among them, the confidence level α (generally set to 0.95) is determined based on experience, and the fault threshold J can be easily calculated.

[0091] Figure 2 : is a fault detection schematic diagram of the present embodiment, the blue line represents the real-time Hotelling statistic, the red dotted line represents the fault threshold, and the red dashed line represents the sample dividing point between normal and faulty data of the power load system. It can be seen that before the fault sample, the Hotelling statistic is basically below the fault threshold, and after the fault sample (i.e., the right part of the red dashed line), the Hotelling statistic exceeds the fault threshold, indicating that the proposed method can detect power load system faults in a timely and accurate manner.

[0092] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

[0093] Although the terminology such as Hotelling statistic and fault threshold is used more frequently in this article, the possibility of using other terms is not excluded. These terms are used only to more conveniently describe and explain the essence of the present invention; interpreting them as any additional limitation is contrary to the spirit of the present invention.

Claims

1. A method for detecting faults in a power load system based on sparse dynamic canonical correlation analysis, characterized in that: The following steps are involved: Step 1: Collect historical operating data of the power load system under normal conditions and normalize them as training data; Step 2: Use the training data to establish a sparse dynamic canonical correlation analysis model, and calculate the normal Hotelling statistic T based on the established sparse dynamic canonical correlation analysis model. 2 ; Step 3: Use T 2 Calculate the fault threshold J; Step 4: Collect the daily operation data of the power load system and normalize it as daily data, and use the daily data to calculate the daily Hotelling statistic T t 2 ; Step 5: Determine whether the power load system has failed: When T t 2 > J, the power load system has failed; when T t 2 < J, the power load system is operating normally.

2. The method for detecting faults in a power load system based on sparse dynamic canonical correlation analysis according to claim 1, characterized in that: In step 2, the sparse dynamic canonical correlation analysis model is established as follows: u(t)∈R l is the normalized input variable, t is the serial number directly corresponding to time, y(t)∈R m is the normalized output variable, l is the number of input variables, m is the number of output variables, and the past input-output vector z is defined as p (t) is: z p (t)=[y T (t-p) … y T (t-1) u T (t-p) … u T (t-1)] T , Define the future input vector u f (t) is: u f (t)=[u T (t) … u T (t+f)] T ; Define the future output vector y f (t) is: y f (t)=[y T (t) … y T (t+f)] T p is the past time lag, and f is the future time lag; Construct the input covariance matrix C zz for: Construct the output covariance matrix C yy for: Construct the cross-covariance matrix C zy for: Among them, Z P , U f and Y f is a Hankel matrix, defined as follows: WITH p =[of p (1) ... with p (N)] U f =[u f (1) … u f (N)] AND f =[and f (1) … and f (N)] Where N is the number of power load data samples; the T in the upper right corner represents matrix transposition; The optimization goal of dynamic canonical correlation analysis is: Among them, W z =[w z (1) w z (2) … w z (s) ],W y =[w y (1) w y (2) … w y (s) ]; w Z and w y is the weight vector obtained after calculating the optimization objective function, I is the unit matrix, and the number in the bracket on the upper right corner of the weight term indicates the number of the most important weight vectors retained; Use singular value decomposition as follows: The s1 and s2 in the lower right corner of the matrix are used to represent the matrix decomposed by column. s is a diagonal matrix containing the maximum s canonical correlation coefficients; The sparse weight matrix W is obtained by the following formula sz and W sy , Among them, rank(C zy )=s is the matrix C zy rank, According to the relevant sparse weight matrix W sy and W sz , and then we can get the residual generator r of sparse dynamic canonical correlation analysis s (t) and Hotelling statistic T 2 (t) is as follows, 3. The method for detecting faults in a power load system based on sparse dynamic canonical correlation analysis according to claim 2, characterized in that: In step 3, the fault threshold J is determined by the following formula: Where P is the probability value of Hotelling statistic, q(T 2 ) is T 2 The probability density function, probability density function q(T 2 ) is estimated by kernel density estimation method, and the confidence level is selected between 0.90-0.

99.

4. A method for detecting faults in a power load system based on sparse dynamic canonical correlation analysis according to claim 1, 2 or 3, characterized in that: The operating data includes power generation, real-time electricity price, temperature, humidity, power consumption, voltage, current and frequency.

Images