A method and device for suppressing resonance of a long-line drive system of a frequency converter

Abstract

The present invention relates to the technical field of inverter resonance control, and provides a method and device for suppressing resonance in a long-line drive system of an inverter. The method comprises obtaining time series data, performing normalization processing, calculating a theoretical resonant frequency, and predicting the resonant frequency using an attention-enhanced long-short-term memory neural network; calculating a frequency error compensation amount, and calculating a harmonic suppression comprehensive performance evaluation index based on the frequency error compensation amount and the predicted resonant frequency; evaluating the harmonic suppression effect based on the harmonic suppression comprehensive performance evaluation index and an index threshold, adjusting the weight coefficient of the attention-enhanced long-short-term memory neural network using an NSGA-II optimization algorithm; and adjusting the center frequency and bandwidth parameters of the inverter long-line drive system based on the predicted resonant frequency to achieve resonance suppression in the inverter long-line drive system. The present invention enhances the feature extraction capability of an LSTM network through an attention mechanism, significantly improving the accuracy of resonant frequency prediction and enhancing the harmonic suppression effect.

Description

Technical Field

[0001] The present invention relates to the technical field of inverter resonance control, and in particular to a method and device for suppressing resonance of an inverter long-line drive system. Background Art

[0002] In long-line inverter drive systems, high-frequency PWM pulses can generate significant resonance during transmission due to the distributed parameter effect of long cable transmission. When the system switching frequency exceeds 2kHz, the resonance amplitude can reach 1.5 to 2 times the fundamental frequency, seriously affecting system stability and equipment life.

[0003] Existing resonance suppression technologies mainly include passive filtering, active damping, predictive control, and intelligent control. Passive filtering uses LC filters for resonance suppression, which has a simple structure but poor adaptability and cannot cope with changes in system parameters. Active damping suppresses resonance by injecting a damping component into the system, but this introduces additional losses and reduces system efficiency. Predictive control performs feedforward control based on the system model, but the computational complexity is large, with typical computation times reaching hundreds of microseconds, making it difficult to achieve real-time control. Intelligent control uses methods such as neural networks, which have certain adaptive capabilities but require a large amount of training data and high resource requirements. These existing technologies either lack precision or have high computational complexity, making it difficult to achieve real-time, accurate, and economical resonance suppression in engineering. Summary of the Invention

[0004] The present invention aims to address at least one of the technical problems existing in the related art. To this end, the present invention provides a method and device for intelligent resonance suppression in a long-line drive system for a frequency converter. By enhancing the feature extraction capabilities of an LSTM network through an attention mechanism, the method significantly improves the accuracy of resonant frequency prediction and enhances harmonic suppression.

[0005] The present invention provides a method for suppressing resonance of a long-line drive system of a frequency converter, comprising:

[0006] S1: Obtain the operating parameters of the inverter long-line drive system in real time through the data acquisition unit, obtain time series data, normalize the time series data, and obtain standard time series data;

[0007] S2: Calculate the theoretical resonant frequency based on the operating parameters of the inverter long-line drive system;

[0008] S3: Predict the resonant frequency of the standard time series data through the attention-enhanced long short-term memory neural network to obtain the predicted resonant frequency;

[0009] S4: Calculate the frequency error compensation amount based on the predicted resonant frequency and the theoretical resonant frequency, and calculate the comprehensive performance evaluation index of harmonic suppression based on the frequency error compensation amount and the predicted resonant frequency;

[0010] S5: Evaluate the harmonic suppression effect based on the harmonic suppression comprehensive performance evaluation index and the index threshold. If the harmonic suppression comprehensive performance evaluation index is greater than or equal to the index threshold, execute step S6; if the harmonic suppression comprehensive performance evaluation index is less than the index threshold, execute step S7;

[0011] S6: Adjust the weight coefficient of the attention-enhanced long short-term memory neural network through the NSGA-II optimization algorithm, and repeat steps S3 to S5;

[0012] S7: Adjust the adaptive multiple notch filter parameters in the inverter long-line drive system according to the predicted resonant frequency to achieve resonance suppression of the inverter long-line drive system.

[0013] Furthermore, in step S2, the operating parameters of the inverter long-line drive system include cable inductance and cable capacitance, and the theoretical resonant frequency of the inverter long-line drive system is calculated based on the cable inductance and cable capacitance;

[0014] The theoretical resonant frequency calculation expression of the inverter long-line drive system is:

[0015]

[0016] in, is the theoretical resonant frequency of the inverter long-line drive system, is the cable inductance per unit length, is the cable capacitance per unit length.

[0017] Furthermore, the attention-enhanced long short-term memory neural network includes an input layer, a first long short-term memory neural network layer, a second long short-term memory neural network layer, an attention mechanism layer, and an output layer.

[0018] Furthermore, step S3 includes:

[0019] S31: Enhance the weights of the first LSTM neural network layer and the second LSTM neural network layer to extract key resonance features through the attention mechanism layer;

[0020] S32: extracting the time-frequency features of the standard time series data through the first long short-term memory neural network layer to obtain the state features of the first hidden layer;

[0021] S33: extracting the frequency and amplitude of the first hidden layer state feature through the second long short-term memory neural network layer to obtain the second hidden layer state feature;

[0022] S34: Perform weighted calculation on the first hidden layer state feature and the second hidden layer state feature according to the weight coefficient to obtain the predicted resonant frequency.

[0023] Furthermore, in step S31, the calculation expression of the attention weight is:

[0024]

[0025] in, For the Step attention weight, For the The first hidden layer state features, For the Step 2 hidden layer state features, W is the first weight matrix, V is the second weight matrix, U is the third weight matrix, is the activation function, is the tangent function.

[0026] Furthermore, the calculation expression for predicting the resonant frequency is:

[0027]

[0028] in, To predict the resonant frequency, is the reference frequency, is the state feature of the first hidden layer, is the state feature of the second hidden layer, is the weight coefficient of the first hidden layer, is the weight coefficient of the second hidden layer.

[0029] Furthermore, in step S4, the frequency error compensation amount is calculated as follows:

[0030]

[0031] in, is the frequency error compensation amount, To predict the resonant frequency, is the theoretical resonant frequency of the inverter long-line drive system;

[0032] The calculation expression of comprehensive performance evaluation index is:

[0033]

[0034] in, is a comprehensive performance evaluation index. is the inverter switching frequency, is the mean of the attention weights.

[0035] Furthermore, the indicator threshold is divided into a first key indicator threshold and a second key indicator threshold by using the K-means clustering method, and the harmonic suppression effect is evaluated according to the first key indicator threshold and the second key indicator threshold;

[0036] when When , the inhibition effect is optimal;

[0037] when When , the inhibitory effect is suboptimal;

[0038] when When , the inhibitory effect is poor;

[0039] in, is a comprehensive performance evaluation index. is the first key indicator threshold, is the second key indicator threshold.

[0040] Furthermore, in step S7, the inverter long-line drive system includes an adaptive multiple notch filter, and the adaptive multiple notch filter parameters include a center frequency and a bandwidth.

[0041] The calculation expression for the corrected center frequency is:

[0042]

[0043] in, To correct the center frequency, is the proportionality coefficient, To predict the resonant frequency, is the frequency error compensation amount, is the reference frequency;

[0044] The calculation expression of bandwidth is:

[0045]

[0046] in, is the bandwidth, is the regulating factor, is the total harmonic distortion.

[0047] The present invention further provides a device for suppressing resonance of a long-line drive system of a frequency converter, which is used to execute the above-mentioned method for suppressing resonance of a long-line drive system of a frequency converter, comprising:

[0048] A data acquisition unit, which collects operating parameters of the inverter long-line drive system through a voltage transformer and a current transformer;

[0049] A local processing unit, wherein the local processing unit performs data preprocessing;

[0050] a central control unit that performs attention-enhanced long short-term memory neural network prediction and harmonic suppression control;

[0051] A human-computer interaction interface displays the system operation status.

[0052] The above one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects:

[0053] By enhancing the feature extraction capabilities of the LSTM network through an attention mechanism, prediction accuracy has increased to over 95%, a 30% improvement over traditional methods. Based on comprehensive performance evaluation indicators, intelligent evaluation is implemented through multi-level criteria, improving resonance suppression effectiveness by over 90%. This comprehensive evaluation system, combined with adaptive closed-loop control, achieves a resonance amplitude attenuation rate exceeding 90%, THD reduction by over 65%, and dynamic response time reduced to less than 5ms. It supports complex operating conditions with cable lengths of 0.5km to 10km, temperatures of -10°C to 50°C, humidity of 30% to 95% RH, and load fluctuations of 10% to 100%. Voltage compatibility reaches ±10% of the rated value, covering all industrial scenarios. Real-time feedback triggers parameter optimization to ensure the system maintains optimal performance. Automatically alerting and switching to safe mode in the event of an anomaly ensures operational reliability.

[0054] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 The present invention provides a flow chart of a method for suppressing resonance of a long-line drive system of a frequency converter.

[0057] Figure 2 The present invention provides a schematic structural diagram of a device for suppressing resonance of a long-line drive system of a frequency converter.

[0058] Reference numerals:

[0059] 101. Data acquisition unit; 102. Local processing unit; 103. Central control unit; 104. Human-computer interaction interface. DETAILED DESCRIPTION

[0060] To make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below. Obviously, the embodiments described are 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 ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0061] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0062] The following combination Figures 1 to 2 The present invention describes an intelligent resonance suppression method and device for a frequency converter long-line drive system.

[0063] like Figure 1 As shown, a method for suppressing resonance of a long-line drive system of a frequency converter includes:

[0064] S1: Obtain the operating parameters of the inverter long-line drive system in real time through the data acquisition unit, obtain time series data, normalize the time series data, and obtain standard time series data;

[0065] In long-line drive systems driven by VFDs, the distributed parameters of long cables are affected by factors such as ambient temperature, humidity, and aging, leading to dynamic changes in resonant frequency. To accurately model the system's impedance characteristics, a data acquisition unit is required to collect real-time operating parameters of the VFD long-line drive system and obtain time series data. This time series data includes voltage and current harmonics, temperature, humidity, and load fluctuation.

[0066] Electrical parameters: The voltage V is collected through a high-precision voltage transformer (accuracy ±0.2%) and the current I is collected through a current transformer (accuracy ±0.5%). The sampling frequency is 10kHz to ensure that the higher harmonics of the PWM pulse (up to the 40th) are captured.

[0067] Environmental parameters: Temperature and humidity are collected through digital sensors (accuracy ±0.3°C, ±2%RH) with a sampling frequency of 1Hz. The sampled data are transmitted to the central controller via RS485.

[0068] Load fluctuation: The load fluctuation rate (10% to 100%) was recorded in real time using a power analyzer (VITREK PA910) with an accuracy of ±0.1%.

[0069] By normalizing time series data, standard time series data can be obtained to improve data quality.

[0070] S2: Calculate the theoretical resonant frequency based on the operating parameters of the inverter long-line drive system;

[0071] The operating parameters of the inverter long-line drive system include cable inductance and cable capacitance. The system impedance characteristics are calculated based on the cable inductance and cable capacitance. The calculation expression of the system impedance characteristics is:

[0072]

[0073] in, is the system impedance characteristic, is the cable inductance per unit length, is the cable capacitance per unit length, is the Laplace operator, is the tangent function;

[0074] Indicates the impedance response of the long-distance cable of the inverter at different frequencies, and its amplitude A maximum value appears at a specific frequency point, and the frequency corresponding to the maximum value point is the actual resonant frequency of the inverter long-line drive system.

[0075] In actual projects, cable distribution parameters are affected by temperature and humidity. For example, temperature increases cause the insulation material to expand, which increases the capacitance C. The theoretical resonant frequency calculation expression of the inverter long-line drive system is:

[0076]

[0077] in, is the theoretical resonant frequency of the inverter long-line drive system, is the cable inductance per unit length, is the cable capacitance per unit length.

[0078] The deviation between the actual resonant frequency and the theoretical resonant frequency is used to train the error compensation of the LSTM model.

[0079] S3: Predict the resonant frequency of the standard time series data through the attention-enhanced long short-term memory neural network to obtain the predicted resonant frequency;

[0080] The attention-enhanced long short-term memory neural network includes an input layer, a first long short-term memory neural network layer, a second long short-term memory neural network layer, an attention mechanism layer, and an output layer;

[0081] Input layer: 32 nodes, receiving standard time series data;

[0082] The first long short-term memory neural network layer (LSTM layer 1): 64 memory cells, extracting time-frequency features such as fundamental amplitude fluctuations and harmonic energy distribution;

[0083] Second long short-term memory neural network layer (LSTM layer 2): 64 memory cells, extracting the time-varying patterns of the resonant frequency, such as frequency drift caused by sudden load changes;

[0084] Attention mechanism layer: 64-dimensional vector, enhancing the weight of key features;

[0085] Output layer: 3 nodes, outputs the predicted resonant frequency.

[0086] S31: Enhance the weights of the first LSTM neural network layer and the second LSTM neural network layer to extract key resonance features through the attention mechanism layer;

[0087] The calculation expression of attention weight is:

[0088]

[0089] in, For the Step attention weight, For the The first hidden layer state features, For the Step 2 hidden layer state features, W is the first weight matrix, V is the second weight matrix, U is the third weight matrix, is the activation function, is the tangent function;

[0090] S32: extracting the time-frequency features of the standard time series data through the first long short-term memory neural network layer to obtain the state features of the first hidden layer;

[0091] S33: extracting the frequency and amplitude of the first hidden layer state feature through the second long short-term memory neural network layer to obtain the second hidden layer state feature;

[0092] S34: Perform weighted calculation on the first hidden layer state feature and the second hidden layer state feature according to the weight coefficient to obtain the predicted resonant frequency.

[0093] The calculation expression for predicting the resonant frequency is:

[0094]

[0095] in, To predict the resonant frequency, is the reference frequency, is the state feature of the first hidden layer, is the state feature of the second hidden layer, is the weight coefficient of the first hidden layer, is the weight coefficient of the second hidden layer.

[0096] The two-step feature extraction combined with the attention mechanism significantly improves the prediction focus and accuracy.

[0097] S4: Calculate the frequency error compensation amount based on the predicted resonant frequency and the theoretical resonant frequency, and calculate the comprehensive performance evaluation index of harmonic suppression based on the frequency error compensation amount and the predicted resonant frequency;

[0098] The calculation expression of frequency error compensation is:

[0099]

[0100] in, is the frequency error compensation amount, To predict the resonant frequency, is the theoretical resonant frequency of the inverter long-line drive system;

[0101] The calculation expression of comprehensive performance evaluation index is:

[0102]

[0103] in, is a comprehensive performance evaluation index. is the inverter switching frequency, is the mean of the attention weights.

[0104] It represents the ratio of the corrected resonant frequency to the switching frequency, reflecting the degree of system resonance risk. The closer the value is to 1, the closer the resonant frequency is to the switching frequency, and the higher the system resonance risk.

[0105] Reflects the stability of attention distribution, is the mean of the attention weights, The larger the value, the more attention is focused on the key features. The smaller it is, the more stable the feature extraction is;

[0106] It reflects the comprehensive status of system resonance risk and feature extraction stability;

[0107] The smaller the value, the lower the risk of system resonance and the more stable the feature extraction, which means the better the suppression effect and the more reliable the system operation.

[0108] The composite indicator design enables the system to ensure the stability of feature extraction while ensuring the resonance suppression effect, avoiding unstable predictions caused by fluctuations in the attention mechanism.

[0109] S5: Evaluate the harmonic suppression effect based on the harmonic suppression comprehensive performance evaluation index and the index threshold. If the harmonic suppression comprehensive performance evaluation index is greater than or equal to the index threshold, execute step S6; if the harmonic suppression comprehensive performance evaluation index is less than the index threshold, execute step S7;

[0110] The indicator threshold is divided into a first key indicator threshold and a second key indicator threshold by using the K-means clustering method, and the harmonic suppression effect is evaluated according to the first key indicator threshold and the second key indicator threshold;

[0111] when When , the inhibition effect is optimal;

[0112] when When , the inhibitory effect is suboptimal;

[0113] when When , the inhibitory effect is poor;

[0114] in, is a comprehensive performance evaluation index. is the first key indicator threshold, is the second key indicator threshold.

[0115] In some specific embodiments of the present invention, more than a thousand sets of experimental data were collected under 15 different cable lengths (0.5km~10km), 4 rated voltage levels (380V, 690V, 6kV and 10kV), and 5 switching frequencies (2kHz~8kHz). For each set of data, the corresponding relationship between the resonance amplitude decay rate, total harmonic distortion (THD) and dynamic response time and the harmonic suppression comprehensive performance evaluation index η was recorded, and K-means clustering was used to calculate the corresponding relationship between the resonance amplitude decay rate, total harmonic distortion (THD) and dynamic response time and the harmonic suppression comprehensive performance evaluation index η. It is divided into multiple intervals, and two key thresholds of 0.05 and 0.08 are determined through statistical analysis. In actual use, they can be fine-tuned within the range of ±10% according to the specific application scenario.

[0116] when When the suppression effect is optimal, the resonance amplitude attenuation is >90%, THD <3%, and dynamic response <5ms;

[0117] when When the suppression effect is suboptimal, the resonance amplitude attenuation is greater than 80%, THD is less than 5%, and the dynamic response is less than 10ms. The weight coefficient of the attention-enhanced long-term and short-term memory neural network is adjusted by the NSGA-II optimization algorithm, and steps S3 to S5 are repeated until .

[0118] S6: Adjust the weight coefficient of the attention-enhanced long short-term memory neural network through the NSGA-II optimization algorithm, and repeat steps S3 to S5;

[0119] In some specific embodiments of the present invention, the first hidden layer weight coefficient is optimized by the NSGA-II optimization algorithm. , weight coefficient of the second hidden layer and regulatory factors .

[0120] Optimization objectives: Minimize total harmonic distortion (THD < 5%); minimize dynamic response time (< 10ms) Constraints: .

[0121] NSGA-II parameters: population size: 100, number of iterations: 500, crossover probability: 0.8; the selection criteria for the Pareto solution are to first meet the THD and response time thresholds, and then minimize .

[0122] Optimization results: .

[0123] Performance improvement: THD dropped from 7.2% to 4.2%, response time shortened from 15.3ms to 8.5ms, Reach the best level.

[0124] S7: Adjust the adaptive multiple notch filter parameters in the inverter long-line drive system according to the predicted resonant frequency to achieve resonance suppression of the inverter long-line drive system.

[0125] The inverter long-line drive system includes an adaptive multi-notch filter, and the adaptive multi-notch filter parameters include center frequency and bandwidth.

[0126] The calculation expression for the corrected center frequency is:

[0127]

[0128] in, To correct the center frequency, is the proportionality coefficient, To predict the resonant frequency, is the frequency error compensation amount, is the reference frequency.

[0129] In some specific embodiments of the present invention, , Balance adjustment speed and stability through step response experimental optimization.

[0130] The calculation expression of bandwidth is:

[0131]

[0132] in, is the bandwidth, is the regulating factor, is the total harmonic distortion.

[0133] In some specific embodiments of the present invention, , when THD=5%, bandwidth B=250Hz, covering ±2 times the resonant frequency fluctuation range.

[0134] After outputting the PWM modulation signal, recollect the voltage and current data and calculate the updated Value; if , maintain current parameters and continuously monitor system status.

[0135] Closed-loop iterative adjustment: If , re-optimize the parameters 、 、 , repeat steps S3 to S5 until the criterion is met.

[0136] Exception handling: When three consecutive iterations fail to converge, the following actions are triggered:

[0137] Enable the preset passive filter (cut-off frequency 1kHz) to suppress resonance;

[0138] Send an alarm signal to the host computer, prompting "system abnormality, manual intervention is required";

[0139] The applicable frequency range of the present invention is 0.5 to 1.0 times of the switching frequency of the frequency converter, and the rated voltage range of the frequency converter long-line drive system is 380V to 10kV. Real-time feedback of values ​​triggers parameter optimization to ensure that the system is continuously in the optimal state. In abnormal situations, it automatically alarms and switches to safe mode to ensure operational reliability.

[0140] The present invention enhances the feature extraction capability of the LSTM network through the attention mechanism, and the prediction accuracy is increased to over 95%, which is 30% higher than the traditional method. According to the comprehensive performance evaluation indicators, intelligent evaluation is achieved through multi-level criteria, which improves the resonance suppression effect by over 90%. Based on the comprehensive evaluation system and adaptive closed-loop control based on multi-level criteria, the resonance amplitude attenuation rate exceeds 90%, THD is reduced by more than 65%, and the dynamic response time is shortened to less than 5ms. It supports complex working conditions with cable lengths of 0.5km to 10km, temperatures of -10℃ to 50℃, humidity of 30% to 95%RH, and load fluctuations of 10% to 100%. The voltage compatibility reaches ±10% of the rated value, covering the needs of all industrial scenarios.

[0141] like Figure 2 As shown, a device for suppressing resonance of a long-line drive system of a frequency converter is used to perform a method for suppressing resonance of a long-line drive system of a frequency converter, comprising:

[0142] The data acquisition unit 101 collects the operating parameters of the inverter long-line drive system through the voltage transformer and the current transformer;

[0143] In some specific embodiments of the present invention, the sampling frequency of voltage is 10kHz, the sampling frequency of current is 10kHz, and the sampling frequency of temperature and humidity is 1kHz, and the temperature and humidity are collected in real time by the temperature and humidity sensor;

[0144] The local processing unit 102 performs data preprocessing;

[0145] Extract fundamental and harmonic components, and calculate cable impedance characteristics and theoretical resonant frequency;

[0146] The central control unit 103 performs attention-enhanced long short-term memory neural network prediction and harmonic suppression control;

[0147] The central control unit 103 executes the attention-enhanced LSTM network to predict the output of the predicted resonant frequency and frequency error compensation , and dynamically adjust the adaptive multiple notch filter parameters to generate PWM modulation signals.

[0148] The human-computer interaction interface 104 displays the system operation status.

[0149] Example:

[0150] Application scenario: A submersible motor at an onshore oilfield is connected to a 10kV / 400kW inverter via a 6km-long XLPE (cross-linked polyethylene) insulated copper-core cable. The distributed parameters (inductance and capacitance) of the cable vary dynamically due to the ambient temperature and humidity (35°C to 45°C, 80% to 90% RH). This causes the PWM voltage output by the inverter to induce high-frequency resonance in the cable, resulting in voltage distortion and equipment overheating.

[0151] System configuration: inverter rated voltage 10kV, switching frequency 4kHz, output power 400kW.

[0152] Long cable parameters: length is 6km, unit inductance L=0.25mH / km, unit capacitance C=0.15uF / km. Total inductance , total capacitance .

[0153] Sensors: ±0.2% voltage transformer, ±0.5% current transformer (sampling frequency 10kHz), ±0.3℃ temperature sensor, ±2%RH humidity sensor (sampling frequency 1Hz).

[0154] Load monitoring: Power analyzer (VITREK PA910, accuracy ±0.1%).

[0155] Data acquisition: The voltage and current at the inverter output are collected and sampled at 10kHz through a high-precision transformer to capture 0-40 harmonic components (such as fundamental wave 50Hz, 5th harmonic component 250Hz, 7th harmonic component 350Hz, etc.).

[0156] Environmental parameters: Real-time recording of temperature and humidity data at the cable wellhead (temperature 35℃~45℃, humidity 80%~90%RH).

[0157] Load fluctuation: monitor the load power change rate of the submersible motor (20% to 100%).

[0158] Low-pass filtering: Filter the voltage and current signals with a cutoff frequency of 8kHz to eliminate high-frequency noise interference.

[0159] Normalization processing: Normalize the harmonic components, temperature and humidity, and load fluctuation rate to a unified dimension to facilitate model input.

[0160] The theoretical model calculates the initial impedance characteristics based on the transmission line model. The theoretical resonant frequency of the inverter long-line drive system is calculated to be approximately 2.6kHz.

[0161] Dynamic correction: adjust the capacitance value according to the real-time temperature and humidity data (the capacitance increases by 4% for every 10°C increase in temperature). The theoretical resonant frequency of the inverter long-line drive system after correction .

[0162] Sweep frequency experiment calibration: Sweep frequency test is performed in the range of 0.5~4kHz, and the actual resonant frequency is measured at 3kHz.

[0163] Dynamically compensate for deviations between the theoretical model and actual operating conditions for subsequent prediction corrections. This ensures the impedance characteristic modeling error is controlled within 5%, providing an accurate benchmark for resonance suppression.

[0164] Implementation process:

[0165] Time window division: Set the time window to 150ms (including 1500 sampling points) to cover the complete cycle of the load mutation (for example, a sudden increase in load from 30% to 80%).

[0166] Input layer (32 nodes): Receives normalized harmonic components, temperature and humidity, and load fluctuation rate.

[0167] LSTM layer 1 (64 units): Extracts fundamental wave amplitude fluctuations and harmonic energy distribution characteristics (for example, when the fundamental wave amplitude drops by 10%, the resonant frequency shifts by 0.5 kHz).

[0168] LSTM layer 2 (64 units): learns the effects of sudden load changes and environmental disturbances (such as capacitance changes caused by increased humidity) on the resonant frequency.

[0169] Attention layer: Dynamically weights key features to enhance sensitivity to sudden load changes and high humidity scenarios.

[0170] Output layer (3 nodes): outputs the predicted resonant frequency , Theoretical resonant frequency of the inverter long-line drive system , frequency error compensation .

[0171] Training data: 800 sets of measured data and 400 sets of simulated data were used, covering scenarios with cable lengths of 5 to 10 km, temperatures of -10°C to 50°C, and load fluctuations of 10% to 100%.

[0172] Optimization results: prediction error RMSE = 0.3Hz, training time 18 minutes, an 85% improvement over the traditional method (RMSE = 2.0Hz).

[0173] The resonant frequency prediction accuracy reaches 97%, and the predicted resonant frequency can be dynamically corrected. , covering the impact of ambient temperature, humidity and load disturbance.

[0174] Adaptive multi-notch filter parameter adjustment

[0175] Purpose: Dynamically suppress resonance and balance response speed and stability.

[0176] Implementation process:

[0177] Center frequency correction:

[0178] The calculation expression for the corrected center frequency is:

[0179]

[0180] in, To correct the center frequency, is the proportionality coefficient, To predict the resonant frequency, is the frequency error compensation amount, is the reference frequency.

[0181] in, .

[0182] The calculation expression of bandwidth is:

[0183]

[0184] in, is the total harmonic distortion, .

[0185] When THD=4.0%, the bandwidth is set to 200Hz, covering the ±2 times resonant frequency fluctuation range.

[0186] Real-time resonance suppression, resonance amplitude attenuation rate> 90% (measured resonance peak value dropped from 100V to 8V); THD dropped from 6.5% to 3.5%, dynamic response time < 4ms.

[0187] Through comprehensive performance indicators Achieve closed-loop iterative optimization to ensure long-term stability of the system.

[0188]

[0189] Obtained by calculation .

[0190] Normal state: If , maintain the current parameters, re-collect data and update the model every 15 minutes.

[0191] Exceeding the standard: If , triggering the NSGA-II algorithm to optimize parameters 、 、 , until the criterion is met.

[0192] If convergence fails after three consecutive iterations, a preset passive filter (cutoff frequency 1 kHz) is enabled to forcibly suppress resonance.

[0193] Send an alarm signal to the host computer: "System abnormality, manual calibration required."

[0194] Expected effect: In high temperature and high humidity environment (45℃, 90%RH), the system automatically relaxes the threshold to , and still maintain the resonance amplitude attenuation rate>85%; the response time under abnormal conditions is <8ms, ensuring equipment safety.

[0195] In this example, the resonant frequency prediction error (RMSE) was 0.3 Hz, with an accuracy of 97%. The resonant amplitude was attenuated by 93%, the THD was reduced to 3.5%, and the dynamic response time was 3.8 ms. In a 6 km cable scenario, the system is compatible with temperatures ranging from -10°C to 50°C and load fluctuations of 20% to 100%. Through real-time feedback and automatic optimization, the frequency of manual intervention was reduced by 60%. The implementation process from data acquisition, model prediction, notch filter adjustment, to closed-loop control was fully demonstrated. This solution provides an efficient and intelligent resonance suppression solution for 6 km long-line submersible motor drives, with significant engineering application value.

[0196] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for suppressing resonance of a long-line drive system of a frequency converter, characterized in that: include: S1: Obtain the operating parameters of the inverter long-line drive system in real time through the data acquisition unit, obtain time series data, normalize the time series data, and obtain standard time series data; S2: Calculate the theoretical resonant frequency based on the operating parameters of the inverter long-line drive system; S3: Predict the resonant frequency of the standard time series data through the attention-enhanced long short-term memory neural network to obtain the predicted resonant frequency; S31: Enhance the weights of the first LSTM neural network layer and the second LSTM neural network layer to extract key resonance features through the attention mechanism layer; S32: extracting the time-frequency features of the standard time series data through the first long short-term memory neural network layer to obtain the state features of the first hidden layer; S33: extracting the frequency and amplitude of the first hidden layer state feature through the second long short-term memory neural network layer to obtain the second hidden layer state feature; S34: performing weighted calculation on the first hidden layer state feature and the second hidden layer state feature according to the weight coefficient to obtain a predicted resonant frequency; S4: Calculate the frequency error compensation amount based on the predicted resonant frequency and the theoretical resonant frequency, and calculate the comprehensive performance evaluation index of harmonic suppression based on the frequency error compensation amount and the predicted resonant frequency; S5: Evaluate the harmonic suppression effect based on the harmonic suppression comprehensive performance evaluation index and the index threshold. If the harmonic suppression comprehensive performance evaluation index is greater than or equal to the index threshold, execute step S6; if the harmonic suppression comprehensive performance evaluation index is less than the index threshold, execute step S7; S6: Adjust the weight coefficient of the attention-enhanced long short-term memory neural network through the NSGA-II optimization algorithm, and repeat steps S3 to S5; S7: Adjust the adaptive multiple notch filter parameters in the inverter long-line drive system according to the predicted resonant frequency to achieve resonance suppression of the inverter long-line drive system.

2. The method for suppressing resonance of a long-line drive system of a frequency converter according to claim 1, characterized in that: In step S2, the operating parameters of the inverter long-line drive system include cable inductance and cable capacitance, and the theoretical resonant frequency of the inverter long-line drive system is calculated based on the cable inductance and cable capacitance; The theoretical resonant frequency calculation expression of the inverter long-line drive system is: ； in, is the theoretical resonant frequency of the inverter long-line drive system, is the cable inductance per unit length, is the cable capacitance per unit length.

3. The method for suppressing resonance of a long-line drive system of a frequency converter according to claim 1, characterized in that: The attention-enhanced long short-term memory neural network includes an input layer, a first long short-term memory neural network layer, a second long short-term memory neural network layer, an attention mechanism layer, and an output layer.

4. The method for suppressing resonance of a long-line drive system of a frequency converter according to claim 1, characterized in that: In step S31, the calculation expression of attention weight is: ； in, For the Step attention weight, For the The first hidden layer state features, For the Step 2 hidden layer state features, W is the first weight matrix, V is the second weight matrix, U is the third weight matrix, is the activation function, is the tangent function.

5. The method for suppressing resonance of a long-line drive system of a frequency converter according to claim 1, characterized in that: The calculation expression for predicting the resonant frequency is: ； in, To predict the resonant frequency, is the reference frequency, is the state feature of the first hidden layer, is the state feature of the second hidden layer, is the weight coefficient of the first hidden layer, is the weight coefficient of the second hidden layer.

6. The method for suppressing resonance of a long-line drive system of a frequency converter according to claim 1, characterized in that: In step S4, the calculation expression of the frequency error compensation amount is: ； in, is the frequency error compensation amount, To predict the resonant frequency, is the theoretical resonant frequency of the inverter long-line drive system; The calculation expression of comprehensive performance evaluation index is: ； in, is a comprehensive performance evaluation index. is the inverter switching frequency, is the mean of the attention weights.

7. The method for suppressing resonance of a long-line drive system of a frequency converter according to claim 1, characterized in that: The indicator threshold is divided into a first key indicator threshold and a second key indicator threshold by using the K-means clustering method, and the harmonic suppression effect is evaluated according to the first key indicator threshold and the second key indicator threshold; when When , the inhibition effect is optimal; when When , the inhibitory effect is suboptimal; when When , the inhibitory effect is poor; in, is a comprehensive performance evaluation index. is the first key indicator threshold, is the second key indicator threshold.

8. The method for suppressing resonance of a long-line drive system of a frequency converter according to claim 1, characterized in that: In step S7, the inverter long-line drive system includes an adaptive multi-notch filter, and the adaptive multi-notch filter parameters include the center frequency and bandwidth. The calculation expression for the corrected center frequency is: ； in, To correct the center frequency, is the proportionality coefficient, To predict the resonant frequency, is the frequency error compensation amount, is the reference frequency; The calculation expression of bandwidth is: ； in, is the bandwidth, is the regulating factor, is the total harmonic distortion.

9. A device for suppressing resonance of a long-line drive system of a frequency converter, characterized in that: The method for suppressing resonance of a long-line drive system of a frequency converter according to any one of claims 1 to 8 comprises: A data acquisition unit, which collects operating parameters of the inverter long-line drive system through a voltage transformer and a current transformer; A local processing unit, wherein the local processing unit performs data preprocessing; a central control unit that performs attention-enhanced long short-term memory neural network prediction and harmonic suppression control; A human-computer interaction interface displays the system operation status.

Images