Short-circuit current zero crossing point prediction method based on least square method

Abstract

The invention relates to the technical field of circuit breaker on-off, in particular to a short-circuit current zero crossing point prediction method based on a least square method, and aims to solve the problems that short-circuit current waveforms cannot be decomposed in a self-adaptive mode, model adaptability is reduced, accurate parameter estimation and rapid convergence cannot be achieved, and short-circuit current zero crossing points cannot be predicted in the prior art. The waveform description capability and the zero crossing point prediction reliability are reduced; the EMD algorithm is adopted to adaptively decompose the short-circuit current waveform, a preset primary function is avoided, the model adaptability is enhanced, sine function superposition is used for high frequency, a polynomial model is used for low frequency, wavelet function linear combination is used for intermediate frequency, the least square method and the Levenberg-Marquardt algorithm are combined, accurate parameter estimation and rapid convergence are achieved, and the final model is high in fitting precision and high in precision. Each IMF component has clear physical significance, and the waveform description capability and the zero crossing point prediction reliability are significantly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of breaker opening, and more specifically, to a short-circuit current zero-crossing prediction method based on the least square method. Background Art

[0002] In a power system, when a breaker receives a tripping command to interrupt a short-circuit current, a traditional breaker will act immediately. However, if the breaker performs a tripping operation at the peak moment of the short-circuit current, a huge instantaneous breaking current will be generated, resulting in a large arc energy, which will cause serious damage to the breaker contacts. In order to reduce the energy of arc combustion, improve the breaking ability of the breaker and extend its service life, the ideal operation is to complete the tripping before the current crosses zero, which can minimize the time and intensity of arc combustion, thus effectively protecting the breaker contacts. Therefore, it is crucial to accurately calculate the zero-crossing time of the short-circuit current.

[0003] The patent application with the publication number CN119312588A discloses a correction method for online calculating the zero-crossing time of a short-circuit current based on the least square method, including: changing the DC decay component respectively according to the preset ratio requirements of a low-frequency power system to perform short-circuit fault simulation calculations, and obtaining the first relationship between the zero-crossing time prediction error and the DC decay component ratio; changing the fault initial phase respectively according to the preset phase adjustment requirements of the low-frequency power system to perform short-circuit fault simulation calculations, and obtaining the second relationship between the short-circuit zero-crossing time prediction error and the fault initial phase; obtaining an error compensation formula for predicting the zero-crossing time of the current according to the first relationship and the second relationship; using the least square method to calculate the predicted short-circuit current zero-crossing time and short-circuit fault current parameters of the short-circuit fault, and recording the calculation time, and calculating the compensated current zero-crossing time; compensating the predicted short-circuit current zero-crossing time according to the compensated current zero-crossing time to obtain the corrected short-circuit current zero-crossing time; However, the above reference patent improves the prediction accuracy of the short-circuit current zero-crossing and the algorithm convergence speed by constructing a zero-crossing prediction error compensation formula, solves the problems of low prediction accuracy and long calculation time, meets the actual requirements of phase selection tripping, but cannot realize real-time and accurate short-circuit fault monitoring and identification, cannot early warn potential risks, cannot provide time guarantee for the timely start of protection measures, and at the same time cannot adaptively decompose the short-circuit current waveform, reducing the model adaptability, cannot realize accurate parameter estimation and fast convergence, reducing the waveform description ability and the reliability of zero-crossing prediction.

[0004] Therefore, we propose a short-circuit current zero-crossing prediction method based on the least square method for the above problems. Summary of the Invention

[0005] The object of the present invention is to provide a method for predicting the zero-crossing point of short-circuit current based on the least squares method, which solves the problems that the prior art cannot achieve real-time and accurate monitoring and identification of short-circuit faults, cannot give early warnings of potential risks, cannot provide time guarantee for the timely activation of protection measures, and at the same time cannot adaptively decompose the short-circuit current waveform, reducing the adaptability of the model, unable to achieve accurate parameter estimation and rapid convergence, reducing the waveform description ability and the reliability of zero-crossing point prediction.

[0006] The object of the present invention is achieved by the following technical solutions: A method for predicting the zero-crossing point of short-circuit current based on the least squares method, comprising the following steps: Step 1: Collect the short-circuit current waveform signal through a current transformer, and perform preprocessing operations on the collected short-circuit current waveform signal; Step 2: Monitor the short-circuit evaluation parameters of the power system in real time, and monitor and identify the short-circuit faults of the power system; Step 3: Decompose the preprocessed short-circuit current waveform signal, and use the least squares method to construct a mathematical model to fit the short-circuit current waveform; Step 4: Based on the model obtained by fitting with the least squares method, predict the zero-crossing time of the short-circuit current; Step 5: Generate corresponding control instructions in advance according to the predicted zero-crossing time and the inherent opening time of the circuit breaker.

[0007] As a preferred embodiment of the present invention, the specific process of monitoring and identifying the short-circuit faults of the power system in step 2 is as follows: Obtain the historical short-circuit evaluation parameters of the power system. The short-circuit evaluation parameters include operating current, operating voltage, and operating frequency, generate a monitoring period, and equally divide the monitoring period into multiple monitoring time periods; Obtain the current mutation rate of the power system in multiple monitoring time periods, and construct a set A of current mutation rates, and record the mean value of the difference between the largest subset and the smallest subset in set A as the current mutation rate difference DTC; Obtain the voltage drop rate of the power system in multiple monitoring time periods, and construct a set B of voltage drop rates, and record the mean value of the difference between the largest subset and the smallest subset in set B as the voltage drop rate difference DDC; Obtain the frequency change rate of the power system in multiple monitoring time periods, and construct a set C of frequency change rates, and record the mean value of the difference between the largest subset and the smallest subset in set C as the frequency change rate difference PBC.

[0008] As a preferred embodiment of the present invention, the current mutation rate difference DTC, the voltage dip rate difference DDC, and the frequency change rate difference PBC are obtained, and the current mutation rate difference DTC, the voltage dip rate difference DDC, and the frequency change rate difference PBC are combined to construct a short-circuit monitoring feature matrix DJ; The constructed short-circuit monitoring feature matrix DJ is used as the input of the machine learning model, and the label vector v is used as the output of the machine learning model. The label vector v indicates whether a short circuit has occurred in the power system. The output of the label vector v is 0 or 1. 0 indicates that no short circuit has occurred in the power system, and 1 indicates that a short circuit has occurred in the power system. Taking the label vector v as the prediction target and minimizing the sum of the prediction errors of all training data as the training target, the machine learning model is trained until the sum of the prediction errors converges and then the training is stopped, obtaining a machine learning model that predicts the label vector v; Obtain the real-time short-circuit evaluation parameters of the power system, process them to construct a real-time short-circuit monitoring feature matrix SJ, and identify whether a short-circuit fault has occurred in the power system through the trained machine learning model; If the label vector v output by the machine learning model is 0, it indicates that no short circuit has occurred in the power system; If the label vector v output by the machine learning model is 1, it indicates that a short circuit has occurred in the power system.

[0009] As a preferred embodiment of the present invention, the specific process of constructing a mathematical model by the least squares method to fit the short-circuit current waveform in step three is as follows: After detecting that a short circuit has occurred in the power system, immediately obtain the preprocessed short-circuit current waveform signal x(t), and use the EMD algorithm to decompose the preprocessed short-circuit current waveform signal; The specific steps of the EMD algorithm are as follows: S1: Find all local maximum and minimum values in the signal; S2: Connect the local maximum and minimum values respectively using the cubic spline interpolation method to obtain the upper envelope and the lower envelope; S3: Calculate the average value of the upper and lower envelopes as the average envelope m 1 (t); S4: Subtract the average envelope from the original signal to obtain the first IMF: h 1 (t)=x(t)-m 1 (t). If h 1 (t) meets the conditions of the IMF, then it is used as the first IMF. Otherwise, h 1 (t) is used as the new original signal, and S1 to S4 are repeated until the IMF conditions are met; S5: Subtract the first IMF from the original signal to obtain the residual r 1(t) = x(t) - h 1 (t); S6: Take the residual r 1 (t) as the new original signal, repeat S1 to S5 until the residual meets the stop condition, obtaining multiple IMFs and the final residual.

[0010] As a preferred embodiment of the present invention, the result of EMD decomposition is represented by the following formula: ; where IMF i (t) is the i-th IMF, representing signals of different frequency components, and r n (t) is the final residual, representing the trend or low-frequency component of the signal, and n is the number of IMFs; After decomposition using the EMD algorithm, a series of IMF components IMF 1 (t), IMF 2 (t), …, IMF n (t) and a residual term r n (t) are obtained. To fit these components using the least squares method, a suitable mathematical model should be selected according to the frequency characteristics and morphology of the IMF components.

[0011] As a preferred embodiment of the present invention, the mathematical model is selected as follows: High-frequency IMFs: These components contain the high-frequency part of the signal and are fitted using the superposition of sine functions: ; where A ik is the amplitude of the k-th sine wave in the i-th IMF, f ik is the frequency of the k-th sine wave in the i-th IMF, φ ik is the phase of the k-th sine wave in the i-th IMF, and K i is the number of sine waves used to fit the i-th IMF; Low-frequency IMFs and residuals: These components contain the low-frequency components or trends of the signal and are fitted using polynomial functions: ; where a k are the polynomial coefficients and P is the order of the polynomial; Medium-frequency IMFs: These components contain the intermediate-frequency components of the signal, which are between the high-frequency and low-frequency components, and are fitted using a linear combination of wavelet functions: ; where ψ jk (t) are wavelet basis functions of different scales (j) and positions (k), and c k are the wavelet coefficients.

[0012] As a preferred embodiment of the present invention, for each selected model, the least squares method is used to estimate the model parameters. When using the sine model to fit the high-frequency IMF, the least squares objective function is: ; where N is the number of data points, and t j is the time of the j-th data point; is the parameter vector of the i-th IMF model; The Levenberg-Marquardt algorithm is used to iteratively update the parameter vector θ i , until convergence; When using the polynomial model to fit the low-frequency IMF, the least squares objective function is: ; Let X be the design matrix, whose elements are , and y be the vector representation of the IMF i (t). Then the least squares solution of the parameter vector a = [a 0 , a 1 , … a p T is: a = (X T X) -1 X T y.

[0013] As a preferred embodiment of the present invention, the residual term represents the trend or low-frequency component of the signal. Therefore, a model similar to that of the low-frequency IMF can be used for fitting. The polynomial fitting process of the low-frequency IMF is the same, except that the IMF i (t) is replaced by r n (t); When using the wavelet model to fit the medium-frequency IMF, the least squares objective function is: ; This objective function is a quadratic function of the wavelet coefficients c. Therefore, an analytical solution can be obtained: c = (Ψ T Ψ) -1 Ψ T y; where Ψ is the design matrix composed of wavelet basis functions, and y is the vector representation of the IMF i (t); Combining the fitting models of all IMF components and the fitting model of the residual term, the final short-circuit current waveform model is obtained: .

[0014] As a preferred embodiment of the present invention, the specific process of predicting the zero-crossing time of the short-circuit current in step four is as follows: Obtain the final short-circuit current waveform model obtained by fitting based on the least squares method. When predicting the zero-crossing time of the short-circuit current, the t value that satisfies x(t) = 0 needs to be found, and the Newton-Raphson method is used for solution. The solution steps of the Newton-Raphson method are as follows: T1: Select an initial guess value t 0 ; T2: Repeat the following steps until the accuracy requirement is met: Calculate , where is the derivative of x(t); T3: The final t n is the approximate value of the zero-crossing time.

[0015] As a preferred embodiment of the present invention, the specific process of generating the corresponding control instruction in advance in step five is as follows: Obtain the predicted zero-crossing time t of the short-circuit current n , obtain the inherent opening time t of the circuit breaker f . In order to safely open near the zero-crossing point, the inherent opening time t of the circuit breaker needs to be considered in advance f , and the calculation of the control instruction issuance time t k is as follows: t k = t n - t f ; After calculating the control instruction issuance time t k , issue the corresponding control instruction before the time of t k .

[0016] Compared with the prior art, the advantages of the present invention are as follows: (1) In the present invention, by analyzing multiple parameters to construct a feature matrix and combining with a machine learning model, real-time and accurate short-circuit fault monitoring and identification are realized, which has the advantages of high automation degree, strong scalability, strong early warning ability, etc., significantly improves the accuracy and efficiency of short-circuit fault identification, and can early warn of potential risks, providing time guarantee for the timely initiation of protection measures; (2) In the present invention, the EMD algorithm is used to adaptively decompose the short-circuit current waveform, avoiding preset basis functions, enhancing the model adaptability. The high-frequency part is superimposed with sine functions, the low-frequency part is modeled with polynomials, and the intermediate-frequency part is linearly combined with wavelet functions. Combining the least squares method and the Levenberg-Marquardt algorithm, accurate parameter estimation and fast convergence are achieved. Finally, the model has a high fitting accuracy, and each IMF component has a clear physical meaning, significantly improving the waveform description ability and the reliability of zero-crossing prediction. Brief Description of the Drawings

[0017] Figure 1 It is a flowchart of the short - circuit current zero - crossing prediction method in the present invention; Figure 2 It is a step - flowchart of the EMD algorithm in the present invention; Figure 3 It is a step - flowchart of the solution by Newton - Raphson method in the present invention. Detailed Embodiments

[0018] Next, in combination with the drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] Embodiment 1: As shown in Figure 1 and Figure 2 The short - circuit current zero - crossing prediction method based on the least - squares method proposed by the present invention includes the following steps: Step 1: Collect the short - circuit current waveform signal through a current transformer, and perform pre - processing operations on the collected short - circuit current waveform signal. The pre - processing operations include but are not limited to filtering and denoising, baseline correction, outlier detection and processing, data smoothing processing, and normalization processing; Performing filtering and denoising, baseline correction, outlier detection, data smoothing, and normalization processing on the short - circuit current waveform signal can significantly improve the data quality and analysis reliability. These pre - processing steps effectively remove noise, compensate for drift, correct incorrect data, reduce fluctuations, and scale the data to a specific range, thereby improving the zero - crossing detection accuracy and stability, and enhancing the performance and safety of the circuit breaker control system.

[0020] Step 2: Real - time monitor the short - circuit evaluation parameters of the power system to monitor and identify the short - circuit faults of the power system; The specific process of monitoring and identifying the short - circuit faults of the power system in Step 2 is as follows: Obtain the historical short - circuit evaluation parameters of the power system. The short - circuit evaluation parameters include operating current, operating voltage, and operating frequency, generate a monitoring period, and equally divide the monitoring period into multiple monitoring time periods; Obtain the current mutation rate of the power system in multiple monitoring time periods. The current mutation rate refers to the rate of change of current with time, that is, the change amount of current per unit time. Thus, construct a set A of current mutation rates, and record the mean value of the difference between the largest subset and the smallest subset in set A as the current mutation rate difference DTC; Obtain the voltage sag rate of the power system during multiple monitoring periods. The voltage sag rate refers to the rate of change of voltage over time, that is, the change in voltage per unit time. Based on this, construct a set B of voltage sag rates, and denote the mean value of the difference between the largest subset and the smallest subset in set B as the voltage sag rate difference DDC; Obtain the frequency change rate of the power system during multiple monitoring periods. The frequency change rate refers to the rate of change of the system frequency over time, that is, the change in frequency per unit time. Based on this, construct a set C of frequency change rates, and denote the mean value of the difference between the largest subset and the smallest subset in set C as the frequency change rate difference PBC; Obtain the current mutation rate difference DTC, the voltage sag rate difference DDC, and the frequency change rate difference PBC. Combine the current mutation rate difference DTC, the voltage sag rate difference DDC, and the frequency change rate difference PBC to construct a short-circuit monitoring feature matrix DJ; Use the constructed short-circuit monitoring feature matrix DJ as the input of the machine learning model, and use the label vector v as the output of the machine learning model. The label vector v represents whether a short circuit has occurred in the power system. The output of the label vector v is 0 or 1. 0 indicates that no short circuit has occurred in the power system, and 1 indicates that a short circuit has occurred in the power system. Using the label vector v as the prediction target and minimizing the sum of the prediction errors of all training data as the training target, train the machine learning model until the sum of the prediction errors converges and then stop training to obtain a machine learning model that predicts the label vector v; Obtain the real-time short-circuit evaluation parameters of the power system, process them to construct a real-time short-circuit monitoring feature matrix SJ, and use the trained machine learning model to identify whether a short-circuit fault has occurred in the power system; If the label vector v output by the machine learning model is 0, it indicates that no short circuit has occurred in the power system; If the label vector v output by the machine learning model is 1, it indicates that a short circuit has occurred in the power system; Construct a feature matrix through multi-parameter analysis, and combine with a machine learning model to achieve real-time and accurate short-circuit fault monitoring and identification. It has the advantages of high automation, strong scalability, and strong early warning ability, significantly improving the accuracy and efficiency of short-circuit fault identification, and being able to give early warnings of potential risks, providing time guarantee for the timely activation of protection measures.

[0021] Step 3: Decompose the preprocessed short-circuit current waveform signal, and use the least squares method to construct a mathematical model to fit the short-circuit current waveform; The specific process of using the least squares method to construct a mathematical model to fit the short-circuit current waveform in Step 3 is as follows: After detecting a short circuit in the power system, immediately obtain the preprocessed short-circuit current waveform signal x(t), and decompose the preprocessed short-circuit current waveform signal using the EMD algorithm; The EMD algorithm is an adaptive data-driven method that decomposes a signal into a series of components called intrinsic mode functions. Each IMF is a locally stationary signal with a clear physical meaning. The specific steps of the EMD algorithm are as follows: S1: Find all local maxima and minima in the signal; S2: Connect the local maxima and minima using cubic spline interpolation respectively to obtain the upper envelope and the lower envelope; S3: Calculate the average of the upper and lower envelopes as the average envelope m 1 (t); S4: Subtract the average envelope from the original signal to obtain the first IMF: h 1 (t)=x(t)-m 1 (t). If h 1 (t) meets the conditions of the IMF (the number of local extrema is finite and the upper and lower envelopes are symmetric about zero mean), then it is taken as the first IMF. Otherwise, h 1 (t) is taken as the new original signal, and repeat S1 to S4 until the IMF conditions are met; S5: Subtract the first IMF from the original signal to obtain the residue r 1 (t)=x(t)-h 1 (t); S6: Take the residue r 1 (t) as the new original signal, and repeat S1 to S5 until the residue meets the stop condition (such as the energy of the residue is less than the preset threshold or the number of iterations reaches the upper limit), to obtain multiple IMFs and the final residue; The result of the EMD decomposition is expressed by the following formula: ; Where IMF i (t) is the i-th IMF, representing signals of different frequency components, r n (t) is the final residue, representing the trend or low-frequency component of the signal, and n is the number of IMFs; After decomposition using the EMD algorithm, a series of IMF components IMF 1 (t), IMF 2 (t), …, IMF n (t) and a residue term r n (t) are obtained. Use the least squares method to fit these components, and a suitable mathematical model should be selected according to the frequency characteristics and morphology of the IMF components; The mathematical model is selected as follows: High-frequency IMFs: These components contain the high-frequency part of the signal and are fitted using a superposition of sine functions: ; where A ik is the amplitude of the k-th sine wave in the i-th IMF, f ik is the frequency of the k-th sine wave in the i-th IMF, φ ik is the phase of the k-th sine wave in the i-th IMF, K i is the number of sine waves used to fit the i-th IMF; Low-frequency IMFs and residuals: These components contain the low-frequency components or trends of the signal and are fitted using polynomial functions: ; where a k are the polynomial coefficients and P is the order of the polynomial; Mid-frequency IMFs: These components contain the intermediate-frequency components of the signal, which are between the high and low frequencies, and are fitted using a linear combination of wavelet functions: ; where ψ jk (t) are wavelet basis functions at different scales (j) and positions (k), c k are the wavelet coefficients, and the choice of wavelet basis function depends on the characteristics of the IMF; For each selected model, the least squares method is used to estimate the model parameters. When using the sine model to fit the high-frequency IMFs, the least squares objective function is: ; where N is the number of data points, t j is the time of the j-th data point; is the parameter vector of the i-th IMF model; The Levenberg-Marquardt algorithm is used to iteratively update the parameter vector θ i , until convergence. The Levenberg-Marquardt algorithm is a commonly used algorithm in the prior art, and the process of iteratively updating the parameter vector θ i will not be elaborated in detail here; When using the polynomial model to fit the low-frequency IMFs, the least squares objective function is: ; Let X be the design matrix, whose elements are , and y be the vector representation of IMF i (t). Then the parameter vector a = [a 0 , a 1,…a p T The least squares solution of is: a = (X T X) -1 X T y; The residual term represents the trend or low-frequency component of the signal. Therefore, a model similar to that of the low-frequency IMF can be used for fitting. The polynomial fitting process of the low-frequency IMF is the same, except that the IMF i (t) is replaced by r n (t); When using the wavelet model to fit the intermediate-frequency IMF, the least squares objective function is: ; This objective function is a quadratic function of the wavelet coefficients c. Therefore, an analytical solution can be obtained: c = (Ψ T Ψ) -1 Ψ T y; where Ψ is the design matrix composed of wavelet basis functions, and y is the vector representation of the IMF i (t); Combining the fitting models of all IMF components and the fitting model of the residual term, the final short-circuit current waveform model is obtained: ; The EMD algorithm is used to adaptively decompose the short-circuit current waveform, avoiding preset basis functions and enhancing the model adaptability. The high-frequency part is superimposed by sine functions, the low-frequency part is modeled by polynomials, and the intermediate-frequency part is linearly combined by wavelet functions. Combining the least squares method and the Levenberg-Marquardt algorithm, accurate parameter estimation and fast convergence are achieved. The final model has a high fitting accuracy, and each IMF component has a clear physical meaning, significantly improving the waveform description ability and the reliability of zero-crossing prediction.

[0022] Example 2: The technical solution of this embodiment of the present invention is different from that of Embodiment 1 in that: As Figure 1 and Figure 3 shown, Step 4: Predict the zero-crossing time of the short-circuit current based on the model obtained by least squares fitting; The specific process of Step 4 for predicting the zero-crossing time of the short-circuit current is as follows: Obtain the final short-circuit current waveform model obtained by least squares fitting. When predicting the zero-crossing time of the short-circuit current, the t value that satisfies x(t) = 0 needs to be found, and the Newton-Raphson method is used for solution. The solution steps of the Newton-Raphson method are as follows: T1: Select an initial guess value t 0 ; ​T2: Repeat the following steps until the accuracy requirement is met: Calculate , where is the derivative of x(t); T3: The final t n is the approximate value of the zero-crossing time; Use the Newton-Raphson method to iteratively solve for the zero-crossing of the short-circuit current waveform model. This method has the advantages of high accuracy, high efficiency, and wide applicability, and can quickly and accurately predict the zero-crossing time.

[0023] Step Five: Generate corresponding control instructions in advance based on the predicted zero-crossing time and the inherent opening time of the circuit breaker; The specific process of generating corresponding control instructions in advance in Step Five is as follows: Obtain the predicted zero-crossing time t n of the short-circuit current, and obtain the inherent opening time t f of the circuit breaker. To safely open near the zero-crossing, the inherent opening time t f of the circuit breaker needs to be considered in advance. The control instruction issuance time t k is calculated as follows: t k = t n - t f ; If the calculated t k is less than zero, it means that the predicted zero-crossing time has passed. At this time, corresponding countermeasures need to be taken according to the actual situation. The content of the countermeasures is as follows: If the system can predict the time of the next zero-crossing, it can wait for the next zero-crossing to perform the opening operation; If missing the current zero-crossing does not cause serious consequences, it can choose to ignore it and wait for the next zero-crossing; In some emergency situations, forced opening may be required, but this is usually not recommended because it may cause greater impact and potential risks; After calculating the control instruction issuance time t k , issue the corresponding control instruction before the time t k . The specific content of the control instruction is: Clearly specify which circuit breaker needs to perform the operation; Clearly indicate the operation to be performed; Include the predicted zero-crossing time t n and the instruction issuance time t k , t n for internal reference in the circuit breaker control system, and t k determines the issuance time of the instruction; According to actual needs, other parameters may be included, such as the tripping mode, current limit, etc.; By pre-calculating the time when the control instruction is issued and providing multiple coping strategies (such as waiting for the next zero-crossing point or ignoring the current zero-crossing point), it is ensured that the circuit breaker trips safely near the zero-crossing point, thereby improving the system reliability, response speed and safety. The complete control instruction contains all necessary information, avoiding the real-time calculation delay.

[0024] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its improved concept, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A short-circuit current zero-crossing point prediction method based on the least squares method, characterized in that: The following steps are involved: Step 1: collecting a short-circuit current waveform signal through a current transformer, and performing a preprocessing operation on the collected short-circuit current waveform signal; Step 2: Real-time monitoring of short-circuit evaluation parameters of the power system, and monitoring and identifying short-circuit faults of the power system; Step 3: Decompose the preprocessed short-circuit current waveform signal, and use the least square method to build a mathematical model to fit the short-circuit current waveform; Step 4: Based on the model obtained by least squares fitting, predict the zero-crossing time of the short-circuit current; Step 5: Generate corresponding control instructions in advance based on the predicted zero-crossing time and the inherent opening time of the circuit breaker.

2. The short-circuit current zero-crossing point prediction method based on the least squares method according to claim 1 is characterized in that: The specific process of monitoring and identifying short-circuit faults in the power system in step 2 is as follows: Obtain historical short-circuit evaluation parameters of the power system, the short-circuit evaluation parameters including operating current, operating voltage and operating frequency, generate a monitoring cycle, and divide the monitoring cycle into multiple monitoring periods; Obtain the current mutation rate of the power system in multiple monitoring periods, thereby constructing a set A of current mutation rates, and record the average of the difference between the largest subset and the smallest subset in set A as the current mutation rate difference DTC; The voltage drop rate of the power system in multiple monitoring periods is obtained to construct a set B of voltage drop rates, and the average of the difference between the largest subset and the smallest subset in set B is recorded as the voltage drop rate difference DDC; The frequency change rate of the power system in multiple monitoring periods is obtained to construct a set C of frequency change rates, and the average of the difference between the largest subset and the smallest subset in set C is recorded as the frequency change rate difference PBC.

3. The short-circuit current zero-crossing point prediction method based on the least squares method according to claim 2 is characterized in that: The current mutation rate difference DTC, the voltage drop rate difference DDC and the frequency change rate difference PBC are obtained, and the current mutation rate difference DTC, the voltage drop rate difference DDC and the frequency change rate difference PBC are combined to construct a short circuit monitoring feature matrix DJ; The constructed short-circuit monitoring feature matrix DJ is used as the input of the machine learning model, and the label vector v is used as the output of the machine learning model. The label vector v indicates whether a short circuit occurs in the power system. The output of the label vector v is 0 or 1. 0 indicates that a short circuit does not occur in the power system, and 1 indicates that a short circuit occurs in the power system. The label vector v is used as the prediction target, and the sum of the prediction errors of all training data is minimized as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the training is stopped to obtain a machine learning model that predicts the label vector v. Obtain the real-time short-circuit assessment parameters of the power system, process them to construct the real-time short-circuit monitoring feature matrix SJ, and use the trained machine learning model to identify whether a short-circuit fault occurs in the power system; If the label vector v output by the machine learning model is 0, it means that there is no short circuit in the power system; If the label vector v output by the machine learning model is 1, it means that a short circuit has occurred in the power system.

4. The short-circuit current zero-crossing point prediction method based on the least squares method according to claim 1, characterized in that: The specific process of step 3 using the least squares method to construct a mathematical model to fit the short-circuit current waveform is as follows: After a short circuit is detected in the power system, the pre-processed short-circuit current waveform signal x(t) is immediately obtained, and the pre-processed short-circuit current waveform signal is decomposed using the EMD algorithm; The specific steps of the EMD algorithm are as follows: S1: Find all local maxima and minima in the signal; S2: Use cubic spline interpolation to connect the local maximum and minimum values ​​to obtain the upper envelope and lower envelope respectively; S3: Calculate the average value of the upper and lower envelopes as the average envelope m1(t); S4: Subtract the average envelope from the original signal to get the first IMF: h1(t)=x(t)-m1(t). If h1(t) meets the IMF conditions, it is used as the first IMF. Otherwise, h1(t) is used as the new original signal and S1 to S4 are repeated until the IMF conditions are met. S5: Subtract the first IMF from the original signal to obtain the residual r1(t)=x(t)-h1(t); S6: Take the residual r1(t) as the new original signal and repeat S1 to S5 until the residual meets the stopping condition, and obtain multiple IMFs and the final residual.

5. The short-circuit current zero-crossing point prediction method based on least squares method according to claim 4 is characterized in that: The result of EMD decomposition is expressed by the following formula: ； IMF i (t) is the i-th IMF, representing the signal with different frequency components, r n (t) is the final residual, representing the trend or low-frequency component of the signal, and n is the number of IMFs; After decomposition using the EMD algorithm, a series of IMF components IMF1(t), IMF2(t), ..., IMF n (t) and a residual term r n (t), use the least squares method to fit these components, and choose a suitable mathematical model based on the frequency characteristics and shape of the IMF components.

6. The short-circuit current zero-crossing point prediction method based on the least squares method according to claim 5, characterized in that: The mathematical model is selected as follows: High-frequency IMFs: These components contain the high-frequency part of the signal and are fitted using a superposition of sinusoidal functions: ； Among them A ik is the amplitude of the kth sine wave in the ith IMF, f ik is the frequency of the kth sine wave in the ith IMF, φ ik is the phase of the kth sinusoid in the ith IMF, K i is the number of sine waves used to fit the ith IMF; Low-frequency IMFs and residuals: These components contain the low-frequency content or trend of the signal and are fitted using a polynomial function: ； where a k are the polynomial coefficients, P is the order of the polynomial; Intermediate Frequency IMF: These components contain the intermediate frequency components of the signal, which are located between high and low frequencies and are fitted using a linear combination of wavelet functions: ； where ψ jk (t) is the wavelet basis function of different scales (j) and positions (k), c k are the wavelet coefficients.

7. The short-circuit current zero-crossing point prediction method based on least squares method according to claim 6, characterized in that: For each selected model, the least squares method is used to estimate the model parameters. When the sinusoidal model is used to fit the high-frequency IMF, the least squares objective function is: ； Where N is the number of data points, t j is the time of the jth data point; is the parameter vector of the ith IMF model; Iteratively update the parameter vector θ using the Levenberg-Marquardt algorithm i , until convergence; When using a polynomial model to fit low-frequency IMF, the least squares objective function is: ； Let X be the design matrix whose elements are , y is IMF i (t) is a vector representation, then the parameter vector a=[a0,a1,…a p ] T The least squares solution of is: a=(X T X) -1 X T y。 8. The short-circuit current zero-crossing point prediction method based on the least squares method according to claim 7, characterized in that: The residual term represents the trend or low-frequency component of the signal, so it can be fitted using a model similar to the low-frequency IMF. The polynomial fitting process is the same as that of the low-frequency IMF, except that the IMF i (t) is replaced by r n (t); When using the wavelet model to fit the intermediate frequency IMF, the least squares objective function is: ； This objective function is a quadratic function of the wavelet coefficient c, so an analytical solution can be obtained: c=(Ψ T (W) -1 P T y; Where Ψ is the design matrix consisting of wavelet basis functions, and y is the IMF i Vector representation of (t); The fitting models of all IMF components and the fitting models of the residual terms are combined to obtain the final short-circuit current waveform model: 。 9. The short-circuit current zero-crossing point prediction method based on least squares method according to claim 1, characterized in that: The specific process of predicting the zero-crossing time of the short-circuit current in step 4 is as follows: The final short-circuit current waveform model obtained by least squares fitting is obtained. When predicting the zero-crossing time of the short-circuit current, it is necessary to find the t value that satisfies x(t)=0. The Newton-Raphson method is used for solution. The steps of the Newton-Raphson method are as follows: T1: Select an initial guess value t0; T2: Repeat the following steps until the accuracy requirement is met: Calculate ,in is the derivative of x(t); T3: Final t n This is the approximate value of the zero-crossing time.

10. The short-circuit current zero-crossing point prediction method based on the least squares method according to claim 1, characterized in that: The specific process of generating the corresponding control instructions in advance in step 5 is as follows: Get the predicted short-circuit current zero-crossing time t n , obtain the inherent opening time t of the circuit breaker f In order to safely open the circuit breaker near the zero crossing point, the inherent opening time t of the circuit breaker needs to be considered in advance. f , control command issuance time t k The calculation of is as follows: t k =t n -t f ; Calculate the time t when the control command is issued k After that, at t k The corresponding control instructions are issued before the time.

Images