Substation secondary circuit signal detection method based on wavelet decomposition rolling learning

Abstract

The present invention provides a substation secondary circuit signal detection method based on wavelet decomposition and rolling learning, the steps are as follows: add a time window to the signal; select the time window in which the signal changes; perform wavelet decomposition on the signal in the selected time window, and decompose it step by step from low frequency to high frequency; obtain wavelet decomposition coefficients of different scales, and then obtain wavelet decomposition coefficients corresponding to different response events; according to the response event corresponding to the signal, select the wavelet decomposition coefficient of the historical signal as a characteristic parameter, and use the response event generated in each online detection as feedback for rolling learning. Beneficial effects of the present invention: through signal wavelet decomposition and rolling learning, the interference of noise signals is further reduced, and the identification of corresponding events of the system substation is enhanced, including the accuracy and real-time performance of event identification. Rolling learning is a kind of online learning. As the running time becomes longer, the accuracy of signal detection can be continuously improved, and it has a good use effect.

Description

Technical Field

[0001] The present invention belongs to the technical field of substation signal detection, and in particular relates to a substation secondary circuit signal detection method based on wavelet decomposition and rolling learning. Background Art

[0002] Automation technology is being used more and more widely in substations. The station-side monitoring background system, the ground-dispatching master station monitoring system, the central-dispatching master station monitoring system, the network-dispatching master station monitoring system, etc. are connected to the substation front-end operating equipment through a large number of telemetry and telesignaling signals to detect the substation secondary circuit signals. However, the complex electromagnetic environment of the substation leads to a poor signal transmission environment and is easily affected by factors such as jitter, poor contact, and mechanical failure. Substation signals are often prone to distortion, glitches, or white noise, which can cause false operation of protection devices and false alarms. In the past, simple manual reset and delayed response methods were mainly used to deal with such problems. However, their processing efficiency is still low, and it is difficult to achieve an effective balance in the identification of interference signals and event signals, which can easily cause delays in the processing of important events in the power grid. Summary of the Invention

[0003] In view of this, the present invention aims to overcome the deficiencies of the above-mentioned problems in the prior art and proposes a substation secondary circuit signal detection method based on wavelet decomposition rolling learning.

[0004] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0005] The substation secondary circuit signal detection method based on wavelet decomposition rolling learning includes the following steps:

[0006] A. Add a time window to the signal;

[0007] B. Select the time window when the signal changes;

[0008] C. Perform wavelet decomposition on the signal within the time window in step B, and decompose it step by step from low frequency to high frequency;

[0009] D. Obtaining wavelet decomposition coefficients of different scales, and then obtaining wavelet decomposition coefficients corresponding to different response events;

[0010] E. According to the response event corresponding to the signal, the wavelet decomposition coefficients of the historical signal are selected as feature parameters, and the response event generated in each online detection is used as feedback for rolling learning.

[0011] Furthermore, the time window in step B includes 2 to the power of the exponential number of signal sampling points.

[0012] Furthermore, in step C, the wavelet decomposition is truncated after reaching a set hard threshold.

[0013] Furthermore, the response event corresponding to the signal change in the substation includes at least one of the following: overcurrent, overvoltage, tripping, and noise.

[0014] Furthermore, the steps for rolling learning in step E are as follows:

[0015] E1. Assume that s1, s2, s3, s4…s q They represent q kinds of response events represented by the signals in the substation. At this time, corresponding to different response events s i , select p coefficients of wavelet decomposition as eigenvector k i =(a i1 ,a i2 ,a i3 ,……,a ip ), where a i1 、a i2 、a i3 、……、a ip Indicates response event s i p coefficients obtained after p-fold decomposition truncation;

[0016] E2. Define c i In response to events i The eigenvector centroid of has the following expression,

[0017] c i =(k i1 +k i2 …+k it ) / t,

[0018] Among them, k i1 、k i2 ,…,k it The vectors are historical data that correspond to the response events s in the learning process. i t eigenvectors of ;

[0019] E3. During the online detection process, the sampled current signal is decomposed by wavelet to obtain the feature vector k u , and obtain the Euclidean distance between it and the centroid of the feature vector of each known response event;

[0020] E4. Determine the Euclidean distances obtained in step E3, and select the response event corresponding to the nearest eigenvector centroid as the response event corresponding to the current signal. Assume that the eigenvector k corresponding to the sampled current signal is u and respond to events i The eigenvector centroid c i If the distance is short, then the response event corresponding to the current signal sampled is s i ;

[0021] E5. Perform rolling learning based on the online detection results, and convert the feature vector k corresponding to the current signal u Dynamic updates in response to events i The eigenvector centroid c i , at this time, the response event s i The eigenvector centroid c i There are the following expressions,

[0022] c i =(k i1 +k i2 …+k it +k u ) / (t+1).

[0023] Compared with the prior art, the present invention has the following advantages:

[0024] The substation secondary circuit signal detection method based on wavelet decomposition and rolling learning described in the present invention further reduces the interference of noise signals through signal wavelet decomposition and rolling learning, and strengthens the recognition of corresponding events in the system substation, including the accuracy and real-time performance of event recognition. Rolling learning is a kind of online learning. As the running time increases, the signal detection accuracy can be continuously improved, and it has a good use effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings, which constitute part of the present invention, are provided to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are provided to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0026] Figure 1 This is a schematic diagram of the principle of a substation secondary circuit signal detection method based on wavelet decomposition rolling learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other.

[0028] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0029] like Figure 1 As shown, the substation secondary circuit signal detection method based on wavelet decomposition rolling learning includes the following steps:

[0030] A. Add time windows to substation signals;

[0031] B. Select the time window when the signal changes, and slide the window as time extends;

[0032] C. Perform discrete wavelet decomposition on the signal within the time window in step B, and decompose it step by step from low frequency to high frequency;

[0033] D. Obtain wavelet decomposition coefficients of different scales, and then obtain wavelet decomposition coefficients corresponding to different response events. Different signals correspond to different response events.

[0034] E. According to the different response events corresponding to the signal, the wavelet decomposition coefficients of the historical signal are selected as feature parameters, and the response events generated in each online detection are used as feedback for rolling learning.

[0035] The time window in step B contains 2 exponential power signal sampling points, which is set based on the requirements of discrete wavelet.

[0036] In step C, the wavelet decomposition is truncated after reaching the set hard threshold. For example, after p decompositions and truncations, p coefficients a1, a2, a3, ..., a p .

[0037] Signal changes in a substation correspond to at least one of the following response events: overcurrent, overvoltage, tripping, and noise. Stable signals do not correspond to any event; only changes in the signal indicate that the system may have generated different types of response events.

[0038] The steps for rolling learning in step E are as follows:

[0039] E1. Assume that s1, s2, s3, s4…s q They represent q kinds of response events represented by the signals in the substation. At this time, corresponding to different response events s i , select p coefficients of wavelet decomposition as eigenvector k i =(a i1 ,a i2 ,a i3 ,……,a ip ), where a i1 、a i2 、a i3 、……、a ip Indicates response event s i The p coefficients obtained after p-time decomposition truncation will respond to the event s i and the corresponding eigenvector k i It is recorded as the following mapping relationship, F(k i )=s i ;

[0040] E2. Define c i In response to events i The eigenvector centroid of has the following expression,

[0041] ci =(k i1 +k i2 …+k it ) / t,

[0042] Among them, k i1 、k i2 ,…,k it The vectors are historical data that correspond to the response events s in the learning process. i This is because there may be multiple eigenvectors k in different situations due to factors such as noise. i The values ​​correspond to the same response event s i situation;

[0043] E3. During the online detection process, the sampled current signal is decomposed by wavelet to obtain the feature vector k u , and obtain the Euclidean distance between it and the center of gravity of the feature vector of each known response event; In this embodiment, for example, in three-dimensional space, the feature vector k corresponding to the current signal sampled is u The coordinates of (x u ,y u ,z u ), known response events s i The eigenvector centroid c i The coordinates of (x i ,y i ,z i ), then the eigenvector k u to the centroid of the eigenvector c i The Euclidean distance ρ i The expression is as follows,

[0044]

[0045] E4. Determine the Euclidean distances obtained in step E3, and select the response event corresponding to the nearest eigenvector centroid as the response event corresponding to the current signal. Assume that the eigenvector k corresponding to the sampled current signal is u and respond to events i The eigenvector centroid c i If the distance is short, then the response event corresponding to the current signal sampled is s i In this embodiment, by comparing the characteristic vector k corresponding to the current signal u The Euclidean distance from the center of gravity of the feature vector of each known response event is selected, and the response event corresponding to the center of gravity of the feature vector with the closest Euclidean distance is selected as the currently sampled feature vector k u The corresponding response events, i.e., the similarity is judged based on the Euclidean distance in the multidimensional space;

[0046] E5. Perform rolling learning based on the online detection results, and convert the feature vector k corresponding to the current signal u Dynamic updates in response to events i The eigenvector centroid c i , at this time, the response event s i The eigenvector centroid c i There are the following expressions,

[0047] c i =(k i1 +k i2 …+k it +k u ) / (t+1),

[0048] This will continuously improve the detection capability of this method.

[0049] This signal detection method further reduces the interference of noise signals through signal decomposition and rolling learning, and strengthens the identification of corresponding events in the system substation, including the accuracy and real-time performance of event identification. Rolling learning is a type of online learning that can continuously improve signal detection accuracy as the running time increases, and has a good use effect.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A substation secondary circuit signal detection method based on wavelet decomposition rolling learning, characterized in that: The following steps are involved: A. Add a time window to the signal; B. Select the time window in which the signal changes; C. Perform wavelet decomposition on the signal in the time window in step B, and decompose it step by step from low frequency to high frequency; D. Obtaining wavelet decomposition coefficients of different scales, and then obtaining wavelet decomposition coefficients corresponding to different response events; E. According to the response event corresponding to the signal, the wavelet decomposition coefficient of the historical signal is selected as the characteristic parameter, and the response event generated in each online detection is used as feedback for rolling learning; The steps for rolling learning in step E are as follows, E1. Assume s1, s2, s3, s4…s q They represent the q response events represented by the signals in the substation. At this time, corresponding to different response events s i , select p coefficients of wavelet decomposition as eigenvector k i =(a i1 ,a i2 ,a i3 ,……,a ip ), where a i1 、a i2 、a i3 ,……,a ip Indicates the response event i p coefficients obtained after p-fold decomposition truncation; E2. Definition of c i In response to events i The eigenvector centroid of has the following expression, c i =(k i1 +k i2 …+k it ) / t, Among them, k i1 , k i2 , …, k it The vector is the historical data learning process corresponding to the response event s i t eigenvectors of ; E3. During the online detection process, the sampled current signal is decomposed by wavelet to obtain the feature vector k u , and obtain the Euclidean distance between it and the centroid of the feature vector of each known response event; E4. Determine the Euclidean distances obtained in step E3, select the response event corresponding to the centroid of the eigenvector closest to the response event corresponding to the current signal, and assume that the eigenvector k corresponding to the sampled current signal u and respond to events i The eigenvector centroid c i If the distance is short, then the response event corresponding to the current signal sampled is s i ; E5. Perform rolling learning based on the online detection results and convert the feature vector k corresponding to the current signal u Dynamic updates in response to events i The eigenvector centroid c i , at this time, the response event s i The eigenvector centroid c i There are the following expressions, c i =(k i1 +k i2 …+k it +k u ) / (t+1)。 2. The substation secondary circuit signal detection method based on wavelet decomposition rolling learning according to claim 1 is characterized in that: In step B, the time window includes exponentially two signal sampling points.

3. The substation secondary circuit signal detection method based on wavelet decomposition rolling learning according to claim 1 is characterized in that: In step C, the wavelet decomposition is truncated after reaching the set hard threshold.

4. The substation secondary circuit signal detection method based on wavelet decomposition rolling learning according to claim 1 is characterized in that: The response event corresponding to the signal change in the substation includes at least one of the following: overcurrent, overvoltage, tripping, and noise.

Images