Fault diagnosis method for medium voltage AC / DC hybrid system based on machine learning

Abstract

The present invention relates to the field of cable fault detection technology, and specifically to a fault diagnosis method for a medium-voltage AC / DC hybrid system based on machine learning; obtaining the abnormal difference degree at any time according to the difference characteristics of the zero-sequence current value and the difference characteristics of the zero-sequence current direction in the zero-sequence current time series; obtaining the power imbalance degree at any time according to the zero-sequence current value at any time and the distribution characteristics of the abnormal difference degree between the preset adjacent time period; performing time series decomposition on the power imbalance degree sequence to obtain the power imbalance trend values ​​at different times; obtaining the power abnormality degree and abnormal reference value at any time according to the difference characteristics of the power imbalance degree at any time and the power imbalance trend value at any time and the adjacent historical time. The present invention obtains the power abnormality index according to the difference characteristics of the power abnormality degree and the abnormal reference value at any time and performs fault detection on the medium-voltage AC / DC hybrid system, thereby improving the accuracy of fault detection.

Description

Technical Field

[0001] The present invention relates to the technical field of cable fault detection, and in particular to a medium voltage AC / DC hybrid system fault diagnosis method based on machine learning. Background Art

[0002] The medium voltage AC / DC hybrid system is a power system that combines AC and DC transmission technologies. The medium voltage AC / DC hybrid system needs to convert AC to DC. During the conversion process, the load balance of the three-phase line can be reflected by monitoring the change of zero-sequence current. When there is a load imbalance, the stability and safety of the AC / DC hybrid system will decrease. Therefore, it is necessary to monitor the zero-sequence current characteristics to improve the operational reliability of the AC / DC hybrid system. At present, current transformers are usually used to monitor the zero-sequence current of the three-phase line. When the zero-sequence current is not 0, it means that there is a load imbalance and timely maintenance is required. However, this method has certain limitations. For example, the switching of electrical equipment and sudden load changes will lead to the existence of zero-sequence current. This traditional method ignores the normal dynamic changes of the load in the power grid, which is prone to false detection, and ultimately leads to reduced detection accuracy of load imbalance. Summary of the invention

[0003] In order to solve the above-mentioned technical problem of low accuracy in judging load imbalance by monitoring zero-sequence current through current transformer, the object of the present invention is to provide a medium voltage AC / DC hybrid system fault diagnosis method based on machine learning, and the technical solution adopted is as follows:

[0004] Obtain zero-sequence current time series of AC / DC hybrid system;

[0005] In the zero-sequence current time series, the abnormal difference degree at any moment is obtained according to the difference characteristics of the zero-sequence current value at any moment and the adjacent historical moments, and the difference characteristics of the zero-sequence current direction; the power imbalance degree at any moment is obtained according to the distribution characteristics of the zero-sequence current value and the abnormal difference degree at any moment and the preset adjacent time period;

[0006] Constructing a power imbalance degree sequence according to the power imbalance degree at all times; performing time series decomposition on the power imbalance degree sequence to obtain power imbalance trend values ​​at different times; obtaining the power anomaly degree at any time according to the power imbalance degree at any time and the difference characteristics of the power imbalance trend values ​​at any time and adjacent historical times;

[0007] An abnormality reference value is obtained according to the power abnormality at a historical moment; an power abnormality index is obtained according to the difference characteristics between the power abnormality at any moment and the abnormality reference value; and fault detection is performed on a medium voltage AC / DC hybrid system according to the power abnormality index.

[0008] Furthermore, the step of obtaining the abnormal difference degree at any moment in the zero-sequence current time series according to the difference characteristics of the zero-sequence current value at any moment and the adjacent historical moments and the difference characteristics of the zero-sequence current direction includes:

[0009] The absolute value of the difference between the zero-sequence current value at any moment and the adjacent historical moment is calculated to obtain the zero-sequence current difference characteristic value; the absolute value of the difference between the zero-sequence current angle at any moment and the adjacent historical moment is calculated to obtain the direction difference characteristic value; the product of the zero-sequence current difference characteristic value and the direction difference characteristic value is calculated to obtain the abnormal difference degree at any moment.

[0010] Furthermore, the step of obtaining the power imbalance degree at any moment according to the distribution characteristics of the zero-sequence current value and the abnormal difference degree at any moment and the preset adjacent time period includes:

[0011] The average value of the zero-sequence current value at any moment and the preset adjacent time period is calculated to obtain the average characteristic value of the zero-sequence current; the coefficient of variation of the zero-sequence current value at any moment and the preset adjacent time period is calculated and negatively correlated to obtain the difference stable characteristic value; the product of the zero-sequence current average characteristic value and the difference stable characteristic value is calculated to obtain the power imbalance degree at any moment.

[0012] Furthermore, the step of obtaining the power anomaly degree at any moment according to the power imbalance degree at any moment and the difference characteristics of the power imbalance trend values ​​at any moment and adjacent historical moments includes:

[0013] The difference between the power imbalance trend value at the arbitrary moment and the adjacent historical moment is calculated and positively correlated to obtain a trend characteristic value; the product of the power imbalance degree at the arbitrary moment and the trend characteristic value is calculated to obtain the power anomaly degree at the arbitrary moment.

[0014] Furthermore, the step of obtaining an abnormal reference value according to the degree of power abnormality at a historical moment includes:

[0015] The sum of the minimum value of the power abnormality and a preset minimum positive number is calculated to obtain the abnormal reference value.

[0016] Furthermore, the step of obtaining the power anomaly index according to the difference characteristics between the power anomaly degree at any time and the abnormal reference value includes:

[0017] The ratio of the power anomaly degree at any time to the abnormal reference value is calculated to obtain a proportionality coefficient; the product of the proportionality coefficient and the power anomaly degree at any time is calculated and normalized to obtain a power anomaly index at any time.

[0018] Furthermore, the step of performing fault detection on the medium voltage AC / DC hybrid system according to the power anomaly index includes:

[0019] When the power anomaly index at the latest moment exceeds the preset anomaly threshold, a power load imbalance occurs in the medium voltage AC / DC hybrid system.

[0020] Furthermore, the step of performing time series decomposition on the power imbalance degree sequence to obtain power imbalance trend values ​​at different times includes:

[0021] The power imbalance degree sequence is time-series decomposed by using the STL time-series decomposition algorithm, and the trend item obtained by the decomposition is used as the power imbalance trend value.

[0022] The present invention has the following beneficial effects:

[0023] In the present invention, obtaining the abnormal difference can characterize the difference characteristics of the zero-sequence current of the three-phase line at different times; there are differences in the continuous characteristics of the zero-sequence current caused by load mutation and load imbalance, and then the power imbalance degree is obtained according to the abnormal difference and the zero-sequence current, and the power imbalance degree can characterize the load balance of the three-phase line. Obtaining the power imbalance degree sequence can be used to analyze the development trend of the load imbalance situation and further determine the abnormal degree of the load balance situation; obtaining the power imbalance trend value can characterize the trend size of the load imbalance situation at different times. Obtaining the power abnormality can accurately characterize the severity of the load imbalance according to the power imbalance degree and the development trend; obtaining the abnormal baseline value can characterize the load balance characteristics of the three-phase line under normal conditions at historical moments, and then obtaining the power abnormality index can accurately reflect the severity of the load imbalance of the three-phase line at the current moment, effectively reducing the situation where the false alarm frequency is too high due to load mutation and affecting the operating reliability of the medium-voltage AC / DC hybrid system, and improving the accuracy of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 A flow chart of a medium voltage AC / DC hybrid system fault diagnosis method based on machine learning is provided in accordance with an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, characteristics and effects of a medium voltage AC / DC hybrid system fault diagnosis method based on machine learning proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0027] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0028] A specific scheme of a medium voltage AC / DC hybrid system fault diagnosis method based on machine learning provided by the present invention is described in detail below with reference to the accompanying drawings.

[0029] See also Figure 1 , which shows a flow chart of a medium voltage AC / DC hybrid system fault diagnosis method based on machine learning provided by an embodiment of the present invention, the method comprising the following steps:

[0030] Step S1, obtaining a zero-sequence current time series of an AC / DC hybrid system.

[0031] In an embodiment of the present invention, the implementation scenario is to detect the load imbalance of a medium-voltage AC / DC hybrid system to improve the operational reliability of the medium-voltage AC / DC hybrid system. First, the zero-sequence current time series of the AC / DC hybrid system is obtained, and current transformers are arranged on the three-phase lines of the AC / DC hybrid system to obtain the zero-sequence current at each acquisition moment, and the zero-sequence current time series is constructed according to the zero-sequence current at each acquisition moment in chronological order. In a normal three-phase system, the vector sum of the three-phase currents is 0, and the corresponding zero-sequence current is 0; when asymmetric operation and single-phase operation load imbalance occur, the vector sum of the three-phase currents passing through the current transformer is not 0, and the generated induced current is a zero-sequence current.

[0032] Step S2, in the zero-sequence current time series, the abnormal difference degree at any moment is obtained according to the difference characteristics of the zero-sequence current value at any moment and the adjacent historical moments, and the difference characteristics of the zero-sequence current direction; the power imbalance degree at any moment is obtained according to the distribution characteristics of the zero-sequence current value and the abnormal difference degree at any moment and the preset adjacent time period.

[0033] When electrical equipment switching and load mutation occur in the power grid, transient zero-sequence current may occur. Traditional monitoring methods may ignore the normal load changes in such power grids, which increases the possibility of false alarms for transient zero-sequence currents in the power grid, increases the response frequency of related protection equipment, and reduces the operational stability of the power system. Since transient zero-sequence currents may be caused by switching of electrical equipment and other actions, and real load imbalance will cause zero-sequence currents for a longer period of time, it can be analyzed based on the difference characteristics of zero-sequence currents at adjacent moments. Therefore, in the zero-sequence current time series, the abnormal difference degree at any moment is obtained based on the difference characteristics of the zero-sequence current value at any moment and the adjacent historical moments and the difference characteristics of the zero-sequence current direction.

[0034] Preferably, in an embodiment of the present invention, the step of obtaining the abnormal difference degree includes: calculating the absolute value of the difference between the zero-sequence current value at any moment and the adjacent historical moment, and obtaining the zero-sequence current difference characteristic value; when the zero-sequence current difference characteristic value is larger, it means that the difference between the zero-sequence current value at any moment and the adjacent historical moment is larger. Calculate the absolute value of the difference between the zero-sequence current angle at any moment and the adjacent historical moment, and obtain the direction difference characteristic value; the zero-sequence current is calculated from the three-phase current, so the zero-sequence current is a vector with magnitude and direction. When the direction difference characteristic value is larger, it means that the direction difference between the zero-sequence current at any moment and the adjacent historical moment is larger. Calculate the product of the zero-sequence current difference characteristic value and the direction difference characteristic value to obtain the abnormal difference degree at any moment; when the abnormal difference degree is larger, it means that the difference between the zero-sequence current characteristics at any moment and the adjacent historical moment is larger.

[0035] Furthermore, if a phase of the three-phase electricity is unbalanced, the zero-sequence current will show a certain trend in a local time period, and the abnormal difference within the time period will be relatively stable; if an instantaneous random zero-sequence current occurs, the abnormal difference within a period of time will fluctuate greatly. Therefore, the power imbalance at any moment is obtained according to the distribution characteristics of the zero-sequence current value and the abnormal difference between any moment and the preset adjacent time period; preferably, in an embodiment of the present invention, the step of obtaining the power imbalance includes: calculating the average value of the zero-sequence current value between any moment and the preset adjacent time period, and obtaining the average characteristic value of the zero-sequence current; in an embodiment of the present invention, the preset adjacent time period is composed of the 50 most recent acquisition moments before and after the arbitrary moment; when the average characteristic value of the zero-sequence current is larger, it means that the three-phase line load is more likely to be unbalanced in the time period, and when the average characteristic value of the zero-sequence current is smaller, it means that it is more likely to be an instantaneous random zero-sequence current. Calculate the coefficient of variation of the zero-sequence current value at any moment and the preset adjacent time period and negatively map them to obtain the difference stability characteristic value; it should be noted that the coefficient of variation belongs to the prior art, and the specific calculation steps are no longer repeated. When the coefficient of variation is larger, it means that the abnormal difference in the time period is more discrete, and the difference stability characteristic value is smaller, the more likely it is an instantaneous random zero-sequence current. When the coefficient of variation is smaller, it means that the abnormal difference in the time period is more stable, and the difference stability characteristic value is larger, the more likely it is a three-phase line load imbalance. Calculate the product of the average characteristic value of the zero-sequence current and the difference stability characteristic value to obtain the power imbalance at any moment; when the power imbalance is larger, it means that the load imbalance of the three-phase line is more likely to occur in the local time period at any moment.

[0036] Step S3, constructing a power imbalance degree sequence according to the power imbalance degrees at all times; decomposing the power imbalance degree sequence in time series to obtain power imbalance trend values ​​at different times; and obtaining the power anomaly degree at any time according to the power imbalance degree at any time and the difference characteristics of the power imbalance trend values ​​at any time and the adjacent historical times.

[0037] Since the short-term zero-sequence current in the power system has a small impact on the power system, if the load imbalance degree at the current moment has a significant growth trend and severity compared to the historical moments, it means that the load imbalance situation has an increasingly greater impact on the power system and requires timely warning. Therefore, a power imbalance degree sequence is constructed according to the power imbalance degree at all times, and the severity of the load imbalance is judged according to the power imbalance degree; then the power imbalance degree sequence is time-series decomposed to obtain the power imbalance trend value at different times. Preferably, in an embodiment of the present invention, the step of obtaining the current imbalance trend value includes: time-series decomposition of the power imbalance degree sequence through the STL time-series decomposition algorithm, and the trend item obtained by the decomposition is used as the power imbalance trend value; it should be noted that the STL time-series decomposition algorithm belongs to the prior art, and the specific steps are not repeated. The algorithm can decompose the original sequence into trend items, seasonal items and residual items, and the power imbalance trend value can reflect the changing trend of the power imbalance degree at different times.

[0038] Furthermore, if the power imbalance at any moment is larger and there is an obvious growth trend, it means that the load imbalance of the three-phase line is getting more and more serious. Therefore, the power abnormality at any moment is obtained according to the difference characteristics of the power imbalance at any moment and the power imbalance trend value at any moment and the adjacent historical moment; preferably, in an embodiment of the present invention, the step of obtaining the power abnormality includes: calculating the difference between the power imbalance trend value at any moment and the adjacent historical moment and positively mapping it to obtain the trend characteristic value; when the power imbalance trend value at any moment is larger than the power imbalance trend value at the adjacent historical moment, the larger the trend characteristic value is, which means that the load imbalance of the three-phase line has a trend of getting more and more serious. Calculate the product of the power imbalance at any moment and the trend characteristic value to obtain the power abnormality at any moment; when the power imbalance and the trend characteristic value are larger, the power abnormality is larger, which means that the load imbalance of the three-phase line is getting more and more serious at any moment. The formula for obtaining the power abnormality includes:

[0039]

[0040] In the formula, represents the power anomaly at the tth moment, represents the power imbalance at the tth moment, represents the power imbalance trend value at the tth moment, represents the power imbalance trend value at the t-1th moment, represents an exponential function with a natural constant as base, Indicates the trend characteristic value.

[0041] Step S4, obtaining an abnormal reference value according to the power abnormality at a historical moment; obtaining a power abnormality index according to the difference characteristics between the power abnormality at any moment and the abnormal reference value; and performing fault detection on the medium voltage AC / DC hybrid system according to the power abnormality index.

[0042] The smaller the power anomaly is, the more balanced the load on the three-phase line is, and the more stable the operation of the medium-voltage AC / DC hybrid system is. Therefore, the abnormal reference value can be obtained according to the power anomaly at the historical moment; preferably, in an embodiment of the present invention, the step of obtaining the abnormal reference value includes: calculating the sum of the minimum value of the power anomaly and a preset minimum positive number to obtain the abnormal reference value; the abnormal reference value represents the power anomaly when the load on the three-phase line is most balanced. In an embodiment of the present invention, the preset minimum positive number is 0.1, and the implementer can determine it by himself according to the implementation scenario, in order to avoid the situation where the power anomaly is 0.

[0043] Further, the power anomaly index can be obtained according to the difference characteristics between the power anomaly degree and the abnormal reference value at any time; preferably, in an embodiment of the present invention, the step of obtaining the power anomaly index includes: calculating the ratio of the power anomaly degree to the abnormal reference value at any time to obtain the proportional coefficient; when the proportional coefficient is larger, it means that the power anomaly degree at any time is larger than the abnormal reference value, and the load of the three-phase line is more unbalanced. The product of the proportional coefficient and the power anomaly degree at any time is calculated and normalized to obtain the power anomaly index at any time; calculating the product of the proportional coefficient and the power anomaly degree can avoid the abnormal reference value being too large and causing the proportional coefficient to be small, affecting the detection accuracy; therefore, the larger the power anomaly index is, the more unbalanced the load of the three-phase line at any time. Then, the medium-voltage AC / DC hybrid system can be fault-detected according to the power anomaly index. When the power anomaly index at the latest moment exceeds the preset abnormal threshold, the medium-voltage AC / DC hybrid system has an unbalanced power load; in an embodiment of the present invention, the preset abnormal threshold is 0.5, and the implementer can determine it according to the implementation scenario. Detecting the load condition according to the power anomaly index can effectively avoid the instantaneous random zero-sequence current causing the load detection device to have an excessively high false alarm frequency, affecting the operating stability of the medium-voltage AC / DC hybrid system, and ultimately improving the accuracy of fault diagnosis.

[0044] In summary, the embodiment of the present invention provides a method for fault diagnosis of a medium-voltage AC / DC hybrid system based on machine learning; in the zero-sequence current time series, the degree of abnormal difference at any time is obtained according to the difference characteristics of the zero-sequence current value and the difference characteristics of the zero-sequence current direction; the degree of power imbalance at any time is obtained according to the distribution characteristics of the zero-sequence current value and the degree of abnormal difference at any time and the preset adjacent time period; the power imbalance sequence is decomposed in time series to obtain the power imbalance trend values ​​at different times; the degree of power abnormality and the abnormal reference value at any time are obtained according to the difference characteristics of the power imbalance at any time and the power imbalance trend value at any time and the adjacent historical time. The present invention obtains the power abnormality index according to the difference characteristics of the power abnormality and the abnormal reference value at any time and performs fault detection on the medium-voltage AC / DC hybrid system, thereby improving the accuracy of fault detection.

[0045] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0046] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A fault diagnosis method for a medium voltage AC / DC hybrid system based on machine learning, characterized in that: The method comprises the following steps: Obtain zero-sequence current time series of AC / DC hybrid system; In the zero-sequence current time series, the abnormal difference degree at any moment is obtained according to the difference characteristics of the zero-sequence current value at any moment and the adjacent historical moments, and the difference characteristics of the zero-sequence current direction; the power imbalance degree at any moment is obtained according to the distribution characteristics of the zero-sequence current value and the abnormal difference degree at any moment and the preset adjacent time period; Constructing a power imbalance degree sequence according to the power imbalance degree at all times; performing time series decomposition on the power imbalance degree sequence to obtain power imbalance trend values ​​at different times; obtaining the power anomaly degree at any time according to the power imbalance degree at any time and the difference characteristics of the power imbalance trend values ​​at any time and adjacent historical times; An abnormality reference value is obtained according to the power abnormality at a historical moment; an power abnormality index is obtained according to the difference characteristics between the power abnormality at any moment and the abnormality reference value; and fault detection is performed on a medium voltage AC / DC hybrid system according to the power abnormality index.

2. A method for fault diagnosis of a medium voltage AC / DC hybrid system based on machine learning according to claim 1, characterized in that: The step of obtaining the abnormal difference degree at any moment according to the difference characteristics of the zero-sequence current value at any moment and the difference characteristics of the zero-sequence current direction in the zero-sequence current time series between any moment and the adjacent historical moments comprises: The absolute value of the difference between the zero-sequence current value at any moment and the adjacent historical moment is calculated to obtain the zero-sequence current difference characteristic value; the absolute value of the difference between the zero-sequence current angle at any moment and the adjacent historical moment is calculated to obtain the direction difference characteristic value; the product of the zero-sequence current difference characteristic value and the direction difference characteristic value is calculated to obtain the abnormal difference degree at any moment.

3. A method for fault diagnosis of a medium voltage AC / DC hybrid system based on machine learning according to claim 1, characterized in that: The step of obtaining the power imbalance degree at any time according to the distribution characteristics of the zero-sequence current value and the abnormal difference degree at any time and the preset adjacent time period includes: The average value of the zero-sequence current value at any moment and the preset adjacent time period is calculated to obtain the average characteristic value of the zero-sequence current; the coefficient of variation of the zero-sequence current value at any moment and the preset adjacent time period is calculated and negatively correlated to obtain the difference stable characteristic value; the product of the zero-sequence current average characteristic value and the difference stable characteristic value is calculated to obtain the power imbalance degree at any moment.

4. The method for fault diagnosis of a medium voltage AC / DC hybrid system based on machine learning according to claim 1, characterized in that: The step of obtaining the power abnormality at any moment according to the power imbalance at any moment and the difference characteristics of the power imbalance trend values ​​at any moment and adjacent historical moments comprises: The difference between the power imbalance trend value at the arbitrary moment and the adjacent historical moment is calculated and positively correlated to obtain a trend characteristic value; the product of the power imbalance degree at the arbitrary moment and the trend characteristic value is calculated to obtain the power anomaly degree at the arbitrary moment.

5. The method for fault diagnosis of a medium voltage AC / DC hybrid system based on machine learning according to claim 1, characterized in that: The step of obtaining an abnormal reference value according to the degree of power abnormality at a historical moment includes: The sum of the minimum value of the power abnormality and a preset minimum positive number is calculated to obtain the abnormal reference value.

6. A method for fault diagnosis of a medium voltage AC / DC hybrid system based on machine learning according to claim 1, characterized in that: The step of obtaining the power anomaly index according to the difference characteristics between the power anomaly degree at any time and the abnormal reference value comprises: The ratio of the power anomaly degree at any time to the abnormal reference value is calculated to obtain a proportionality coefficient; the product of the proportionality coefficient and the power anomaly degree at any time is calculated and normalized to obtain a power anomaly index at any time.

7. The method for fault diagnosis of a medium voltage AC / DC hybrid system based on machine learning according to claim 1, characterized in that: The step of performing fault detection on the medium voltage AC / DC hybrid system according to the power anomaly index comprises: When the power anomaly index at the latest moment exceeds the preset anomaly threshold, a power load imbalance occurs in the medium voltage AC / DC hybrid system.

8. The method for fault diagnosis of a medium voltage AC / DC hybrid system based on machine learning according to claim 1, characterized in that: The step of performing time series decomposition on the power imbalance degree sequence to obtain power imbalance trend values ​​at different times includes: The power imbalance degree sequence is time-series decomposed by using the STL time-series decomposition algorithm, and the trend item obtained by the decomposition is used as the power imbalance trend value.

Images