A method for accurate detection of motor current

Abstract

This invention provides a method for accurate detection of motor current, relating to the field of motor control technology. The method includes: acquiring a real-time current sampling signal at a preset position of a target motor; denoising the real-time current sampling signal to obtain a denoised signal; inputting the denoised signal into an FPGA module for multi-channel parallel data processing to obtain a frequency domain signal and a filtered signal; the multi-channel parallel data processing includes Fast Fourier Transform and Kalman filtering; based on a constructed current prediction model, predicting future current fluctuations according to the frequency domain signal and the filtered signal to obtain a prediction result; and adjusting the sampling frequency and number of sampling points of the real-time current sampling signal according to the prediction result. The method described in this invention significantly reduces the time delay of current detection, improves the real-time performance and accuracy of detection, enhances the system's monitoring capability of the motor's operating status, and has broad application prospects.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of motor control, in particular to a motor current accurate detection method. BACKGROUND

[0002] In the process of motor control, the detection of current plays an important role. Accurate detection of motor current helps to optimize the performance of the motor, improve energy efficiency, and prevent overload and other potential faults.

[0003] At present, the invention patent with application number 202310318348.4 discloses a three-phase motor current detection method. The steps of the method include: obtaining the sampling value of three-phase voltage; calculating the modulation signal of three-phase voltage according to the sampling value; determining the predetermined position of the next current sampling point; adjusting the actual position of the current sampling point according to the interval where the predetermined position is located; sampling the three-phase voltage at the actual position. Through the above technical scheme, the technical solution adjusts the sampling point interval for each current detection, so that the current spikes can be avoided during sampling while ensuring the retention of the change characteristics of the current, thereby improving the accuracy of the sampled current.

[0004] The motor current detection method has certain theoretical basis and potential advantages, but still faces some problems in actual application. When sampling voltage, calculating modulation signal and determining the position of sampling point, it introduces a certain time delay. This delay may affect the real-time performance under high dynamic load, making the detection not timely enough and reducing the system response speed. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a motor current accurate detection method, which significantly reduces the time delay of current detection and improves the real-time performance and accuracy of detection.

[0006] To achieve the above-mentioned purpose, the present application provides the following scheme:

[0007] A motor current accurate detection method, comprising:

[0008] obtaining a real-time current sampling signal at a preset position of a target motor;

[0009] performing denoising processing on the real-time current sampling signal to obtain a denoised signal;

[0010] inputting the denoised signal into an FPGA module for multi-channel data parallel processing to obtain a frequency domain signal and a filtered signal; the multi-channel data parallel processing includes fast Fourier transform and Kalman filtering;

[0011] Based on the constructed current prediction model, future current fluctuation prediction is performed according to the frequency domain signal and the filtered signal, and a prediction result is obtained.

[0012] The sampling point frequency and the sampling point number of the real-time current sampling signal are adjusted according to the prediction result.

[0013] Preferably, the real-time current sampling signal is denoised to obtain a denoised signal, including:

[0014] The real-time current sampling signal is decomposed into sub-signals of different frequencies using a basis function;

[0015] Transform coefficients corresponding to the sub-signals of different frequencies are obtained;

[0016] The variance of the transform coefficients at each decomposition scale is calculated;

[0017] A signal preprocessing model is constructed based on the variance of the transform coefficients;

[0018] The real-time current sampling signal is preprocessed using the signal preprocessing model to obtain the denoised signal.

[0019] Preferably, the signal preprocessing model is constructed based on the variance of the transform coefficients, including:

[0020] The standard deviation of abnormal noise is estimated according to the variance of the transform coefficients;

[0021] A coefficient removal threshold is constructed according to the standard deviation of the noise; wherein the coefficient removal threshold is: wherein t represents the coefficient removal threshold, σ0 represents the variance of the transform coefficients, σ p represents the standard deviation of abnormal noise, median(d p ) represents the median of the transform coefficients at the pth decomposition scale;

[0022] The signal preprocessing model is constructed using the coefficient removal threshold.

[0023] Preferably, the formula of the signal preprocessing model is:

[0024]

[0025] wherein α represents the transform coefficient, w j,k represents the transform coefficient at the jth decomposition scale, m represents the convergence coefficient, sgn represents the sign function, and w j,k represents the transform coefficient after removing outliers.

[0026] Preferably, the real-time current sampling signal is preprocessed using the signal preprocessing model to obtain the denoised signal, including:

[0027] setting initial transform coefficients and convergence coefficients for the signal preprocessing model to obtain a signal preprocessing model with setting completed;

[0028] processing the transform coefficients by using the signal preprocessing model with setting completed to obtain transform coefficients with abnormal values removed;

[0029] reconstructing the transform coefficients with abnormal values removed to obtain a reconstructed signal;

[0030] calculating a signal-to-noise ratio between the reconstructed signal and the original real-time current sampling signal;

[0031] when the signal-to-noise ratio is not in a preset range, resetting the transform coefficients and the convergence coefficients until the signal-to-noise ratio is kept in the preset range.

[0032] Preferably, the formula of the fast Fourier transform is:

[0033]

[0034] wherein X(k) is the kth component of the frequency domain signal, x(n) is the nth sample of the time domain current signal, N is the total number of sampling points, and j is an imaginary unit.

[0035] Preferably, the formula of the prediction stage of the Kalman filter comprises:

[0036]

[0037] P k∣k-1 =A·P k-1∣k-1 ·A T +Q

[0038] The formula of the update stage of the Kalman filter comprises:

[0039] K k =P k∣k-1 ·H T ·(H·P k∣k-1 ·H T +R) -1

[0040]

[0041] P k∣k =(I-K k ·H)·P k∣k-1

[0042] wherein is a prior state estimate, is a posterior state estimate, P k∣k is a posterior error covariance matrix, and P k∣k-1is the prior error covariance matrix, K k is the Kalman gain, z k is the measurement value, A, B, H, Q, R are state transition, control input, observation matrix, process noise covariance, measurement noise covariance matrix, respectively, u k is the control input.

[0043] Preferably, based on the constructed current prediction model, the future current fluctuation prediction is performed according to the frequency domain signal and the filtered signal, and a prediction result is obtained, including:

[0044] An autoregressive prediction model is constructed according to the filtered signal; the formula of the autoregressive prediction model is: wherein, is the predicted future signal value, c is a constant term, φ i is the autoregressive coefficient of the model, is the input filtered data, p is the autoregressive order, ε(t'+1) is the error term, t' represents the time variable, i represents a certain relative position relative to the current time point; the autoregressive order is determined by the frequency domain signal;

[0045] The sequence of the filtered signal is taken as input to train the LSTM model; the LSTM model wherein, h t is the hidden state of the LSTM unit, LSTM(·) represents a nonlinear transformation of the LSTM operation, is the past signal data, h t′-1 is the previous hidden state, is the predicted value obtained by linearly transforming the LSTM output through the Dense layer.

[0046] The parameters of the autoregressive prediction model and the LSTM model are optimized using the prediction error; the target formula of the optimization is: wherein, J is the cost function, N is the total number of data points, is the model prediction output, x(t') is the true target signal value, λ is a regularization parameter, and Θ is a set of model trainable parameters;

[0047] The prediction result is obtained based on a fusion strategy; the formula of the prediction result is: wherein, α' and β' are the prediction weights of the autoregressive prediction model and the prediction weights of the LSTM model, respectively.

[0048] Preferably, the sampling point frequency and the sampling point number of the real-time current sampling signal are adjusted according to the prediction result, including:

[0049] Determine a mean square error according to the prediction result and the actual value; a formula of the mean square error is: Wherein, MSE is the mean square error, z 真实 (t') is an actual value of a real-time current sampling signal;

[0050] Judge whether the mean square error MSE is greater than a preset error upper threshold value, if yes, adjust a sampling point frequency of the real-time current sampling signal according to a formula f new =f old ×(1+α”) and increase the sampling point quantity; wherein, f new is the adjusted sampling point frequency, f old is an original sampling point frequency, and α” is a preset increment factor;

[0051] Judge whether the mean square error MSE is less than a preset error lower threshold value, if yes, adjust the sampling point frequency of the real-time current sampling signal according to a formula f new =f old ×(1-β”) and decrease the sampling point quantity; wherein, β” is a preset decrement factor.

[0052] According to the specific embodiments provided by the application, the following technical effects are disclosed:

[0053] The application provides a motor current accurate detection method, comprising: acquiring a real-time current sampling signal at a preset position of a target motor; carrying out denoising processing on the real-time current sampling signal to obtain a denoising signal; inputting the denoising signal into an FPGA module to carry out multi-channel data parallel processing, to obtain a frequency domain signal and a filtering signal; the multi-channel data parallel processing comprises fast Fourier transform and Kalman filtering; based on a constructed current prediction model, carrying out future current fluctuation prediction according to the frequency domain signal and the filtering signal to obtain a prediction result; and adjusting a sampling point frequency and a sampling point quantity of the real-time current sampling signal according to the prediction result. By the method, time delay of current detection is significantly reduced, real-time performance and accuracy of detection are improved, monitoring capability of the system on the motor running state is enhanced, and the method has wide application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings described in the following only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0055] Figure 1 The method flowchart provided by the embodiments of the present application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be apparently and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0057] In order to make the above objectives, characteristics and advantages of the present application more apparent, comprehensible and easier to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0058] Figure 1 The method flowchart provided by the embodiments of the present application is shown in Figure 1 The present application provides an accurate motor current detection method, which comprises the following steps:

[0059] Step 100: acquiring a real-time current sampling signal at a preset position of a target motor;

[0060] Step 200: performing denoising processing on the real-time current sampling signal to obtain a denoised signal;

[0061] Step 300: inputting the denoised signal into an FPGA module for multi-channel data parallel processing to obtain a frequency domain signal and a filtered signal; the multi-channel data parallel processing comprises fast Fourier transform and Kalman filtering;

[0062] Step 400: based on a constructed current prediction model, performing future current fluctuation prediction according to the frequency domain signal and the filtered signal to obtain a prediction result;

[0063] Step 500: adjusting the sampling point frequency and the sampling point quantity of the real-time current sampling signal according to the prediction result.

[0064] Specifically, the key factors of the preset position are first determined in the present embodiment, which are as follows:

[0065] (1) Working condition of the motor

[0066] Normal working area: positions where the motor often appears in normal operation are selected, and these positions provide the most representative signals.

[0067] Critical operating point: positions where the motor working condition is close to the limit or changes dramatically (such as acceleration or deceleration stage) are selected to capture dynamic characteristics.

[0068] (2) Physical and mechanical characteristics

[0069] Torque and Speed: Key positions based on torque and speed characteristics, such as the point of maximum torque or speed, are considered for good performance measurements.

[0070] Vibration and Noise: Areas where the motor structure may vibrate, which can affect real-time signals, are considered.

[0071] (3) Transitional and Faulty States

[0072] State Switching Points: Points where the motor transitions from one steady state to another are selected to obtain information during the transition.

[0073] Fault Modes: Parts of the position where known fault states occur are selected for predicting and preventing faults.

[0074] Further, the technical method for determining the preset position in the embodiment is as follows:

[0075] Use historical current sampling data analysis to identify key positions. Analyze trends and characteristics to select suitable sampling positions. For example, the embodiment uses a machine learning model to identify the sampling position that has the greatest impact on the prediction result.

[0076] Use simulation tools to identify suitable observation positions by modeling the motor system. Use experimental test methods to find points where output changes are obvious, and adjust according to different working conditions.

[0077] Analyze current signals in real time, dynamically adjust sampling positions through feedback control systems, and implement adaptive strategies to adjust sampling positions according to real-time working condition changes.

[0078] Further, the implementation steps of the embodiment are as follows:

[0079] (1) Use historical data and simulation tools to determine potential preset positions.

[0080] (2) Verify the effectiveness of these positions through actual experiments to ensure high-quality sampling signals.

[0081] (3) Adjust and optimize the preset sampling positions according to real-time feedback to respond to dynamic changes.

[0082] (4) Continuously monitor motor state and prediction error, and update preset positions in a timely manner if error increases or working conditions change.

[0083] In particular, the selection of motor preset sampling positions in the embodiment is a dynamic and finely considered process involving multiple factors such as motor physical characteristics, working conditions, and fault modes. By comprehensively using historical analysis, simulation technology, and real-time adjustment strategies, the most effective sampling positions can be ensured, improving sampling effectiveness and prediction accuracy.

[0084] Preferably, the real-time current sampling signal is denoised to obtain a denoised signal, comprising:

[0085] The real-time current sampling signal is decomposed into sub-signals of different frequencies using a base function;

[0086] Obtain the transform coefficients corresponding to the sub-signals of different frequencies;

[0087] Calculate the variance of the transform coefficients at each decomposition scale;

[0088] Construct a signal preprocessing model based on the variance of the transform coefficients;

[0089] Preprocess the real-time current sampling signal using the signal preprocessing model to obtain the denoised signal.

[0090] Preferably, the signal preprocessing model is constructed based on the variance of the transform coefficients, comprising:

[0091] Estimate the standard deviation of abnormal noise according to the variance of the transform coefficients;

[0092] Construct a coefficient removal threshold according to the standard deviation of the noise; wherein the coefficient removal threshold is: where t represents the coefficient removal threshold, σ0 represents the variance of the transform coefficients, σ p represents the standard deviation of abnormal noise, median(d p ) represents the median of the transform coefficients at the pth decomposition scale;

[0093] Construct a signal preprocessing model using the coefficient removal threshold.

[0094] Preferably, the formula of the signal preprocessing model is:

[0095]

[0096] where α represents the transform coefficient, w j,k represents the transform coefficient at the jth decomposition scale, m represents the convergence coefficient, sgn represents the sign function, and w j,k represents the transform coefficient after removing outliers.

[0097] Specifically, the original signal is transformed into sub-signals of different frequencies according to the principle of wavelet transform, and the above sequence preprocessing model is continuous at t, the function value is (1-α)t, and the first derivative is 1-α at t. The convergence speed can be adjusted by m, and the larger m, the faster the convergence. The contraction degree of the wavelet coefficient within ±t can be adjusted by α, so that the above function can be transformed between the conventional hard threshold function and the soft threshold function, and the characteristics of the original signal can be better understood, so that the interference of noise can be perfectly removed.

[0098] Preferably, the real-time current sampling signal is preprocessed using the signal preprocessing model to obtain the denoised signal, including:

[0099] Setting initial transform coefficients and convergence coefficients for the signal preprocessing model to obtain a signal preprocessing model with setting completed;

[0100] Processing the transform coefficients using the signal preprocessing model with setting completed to obtain transform coefficients with outliers removed;

[0101] Reconstructing the transform coefficients with outliers removed to obtain a reconstructed signal;

[0102] Calculating the signal-to-noise ratio between the reconstructed signal and the original real-time current sampling signal;

[0103] When the signal-to-noise ratio is not in the preset range, the transform coefficients and the convergence coefficients are re-set until the signal-to-noise ratio remains in the preset range.

[0104] Optionally, in the embodiment, the current sensor collects current data in real time while the motor is running, and outputs the data in the form of an analog signal to a signal conditioning circuit. The conditioning circuit filters and amplifies the analog signal to preliminarily remove noise and converts it to a level suitable for ADC (analog-to-digital converter) processing.

[0105] Further, the embodiment converts the conditioned analog signal into a digital signal through a high-speed ADC and transmits it to an FPGA. The parallel processing architecture is configured in the FPGA to enable it to process data from multiple channels simultaneously, significantly improving signal processing speed. Fast Fourier Transform (FFT) is performed using the FPGA to analyze the current spectrum, and a Kalman filter is applied for real-time filtering to improve signal purity.

[0106] Further, the embodiment utilizes the parallel processing capability of the FPGA to accelerate the processing of multiple current signals and reduce delay.

[0107] Preferably, the formula of the fast Fourier transform is:

[0108]

[0109] where X(k) is the kth component of the frequency domain signal, x(n) is the nth sample of the time domain current signal, N is the total number of sampling points, and j is the imaginary unit.

[0110] In the FPGA, an adaptive adjustment mechanism can be set up to change N according to the severity of signal fluctuation, so as to improve the calculation efficiency. For example, a smaller N is used when the measured signal changes less, while a larger N is used when the changes are larger, so as to retain more spectral information.

[0111] Preferably, the formula of the prediction stage of the Kalman filter comprises:

[0112]

[0113] P k∣k-1 = A·P k-1∣k-1 ·A T + Q

[0114] The formula of the update stage of the Kalman filter comprises:

[0115] K k = P k∣k-1 ·H T ·(H·P k∣k-1 ·H T + R) -1

[0116]

[0117] P k∣k = (I-K k ·H)·P k∣k-1

[0118] wherein, is the prior state estimate, is the posterior state estimate, P k∣k is the posterior error covariance matrix, P k∣k-1 is the prior error covariance matrix, K k is the Kalman gain, z k is the measurement value, A, B, H, Q, R are state transition, control input, observation matrix, process noise covariance, measurement noise covariance matrix respectively, u k is the control input.

[0119] Demonstratively, the present embodiment uses a real-time data model to intelligently adjust A and Q, so that the filter can adapt to the dynamically changing current signal, further improving the filtering effect. Based on the parallel ability of FPGA, the gain calculation is optimized, so that the calculation is more efficient, thereby meeting the real-time requirement.

[0120] Specifically, the fast Fourier transform (FFT) of the present embodiment is to convert the current signal in the time domain into a frequency domain signal to analyze the frequency spectrum characteristics of the current, so as to identify whether there is an abnormal frequency component (such as a harmonic or noise). The Kalman filter is to process the noise interference and extract the effective information in the signal to obtain a purer current signal. This is very crucial for effectively isolating random noise and interference in the current signal.

[0121] Further, after FFT processing, the frequency domain signal obtained is used to further analyze the running state and characteristics of the motor. The frequency domain information can help identify whether the motor has abnormal conditions, such as abnormal vibration or load change. After Kalman filter processing, the system can obtain a relatively clean and stable current signal, which removes the noise effect and makes the signal remain true.

[0122] Optionally, through the innovative formula and method of FFT and Kalman filter described above, the FPGA can effectively process parallel signals and generate frequency domain information and filtered pure signals of the current in real time. These processing results can be used in the real-time analysis and feedback stage to help the system identify and respond to dynamic current changes more quickly.

[0123] Preferably, based on the constructed current prediction model, future current fluctuation prediction is performed according to the frequency domain signal and the filtered signal to obtain a prediction result, including:

[0124] An autoregressive prediction model is constructed according to the filtered signal; the formula of the autoregressive prediction model is: wherein, is a predicted future signal value, c is a constant term, and φ i is an autoregressive coefficient of the model, is input filtered data, p is an autoregressive order, and ε(t'+1) is an error term, t' represents a time variable, and i represents a certain relative position relative to the current time point; the autoregressive order is determined by the frequency domain signal;

[0125] The sequence of the filtered signal is taken as input to train an LSTM model; the LSTM model wherein, h t is a hidden state of the LSTM unit, LSTM(·) represents a nonlinear transformation of the LSTM operation, is past signal data, h t′-1 is a previous hidden state, is a prediction value obtained by linearly transforming the LSTM output through a Dense layer.

[0126] The parameters of the autoregressive prediction model and the LSTM model are optimized using the prediction error; the target formula of the optimization is: where J is the cost function, N is the total number of data points, is the model predicted output, x(t') is the true target signal value, λ is the regularization parameter, Θ is the set of model trainable parameters;

[0127] The prediction result is obtained based on the fusion strategy; the formula of the prediction result is: where α' and β' are the prediction weights of the autoregressive prediction model and the LSTM model, respectively. The prediction weights of the autoregressive prediction model and the LSTM model can be selected as 0.3 and 0.7, respectively.

[0128] Optionally, for a signal with multiple significant frequency components, the embodiment takes into account the proportion of these frequencies in the total energy, and calculates the total effect of combining the primary and secondary frequencies:

[0129]

[0130] The energy percentage of each frequency is calculated, and p is dynamically adjusted based on this information to reflect the complex frequency composition:

[0131]

[0132] where E total is the total energy of the signal, which is the sum of the energies at all frequencies, X(f i ) is the frequency domain representation of f i , M is the number of frequency components to be considered primarily, f s is the sampling frequency, and f i is the i-th significant frequency component

[0133] Preferably, the sampling point frequency and the number of sampling points of the real-time current sampling signal are adjusted according to the prediction result, including:

[0134] The mean square error is determined according to the prediction result and the actual value; the formula of the mean square error is: where MSE is the mean square error, z 真实 (t') is the actual value of the real-time current sampling signal;

[0135] It is judged whether the mean square error MSE is greater than a preset upper error threshold, if yes, the sampling point frequency of the real-time current sampling signal is adjusted according to the formula f new = f old ×(1+α”), and the number of sampling points is increased; where f new is the adjusted sampling point frequency, f old is the original sampling point frequency, and α” is a preset increment factor;

[0136] determining whether the mean square error (MSE) is less than a preset lower error threshold value, and if yes, adjusting the sampling point frequency of the real-time current sampling signal according to a formula f new =f old and reducing the number of sampling points, wherein β" is a preset reduction factor.

[0137] Specifically, the upper error threshold value in the embodiment is that when the prediction error exceeds the value, it is considered that the sampling is insufficient, and the sampling rate needs to be increased. The lower error threshold value is that when the error is lower than the value and the signal is smooth, the sampling rate can be reduced to reduce data redundancy.

[0138] Optionally, in the embodiment, the system response is observed under different increment and reduction factors through experiments, and the effects are recorded. According to these observation values, the best factor value is found by adjusting. Exemplarily, the preset increment factor and the preset reduction factor in the embodiment are respectively set as:

[0139] Increment factor: increase 1 kHz each time the sampling frequency is adjusted (for example, the current sampling frequency is 44.1 kHz, and after the first adjustment, it is set to 45.1 kHz).

[0140] Reduction factor: reduce 1 kHz each time the sampling frequency is adjusted (for example, the frequency needs to be reduced in the adjustment process, such as from 48 kHz to 47 kHz).

[0141] The beneficial effects of the present application are as follows:

[0142] In the present application, starting from the accuracy of signal acquisition, through high-speed data processing and intelligent analysis, efficient detection and feedback of current signal are realized. High-precision sensors ensure the accuracy of basic data, laying a foundation for subsequent FPGA high-speed processing; FPGA undertakes efficient parallel processing tasks, speeds up data processing speed, so that the detection result can be quickly analyzed and fed back; the prediction algorithm and the dynamic adjustment complement each other, improving the adaptive ability and response speed of the system. Real-time feedback ensures that the overall system still maintains the best performance under the constantly changing motor operating conditions.

[0143] In the present application, starting from the accuracy of signal acquisition, through high-speed data processing and intelligent analysis, efficient detection and feedback of current signal are realized. High-precision sensors ensure the accuracy of basic data, laying a foundation for subsequent FPGA high-speed processing; FPGA undertakes efficient parallel processing tasks, speeds up data processing speed, so that the detection result can be quickly analyzed and fed back; the prediction algorithm and the dynamic adjustment complement each other, improving the adaptive ability and response speed of the system. Real-time feedback ensures that the overall system still maintains the best performance under the constantly changing motor operating conditions.

[0144] The principles and implementation manners of the present application are described by using specific examples in the present application, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the art, the specific implementation manners and application ranges will be changed according to the idea of the present application. In conclusion, the content of the present specification should not be understood as the limitation of the present application.

Claims

1. A method for accurate detection of motor current, characterized in that, The method comprises the following steps: obtaining a real-time current sampling signal at a preset position of a target motor; performing denoising processing on the real-time current sampling signal to obtain a denoised signal; inputting the denoised signal into an FPGA module for multi-channel data parallel processing to obtain a frequency domain signal and a filtered signal; the multi-channel data parallel processing comprises fast Fourier transform and Kalman filtering; based on a constructed current prediction model, predicting future current fluctuation according to the frequency domain signal and the filtered signal to obtain a prediction result; adjusting the sampling point frequency and the sampling point quantity of the real-time current sampling signal according to the prediction result; predicting future current fluctuation according to the frequency domain signal and the filtered signal to obtain a prediction result, comprising: An autoregressive prediction model is constructed according to the filtered signal; the formula of the autoregressive prediction model is: ; wherein, is a predicted future signal value, is a constant term, is an autoregressive coefficient of the model, is input filtered data, p is an autoregressive order, is an error term, denotes a time variable, denotes a certain relative position with respect to a current time point; the autoregressive order is determined by the frequency domain signal; filtering signal as input; the LSTM model is trained with sequences of the filtered signal ; wherein, is the hidden state of the LSTM unit, LSTM(·) denotes a non-linear transformation of the LSTM operation, is the past signal data, is the previous hidden state, is the predicted value obtained by linearly transforming the LSTM output through a Dense layer; parameters of the autoregressive prediction model and the LSTM model using prediction error; the target formula of optimization is: ; wherein, is a cost function, is the total number of data points, is the model prediction output, is the true target signal value, is a regularization parameter, is a set of model trainable parameters; obtaining the prediction result based on the fusion strategy; a formula of the prediction result is: ; wherein, ' and ' are respectively a prediction weight of an autoregressive prediction model and a prediction weight of an LSTM model. adjusting the sampling point frequency and the sampling point quantity of the real-time current sampling signal according to the prediction result, comprising: Determine the mean square error according to the prediction result and the actual value; the formula of the mean square error is: ; wherein, is the mean square error, is the actual value of the real-time current sampling signal; judging the mean square error preset If yes, the sampling point frequency of the real-time current sampling signal is adjusted according to the formula and the number of sampling points is increased; wherein, is the adjusted sampling point frequency, is the original sampling point frequency, is a preset increment factor; judging mean square error a preset error lower threshold value, if yes, then according to the formula adjusting the sampling point frequency of the real-time current sampling signal and reducing the sampling point quantity; wherein, is a preset decrement factor.

2. The method of claim 1, wherein, performing denoising processing on the real-time current sampling signal to obtain a denoised signal, comprising: decomposing the real-time current sampling signal into sub-signals of different frequencies using a basis function; obtaining transform coefficients corresponding to the sub-signals of different frequencies; calculating the variance of the transform coefficients at each decomposition scale; constructing a signal preprocessing model based on the variance of the transform coefficients; preprocessing the real-time current sampling signal using the signal preprocessing model to obtain the denoised signal.

3. The method of claim 2, wherein, The signal preprocessing model is constructed based on the variance of the transform coefficients, comprising: estimating the standard deviation of abnormal noise according to the variance of the transform coefficients; The coefficient removal threshold is constructed according to a standard deviation of the noise; wherein the coefficient removal threshold is: ; wherein, represents the coefficient removal threshold, represents a variance of the transform coefficient, represents a standard deviation of the abnormal noise, represents a median value of the transform coefficient at the p th decomposition scale; constructing a signal preprocessing model using the coefficient removal threshold.

4. The method of claim 3, wherein, The formula of the signal preprocessing model is: ； wherein, denotes a transform coefficient, denotes a transform coefficient at the j-th decomposition scale, denotes a convergence coefficient, denotes a sign function, denotes a transform coefficient with outliers removed.

5. The method of claim 2, wherein, The signal preprocessing model is used to preprocess the real-time current sampling signal to obtain the denoised signal, comprising: setting initial transform coefficients and convergence coefficients for the signal preprocessing model to obtain a set signal preprocessing model; processing the transform coefficients using the set signal preprocessing model to obtain transform coefficients with abnormal values removed; reconstructing the transform coefficients with abnormal values removed to obtain a reconstructed signal; calculating the signal-to-noise ratio between the reconstructed signal and the original real-time current sampling signal; when the signal-to-noise ratio is not within a preset range, resetting the transform coefficients and the convergence coefficients until the signal-to-noise ratio remains within the preset range.

6. The method of claim 1, wherein, The formula of the fast Fourier transform is: ； wherein is the kth component of the frequency domain signal, is the kth component of the frequency domain signal, is the kth sample of the time domain current signal, is the kth sample of the time domain current signal, is the total number of sampling points, is the imaginary unit.

7. The method of claim 1, wherein, The formula of the prediction stage of the Kalman filter comprises: ； The formula of the update stage of the Kalman filter comprises: ； wherein, is a prior state estimate, is a posterior state estimate, is a posterior error covariance matrix, is a prior error covariance matrix, is a Kalman gain, is a measurement, are a state transition, a control input, an observation matrix, a process noise covariance, a measurement noise covariance matrix, respectively, is a control input.

Images