Method for extracting effective direct current signals from large number of noisy points and ripples

Abstract

The invention provides a method for extracting an effective direct current signal from a large number of noisy points and ripples. The method comprises the following steps: preprocessing a signal output by a high-voltage cable insulation state monitoring current sensor by adopting a 4-order active low-pass filter circuit; performing signal-noise separation operation on the output signal of the hardware filtering step by adopting a wavelet threshold denoising algorithm, and dividing the output signal into a first wavelet coefficient set and a second wavelet coefficient set according to a preset threshold; performing sliding median filtering on the output signal of the noise reduction processing sub-step to suppress Gaussian noise points and ultra-high and ultra-low noise points in the signal; recursive average filtering is carried out on the output signal of the median filtering sub-step; and through multiple times of iteration processing, analyzing and comparing the similarity of the data to obtain a finally extracted effective direct current signal. The high-order active low-pass filter circuit and the advanced software filtering algorithm are combined, the direct current effective component in the signal can be effectively extracted, and the accuracy of insulation monitoring is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system and electrical equipment monitoring, and in particular to a method for extracting effective DC signals from a large number of noise points and ripples. Background Art

[0002] The stable operation of power systems and electrical equipment is of great significance to ensuring power supply and safe production. However, electrical equipment failures occur frequently, most of which are caused by insulation damage. In order to measure the insulation performance of electrical equipment, the traditional method usually requires the use of a megohmmeter to make a judgment during a power outage. This method is not only cumbersome to operate, but also cannot achieve online monitoring, making it difficult to detect and handle insulation failures in a timely manner.

[0003] In order to overcome the shortcomings of traditional methods, the world has begun to use partial discharge and leakage current measurement methods to achieve online measurement of equipment insulation resistance. The partial discharge method determines the degree of insulation damage by analyzing the high-frequency pulse component generated by partial discharge of insulation, but this method can only be used as a reference and cannot provide accurate insulation resistance values.

[0004] In the actual application and debugging of the high-voltage cable insulation status monitoring current sensor, we found that the collected waveform contains a large amount of high-frequency noise, low-frequency noise, AC components and DC effective components. The existence of these noise points and ripples makes it extremely difficult to extract effective DC signals. How to extract accurate and effective DC signals from a large amount of noise points and ripples has become a major difficulty and challenge in the development of system insulation monitoring devices.

[0005] There are many deficiencies in the existing technology when dealing with this problem. First, the hardware filtering method is relatively simple, usually using passive low-pass filtering or low-order active low-pass filtering. These methods have poor filtering effects and it is difficult to effectively remove high-frequency noise and low-frequency noise in the waveform. Secondly, the cutoff frequency setting of the filter circuit is unreasonable. The traditional technology does not perform accurate calculations based on the values ​​of resistance R, capacitance C, etc. in the filter circuit, resulting in unsatisfactory filtering effects. In addition, the traditional technology does not perform "signal-to-noise separation" and treats noise as an effective signal for software calculations, further reducing the accuracy of the signal. Finally, the traditional software filtering algorithm is single and has poor accuracy. It cannot effectively present the true value and cannot meet the requirements of the high-voltage cable insulation monitoring device for signal extraction accuracy. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a method for extracting an effective DC signal from a large number of noise points and ripples. The method combines a high-order active low-pass filter circuit and an advanced software filtering algorithm to effectively extract the DC effective component in the signal and improve the accuracy of insulation monitoring.

[0007] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0008] A method for extracting an effective DC signal from a large amount of noise and ripples comprises the following steps:

[0009] Hardware filtering step: using a 4th-order active low-pass filter circuit to pre-process the signal output by the high-voltage cable insulation status monitoring current sensor, wherein the 4th-order active low-pass filter circuit is composed of two 2nd-order Butterworth filter networks in cascade;

[0010] Software filtering steps:

[0011] Noise reduction processing sub-step: using a wavelet threshold denoising algorithm to perform a signal-to-noise separation operation on the output signal of the hardware filtering step, and dividing it into a first wavelet coefficient set and a second wavelet coefficient set according to a predetermined threshold;

[0012] Median filtering sub-step: Perform sliding median filtering on the output signal of the noise reduction processing sub-step to suppress Gaussian noise and ultra-high and ultra-low noise in the signal;

[0013] Recursive average filtering sub-step: performing recursive average filtering on the output signal of the median filtering sub-step;

[0014] Value determination sub-step: Through multiple iterative processing, the similarity of the data is analyzed and compared to obtain the final extracted effective DC signal.

[0015] According to a method for extracting effective DC signals from a large number of noise points and ripples provided by the present invention, the first second-order Butterworth filter network includes an input end, a first resistor series structure, a first capacitor matrix, a second capacitor matrix, a first operational amplifier and a first output end, the first resistor series structure includes resistors R1, R2, R3 and R4 connected in series in sequence, the first end of the first capacitor matrix is ​​connected between the resistor R2 and the resistor R3, the second end of the first capacitor matrix is ​​connected to the inverting input end and the first output end of the first operational amplifier, the first end of the second capacitor matrix is ​​connected between the resistor R4 and the non-inverting input end of the first operational amplifier, the second end of the second capacitor matrix is ​​grounded, and the output end of the first operational amplifier is connected to the first output end.

[0016] According to a method for extracting effective DC signals from a large number of noise points and ripples provided by the present invention, the second second-order Butterworth filter network includes a second resistor series structure, a third capacitor matrix, a fourth capacitor matrix, a second operational amplifier and a second output end, the second resistor series structure includes resistors R5, R6, R7 and R8 connected in series in sequence, the first end of the third capacitor matrix is ​​connected between the resistor R6 and the resistor R7, the second end of the third capacitor matrix is ​​connected to the inverting input end and the second output end of the second operational amplifier, the first end of the fourth capacitor matrix is ​​connected between the resistor R8 and the non-inverting input end of the second operational amplifier, the second end of the fourth capacitor matrix is ​​grounded, and the output end of the second operational amplifier is connected to the second output end.

[0017] According to a method for extracting an effective DC signal from a large number of noise points and ripples provided by the present invention, the cutoff frequency calculation formula of the second-order Butterworth filter network is:

[0018]

[0019] The cutoff frequency is set to 10 Hz.

[0020] According to a method for extracting an effective DC signal from a large number of noise points and ripples provided by the present invention, the wavelet threshold denoising algorithm specifically includes the following steps:

[0021] The steps of wavelet decomposition of the signal are as follows: select a wavelet basis type and a wavelet decomposition level N, perform N-layer wavelet decomposition on the noisy signal, and obtain the wavelet coefficients of each layer;

[0022] High-frequency coefficient threshold quantization step: The high-frequency wavelet coefficients from the first layer to the Nth layer obtained by decomposition are denoised by using a quantization method, and the noise components in the high-frequency coefficients are removed by setting a threshold while retaining the useful information in the signal;

[0023] The wavelet reconstruction step of the signal: combine the high-frequency wavelet coefficients after threshold quantization processing with the low-frequency wavelet coefficients, reconstruct the signal using the inverse wavelet transform, and obtain the denoised signal output.

[0024] According to a method for extracting effective DC signals from a large number of noise points and ripples provided by the present invention, the wavelet decomposition and the inverse wavelet transform are both implemented based on the Haar wavelet basis, and the mother wavelet of the Haar wavelet basis is expressed as the following formula:

[0025]

[0026] Among them, the corresponding scaling equation is expressed as the following formula:

[0027]

[0028] According to a method for extracting an effective DC signal from a large number of noise points and ripples provided by the present invention, the sliding median filtering specifically comprises the following steps:

[0029] Continuously sample the signal after wavelet threshold denoising to obtain a queue consisting of N data points, where N is an integer greater than 1 and is set to 400;

[0030] In the queue, XMax maximum values ​​and XMin minimum values ​​are removed, where XMax and XMin are both integers less than N / 2 and are equal, specifically set to 3 / 8*N, i.e., 150, to eliminate pulse interference;

[0031] Calculate the arithmetic mean of the remaining N-XMax-XMin data points to obtain the filtered signal output.

[0032] According to a method for extracting an effective DC signal from a large number of noise points and ripples provided by the present invention, the recursive average filtering specifically includes the following steps:

[0033] The signal after sliding median filtering is continuously sampled, and the N consecutive sample values ​​are regarded as a queue of fixed length, where N is set to 400;

[0034] Each time a new data is sampled, the new data is put at the end of the queue, and the oldest data at the head of the queue is removed at the same time, keeping the queue length always N. This process follows the first-in-first-out principle;

[0035] Perform arithmetic mean calculation on the N data points in the queue to obtain a new filtering result and output it;

[0036] Among them, the smoothing filtering processing of the signal is achieved by continuously recursively updating the queue and calculating the arithmetic mean.

[0037] According to a method for extracting an effective DC signal from a large number of noise points and ripples provided by the present invention, the value determination sub-step specifically includes:

[0038] A sampling cycle is set, and in each sampling cycle, the signal processed by the filtering step is sampled and calculated, and a method of multiple iterative sampling is used to improve the accuracy; wherein the sampling process of three sampling cycles is performed continuously, and after each sampling, the three values ​​obtained are compared with each other;

[0039] A deviation accuracy threshold is set. If the deviations between the results of three times are within the deviation accuracy threshold, the data are considered valid and one of the values ​​is selected as the final output value.

[0040] If the deviation between any two of the three sampling results exceeds the deviation accuracy threshold, the data is judged to be invalid and the sampling and comparison process needs to be repeated for three sampling cycles until valid data that meets the deviation accuracy requirements is obtained.

[0041] According to a method for extracting effective DC signals from a large number of noise points and ripples provided by the present invention, the high-frequency wavelet coefficients of the first to Nth layers obtained by decomposition are denoised by using a hard threshold quantization method. It includes:

[0042] A hard threshold T is set, which is determined according to the noise level of the signal and the desired signal preservation degree;

[0043] The high-frequency wavelet coefficients of each layer are judged one by one. If the absolute value of a wavelet coefficient is greater than or equal to the threshold T, the coefficient is retained; if the absolute value of a wavelet coefficient is less than the threshold T, the coefficient is set to zero;

[0044] Through the above hard threshold quantization process, the noise components in the high-frequency wavelet coefficients are removed, while retaining local features such as signal edges;

[0045] The high-frequency wavelet coefficients after hard threshold quantization are reconstructed to obtain the denoised signal.

[0046] It can be seen that compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. The present invention adopts a high-order active low-pass filter circuit to effectively filter out a specific frequency point or frequencies other than the frequency point, thereby significantly reducing irrelevant noise in the waveform and improving the purity of the signal.

[0048] 2. The filter circuit of the present invention is reasonably designed and can achieve a higher attenuation, further improving the signal-to-noise ratio of the signal and providing a solid foundation for subsequent signal processing.

[0049] 3. The present invention can accurately distinguish signals from noise and achieve effective signal-noise separation by introducing a wavelet threshold denoising algorithm. This feature makes it possible to extract effective DC signals in complex noise environments.

[0050] 3. The present invention combines the two filtering technologies of "median filtering method" and "arithmetic mean filtering method", complementing each other's advantages and further improving the smoothness and stability of the signal. This fusion filtering method can not only effectively suppress occasional pulse interference, but also have a good inhibitory effect on periodic interference.

[0051] 4. For occasional pulse interference, the present invention can effectively eliminate the sampling value deviation caused by it through the median filtering method, thereby ensuring the accuracy and reliability of the signal.

[0052] 5. By performing multi-level filtering on the signal, the final output signal of the present invention has a high degree of smoothness, which is particularly suitable for high-frequency oscillation systems and provides a strong guarantee for the stable operation of the system.

[0053] 6. Since the present invention has significant advantages in extracting effective DC signals, it can be widely used in online monitoring and fault diagnosis of power systems and electrical equipment, providing technical support for the safe and stable operation of power systems, and has extremely high practical value and socio-economic value.

[0054] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flowchart of an embodiment of a method for extracting an effective DC signal from a large number of noise points and ripples of the present invention.

[0056] Figure 2 It is a circuit schematic diagram of a 4th-order active low-pass filter circuit in an embodiment of a method for extracting an effective DC signal from a large number of noise points and ripples of the present invention.

[0057] Figure 3 It is a schematic diagram of generating a DC quantity in an embodiment of a method for extracting an effective DC signal from a large number of noise points and ripples of the present invention.

[0058] Figure 4 It is a schematic diagram of adding Gaussian noise to a DC quantity in an embodiment of a method for extracting an effective DC signal from a large number of noise points and ripples of the present invention.

[0059] Figure 5 It is a schematic diagram of performing wavelet denoising, median filtering, and recursive average filtering on target noise data in an embodiment of a method for extracting an effective DC signal from a large number of noise points and ripples of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0062] See also Figure 1 This embodiment provides a method for extracting an effective DC signal from a large number of noise points and ripples, the method comprising the following steps:

[0063] Hardware filtering steps: A 4th-order active low-pass filter circuit is used to pre-process the signal output by the high-voltage cable insulation status monitoring current sensor. The 4th-order active low-pass filter circuit is composed of two cascaded second-order Butterworth filter networks;

[0064] Software filtering steps:

[0065] Noise reduction processing sub-step: using a wavelet threshold denoising algorithm to perform a signal-noise separation operation on the output signal of the hardware filtering step, and dividing it into a first wavelet coefficient set and a second wavelet coefficient set according to a predetermined threshold;

[0066] Median filtering sub-step: Perform sliding median filtering on the output signal of the noise reduction processing sub-step to suppress Gaussian noise and ultra-high and ultra-low noise in the signal;

[0067] Recursive average filtering sub-step: performing recursive average filtering on the output signal of the median filtering sub-step;

[0068] Value determination sub-step: Through multiple iterative processing, the similarity of the data is analyzed and compared to obtain the final extracted effective DC signal.

[0069] In this embodiment, the hardware adopts a 4th order active low pass filter circuit, which is usually used to effectively filter out the frequency point of a specific frequency or the frequency other than the frequency point. In order to achieve a high attenuation, or in other words, the attenuation needs to rise quickly, which requires a combination of multiple orders of filters. Taking Butterworth as an example, an odd-numbered high-order low pass filter circuit is a cascade of a first-order Butterworth low pass filter circuit and several second-order Butterworth filter networks, and an even-numbered high-order low pass filter circuit is a cascade of several second-order Butterworth filter networks.

[0070] Among them, the signal VIN output by the high-voltage cable insulation status monitoring current sensor is as follows: Figure 2The upper part is connected to the input of the 4th-order active low-pass filter circuit. This circuit is a cascade of two second-order Butterworth filter networks with a cutoff frequency of 10Hz. After the signal VIN is filtered by the hardware filter, the output signal OUT2 is as follows: Figure 2 In the lower part, the waveform contains a small amount of low-frequency noise, AC component, and DC effective component.

[0071] Specifically, Figure 2 As shown, the first second-order Butterworth filter network includes an input end, a first resistor series structure, a first capacitor matrix, a second capacitor matrix, a first operational amplifier and a first output end. The first resistor series structure includes resistors R1, R2, R3 and R4 connected in series in sequence. The first end of the first capacitor matrix is ​​connected between resistors R2 and R3, the second end of the first capacitor matrix is ​​connected to the inverting input end and the first output end of the first operational amplifier, the first end of the second capacitor matrix is ​​connected between resistor R4 and the non-inverting input end of the first operational amplifier, the second end of the second capacitor matrix is ​​grounded, and the output end of the first operational amplifier is connected to the first output end.

[0072] The second second-order Butterworth filter network includes a second resistor series structure, a third capacitor matrix, a fourth capacitor matrix, a second operational amplifier and a second output end. The second resistor series structure includes resistors R5, R6, R7 and R8 connected in series in sequence. The first end of the third capacitor matrix is ​​connected between resistors R6 and R7, the second end of the third capacitor matrix is ​​connected to the inverting input end and the second output end of the second operational amplifier, the first end of the fourth capacitor matrix is ​​connected between resistor R8 and the non-inverting input end of the second operational amplifier, the second end of the fourth capacitor matrix is ​​grounded, and the output end of the second operational amplifier is connected to the second output end.

[0073] Among them, the first second-order Butterworth filter network R1 = 51K, C1 = 330nF, R2 = 51K, C2 = 288nF; the second second-order Butterworth filter network R5 = 51K, C5 = 820nF, R6 = 51K, C6 = 108.2nF.

[0074] In this embodiment, the cutoff frequency calculation formula of the second-order Butterworth filter network is:

[0075]

[0076] Among them, according to the formula and actual tests, it can be concluded that the cutoff frequency is 10Hz.

[0077] In this embodiment, the wavelet threshold denoising algorithm specifically includes the following steps:

[0078] The steps of wavelet decomposition of the signal are as follows: select a wavelet basis type (DB4, DB3, ECHO, Haar, etc.) and the level N of wavelet decomposition, perform N-layer wavelet decomposition on the noisy signal, and obtain the wavelet coefficients of each layer.

[0079] High-frequency coefficient threshold quantization step: The high-frequency wavelet coefficients from the first layer to the Nth layer obtained by decomposition are denoised by quantization method, and the noise components in the high-frequency coefficients are removed by setting the threshold, while retaining the useful information in the signal. Among them, the quantization processing methods mainly include hard threshold quantization and soft threshold quantization. The hard threshold method can well retain local features such as signal edges, while the soft threshold processing is relatively smooth, but it will cause distortion such as edge blur.

[0080] The wavelet reconstruction step of the signal: combine the high-frequency wavelet coefficients after threshold quantization processing with the low-frequency wavelet coefficients, reconstruct the signal using the inverse wavelet transform, and obtain the denoised signal output.

[0081] In this embodiment, wavelet decomposition and inverse wavelet transform are both implemented based on Haar wavelet basis, and the mother wavelet of Haar wavelet basis is expressed as the following formula:

[0082]

[0083] Among them, the corresponding scaling equation is expressed as the following formula:

[0084]

[0085] It can be seen that the above steps use the "wavelet threshold denoising algorithm". In the brief moment when the signal appears, the wavelet coefficient will have a modulus maximum, and it will increase with the increase of the decomposition scale and reach a peak. White noise has negative singularity, and its wavelet coefficient maximum and density will decrease with the increase of the decomposition scale. Using this diametrically opposite characteristic, signal-to-noise separation can be performed through wavelet transform. The wavelet coefficients obtained by wavelet transform of the signal contain important time-frequency information. The wavelet coefficients of the real signal are larger, and the wavelet coefficients of the noise are smaller. Select a suitable threshold, retain the larger coefficients, and set the smaller ones to 0, so that the signal can be retained and the noise can be filtered out.

[0086] In this embodiment, the sliding median filtering specifically includes the following steps:

[0087] The signal after wavelet threshold denoising is continuously sampled to obtain a queue of N data points, where N is an integer greater than 1 and is set to 400; in the queue, XMax maximum values ​​and XMin minimum values ​​are removed, and XMax and XMin are both integers less than N / 2 and are equal, specifically set to 3 / 8*N, i.e., 150, to eliminate pulse interference; the arithmetic mean of the remaining N-XMax-XMin data points is calculated to obtain the filtered signal output.

[0088] Specifically, a group of queues are used to remove the maximum and minimum values ​​and then take the average value, which is equivalent to the "median filter method" + "arithmetic mean filter method". Continuously sample N (currently 400 points) data, remove part of the maximum value XMax (currently 3 / 8*N=150) and part of the minimum value XMin (currently 3 / 8*N=150), and then calculate the arithmetic mean of N-XMax-XMin data, among which the middle 100 points are selected, and 150 points are discarded before and after, which is the 3 / 8*N described above.

[0089] In this embodiment, the recursive averaging filtering specifically includes the following steps:

[0090] The signal after sliding median filtering is sampled continuously, and the N continuously obtained sampling values ​​are regarded as a queue of fixed length, where N is set to 400; each time a new data is sampled, the new data is put into the tail of the queue, and the oldest data at the head of the queue is removed at the same time, so that the queue length is always N. This process follows the first-in-first-out principle; the arithmetic mean operation is performed on the N data points in the queue to obtain a new filtering result and output it; wherein, the smoothing filtering processing of the signal is realized by continuously recursively updating the queue and calculating the arithmetic mean.

[0091] It can be seen that the N (currently 400 points) continuously obtained sampling values ​​are regarded as a queue. The length of the queue is fixed at N. Each time a new data is sampled, it is put into the end of the queue, and the original data at the head of the queue is discarded (first-in-first-out principle). The N data in the queue are arithmetic averaged to obtain a new filtering result.

[0092] In this embodiment, the value determination sub-step specifically includes:

[0093] A sampling cycle is set. In each sampling cycle, the signal processed by the filtering step is sampled and calculated, and the method of multiple iterative sampling is used to improve the accuracy; wherein, the sampling process of three sampling cycles is carried out continuously, and after each sampling, the three values ​​obtained are compared with each other; a deviation accuracy threshold is set. If the deviations between the results of the three samplings are all within the deviation accuracy threshold, the data are considered to be valid, and one of the values ​​is selected as the final output value; if the deviations between any two of the results of the three samplings exceed the deviation accuracy threshold, the data is determined to be invalid, and the sampling and comparison process of three sampling cycles needs to be repeated until valid data that meets the deviation accuracy requirements is obtained.

[0094] Specifically, in the value determination sub-step, a relatively accurate value can be taken every 40 seconds. However, since the reference value is constantly changing, it is basically irregular. However, the good thing is that it does not change frequently in a short period of time, and occasionally there will be jumps. Within the 40 seconds of calculation, the data may jump, and at this time, multiple 40-second data comparisons are required. The current algorithm uses three 40-second data comparisons. If the three result deviations are within the normal range, the data is considered valid. The deviation accuracy is 50uA. If it exceeds the range, the value is recalculated.

[0095] In this embodiment, the high-frequency wavelet coefficients of the first to Nth layers obtained by decomposition are denoised using a hard threshold quantization method, including:

[0096] A hard threshold T is set (the current algorithm uses a hard threshold of 0.2), which is determined according to the noise level of the signal and the desired degree of signal retention; the high-frequency wavelet coefficients of each layer are judged one by one, and if the absolute value of a wavelet coefficient is greater than or equal to the threshold T, the coefficient is retained; if the absolute value of a wavelet coefficient is less than the threshold T, the coefficient is set to zero; through the above-mentioned hard threshold quantization processing, the noise component in the high-frequency wavelet coefficients is removed, while retaining local features such as signal edges; the high-frequency wavelet coefficients after hard threshold quantization processing are reconstructed to obtain a denoised signal.

[0097] In actual testing, this embodiment simulates three stages of DC flow, namely, DC flow of 100, 200, and 300, adds Gaussian noise points, ultra-high noise points, and ultra-low noise points to obtain contaminated data, and then uses the above preparation algorithm to restore the original signal as follows:

[0098] 1. First generate 100-200-300 DC, such as Figure 3 Shown

[0099] 2. Add Gaussian noise with an amplitude of 100 to the DC flow, such as Figure 4 shown.

[0100] 3. Perform wavelet denoising + median filtering + sliding average filtering on the target noise data, such as Figure 5 shown.

[0101] Conclusion: The final algorithm completion range of 100 true values ​​is 100.9-105.9 with an error within 6%.

[0102] In practical applications, the algorithm of this embodiment is transplanted to a hardware board to test the output signal of the current sensor for monitoring the insulation status of a high-voltage cable. The transplanted algorithm is used to analyze the extracted features to evaluate the insulation status. The hardware is debugged using debugging tools (such as JTAG, SWD, etc.) to ensure that the algorithm is executed correctly and the sensor data is read correctly. Test cases are written to verify the correctness of the algorithm under different input conditions. The algorithm is integrated with the sensor and the hardware board for overall testing. The algorithm and code are optimized according to the evaluation results to improve performance and efficiency. The optimized algorithm and hardware are deployed in the actual application environment. The running status of the algorithm and the output data of the sensor are continuously monitored to ensure stable operation of the system.

[0103] In summary, the present invention can effectively filter out a specific frequency point or frequencies other than the specific frequency point by adopting a high-order active low-pass filter circuit, thereby significantly reducing irrelevant noise in the waveform and improving the purity of the signal.

[0104] The filter circuit of the present invention is reasonably designed and can achieve a higher attenuation, further improving the signal-to-noise ratio of the signal, and providing a solid foundation for subsequent signal processing.

[0105] The present invention can accurately distinguish signals from noise and realize effective signal-noise separation by introducing a wavelet threshold denoising algorithm. This feature makes it possible to extract effective DC signals in a complex noise environment.

[0106] The present invention combines the two filtering technologies of "median filtering method" and "arithmetic mean filtering method", complementing each other's advantages and further improving the smoothness and stability of the signal. This fusion filtering method can not only effectively suppress occasional pulse interference, but also have a good inhibitory effect on periodic interference.

[0107] With respect to the occasional pulse interference, the present invention can effectively eliminate the sampling value deviation caused by the occasional pulse interference through the median value filtering method, thereby ensuring the accuracy and reliability of the signal.

[0108] By performing multi-level filtering on the signal, the final output signal of the present invention has a high degree of smoothness, is particularly suitable for a high-frequency oscillation system, and provides a strong guarantee for the stable operation of the system.

[0109] Since the present invention has significant advantages in extracting effective DC signals, it can be widely used in online monitoring and fault diagnosis of power systems and electrical equipment, providing technical support for the safe and stable operation of power systems, and has extremely high practical value and socio-economic value.

[0110] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0111] The above-mentioned embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and substitutions made by technicians in this field on the basis of the present invention shall fall within the scope of protection required by the present invention.

Claims

1. A method for extracting effective DC signals from a large number of noise points and ripples, characterized in that: The following steps are involved: Hardware filtering step: using a 4th-order active low-pass filter circuit to pre-process the signal output by the high-voltage cable insulation status monitoring current sensor, wherein the 4th-order active low-pass filter circuit is composed of two cascaded second-order Butterworth filter networks; Software filtering steps: Noise reduction processing sub-step: using a wavelet threshold denoising algorithm to perform a signal-noise separation operation on the output signal of the hardware filtering step, and dividing it into a first wavelet coefficient set and a second wavelet coefficient set according to a predetermined threshold; Median filtering sub-step: Perform sliding median filtering on the output signal of the noise reduction processing sub-step to suppress Gaussian noise and ultra-high and ultra-low noise in the signal; Recursive average filtering sub-steps: Performing recursive averaging filtering on the output signal of the median filtering sub-step; Value determination sub-step: Through multiple iterative processing, the similarity of the data is analyzed and compared to obtain the final extracted effective DC signal.

2. The method according to claim 1, characterized in that: The first second-order Butterworth filter network includes an input end, a first resistor series structure, a first capacitor matrix, a second capacitor matrix, a first operational amplifier and a first output end. The first resistor series structure includes resistors R1, R2, R3 and R4 connected in series in sequence. The first end of the first capacitor matrix is ​​connected between the resistor R2 and the resistor R3. The second end of the first capacitor matrix is ​​connected to the inverting input end and the first output end of the first operational amplifier. The first end of the second capacitor matrix is ​​connected between the resistor R4 and the non-inverting input end of the first operational amplifier. The second end of the second capacitor matrix is ​​grounded. The output end of the first operational amplifier is connected to the first output end.

3. The method according to claim 1, characterized in that: The second second-order Butterworth filter network includes a second resistor series structure, a third capacitor matrix, a fourth capacitor matrix, a second operational amplifier and a second output end. The second resistor series structure includes resistors R5, R6, R7 and R8 connected in series in sequence. The first end of the third capacitor matrix is ​​connected between the resistor R6 and the resistor R7, the second end of the third capacitor matrix is ​​connected to the inverting input end and the second output end of the second operational amplifier, the first end of the fourth capacitor matrix is ​​connected between the resistor R8 and the non-inverting input end of the second operational amplifier, the second end of the fourth capacitor matrix is ​​grounded, and the output end of the second operational amplifier is connected to the second output end.

4. The method according to claim 2, characterized in that: The cutoff frequency calculation formula of the second-order Butterworth filter network is: The cutoff frequency is set to 10 Hz.

5. The method according to claim 1, characterized in that: The wavelet threshold denoising algorithm specifically includes the following steps: The steps of wavelet decomposition of the signal are as follows: select a wavelet basis type and a wavelet decomposition level N, perform N-layer wavelet decomposition on the noisy signal, and obtain the wavelet coefficients of each layer; High-frequency coefficient threshold quantization step: The high-frequency wavelet coefficients from the first layer to the Nth layer obtained by decomposition are denoised by using a quantization method, and the noise components in the high-frequency coefficients are removed by setting a threshold while retaining the useful information in the signal; The wavelet reconstruction step of the signal: combine the high-frequency wavelet coefficients after threshold quantization processing with the low-frequency wavelet coefficients, reconstruct the signal using the inverse wavelet transform, and obtain the denoised signal output.

6. The method according to claim 5, characterized in that: The wavelet decomposition and inverse wavelet transform are both implemented based on the Haar wavelet basis. The mother wavelet of the Haar wavelet basis is expressed as the following formula: Among them, the corresponding scaling equation is expressed as the following formula:

7. The method according to claim 1, characterized in that: The sliding median filtering specifically comprises the following steps: Continuously sample the signal after wavelet threshold denoising to obtain a queue consisting of N data points, where N is an integer greater than 1 and is set to 400; In the queue, XMax maximum values ​​and XMin minimum values ​​are removed, where XMax and XMin are both integers less than N / 2 and are equal, specifically set to 3 / 8*N, i.e., 150, to eliminate pulse interference; Calculate the arithmetic mean of the remaining N-XMax-XMin data points to obtain the filtered signal output.

8. The method according to claim 1, characterized in that: The recursive averaging filtering specifically comprises the following steps: The signal after sliding median filtering is continuously sampled, and the N consecutive sample values ​​are regarded as a queue of fixed length, where N is set to 400; Each time a new data is sampled, the new data is put at the end of the queue, and the oldest data at the head of the queue is removed at the same time, keeping the queue length always N. This process follows the first-in-first-out principle; Perform arithmetic mean calculation on the N data points in the queue to obtain a new filtering result and output it; Among them, the smoothing filtering processing of the signal is achieved by continuously recursively updating the queue and calculating the arithmetic mean.

9. The method according to any one of claims 1 to 8, characterized in that: The value determination sub-step specifically includes: A sampling cycle is set, and in each sampling cycle, the signal processed by the filtering step is sampled and calculated, and a method of multiple iterative sampling is used to improve the accuracy; wherein the sampling process of three sampling cycles is performed continuously, and after each sampling, the three values ​​obtained are compared with each other; A deviation accuracy threshold is set. If the deviations between the results of three times are within the deviation accuracy threshold, the data are considered valid and one of the values ​​is selected as the final output value. If the deviation between any two of the three sampling results exceeds the deviation accuracy threshold, the data is judged to be invalid and the sampling and comparison process needs to be repeated for three sampling cycles until valid data that meets the deviation accuracy requirements is obtained.

10. The method according to claim 5, characterized in that: The high-frequency wavelet coefficients from the first layer to the Nth layer obtained by decomposition are denoised using a hard threshold quantization method, including: A hard threshold T is set, which is determined according to the noise level of the signal and the desired signal preservation degree; The high-frequency wavelet coefficients of each layer are judged one by one. If the absolute value of a wavelet coefficient is greater than or equal to the threshold T, the coefficient is retained; if the absolute value of a wavelet coefficient is less than the threshold T, the coefficient is set to zero; Through the above hard threshold quantization process, the noise components in the high-frequency wavelet coefficients are removed, while retaining local features such as signal edges; The high-frequency wavelet coefficients after hard threshold quantization are reconstructed to obtain the denoised signal.

Images