Motor driving intelligent control system and method based on multi-modal data fusion

Abstract

The invention discloses a motor driving intelligent control system and method based on multi-modal data fusion, and relates to the technical field of motor intelligent control. The system comprises a reference feature construction module, a real-time data processing module, an anomaly recognition and judgment module and an intelligent control and update module. The reference feature construction module is used for extracting features and association relationships and establishing a reference feature set by analyzing multi-mode historical data of a full life cycle of a target motor; the real-time data processing module collects and analyzes real-time data, and extracts real-time features and association relationships; the anomaly recognition and judgment module compares the real-time features with the reference, recognizes anomaly and judges the fluctuation type; and the intelligent control and update module adjusts driving parameters according to a judgment result, periodically updates a reference or reconstructs the reference, realizes personalized and precise intelligent control of the motor, and improves operation stability and safety.

Description

Technical Field

[0001] This invention relates to the field of intelligent motor control technology, specifically to an intelligent control system and method for motor drive based on multimodal data fusion. Background Technology

[0002] In the field of industrial production and automation, electric motors, as core power equipment, are directly related to production efficiency and system safety through accurate monitoring and intelligent control of their operating status. In recent years, multimodal data fusion technology, with its comprehensive analysis capabilities of multi-dimensional information such as current, voltage, vibration, and temperature, has gradually become a key support for intelligent control systems of motor drives. It aims to improve the accuracy of condition assessment and fault diagnosis through correlation mining of cross-modal data.

[0003] However, in practical applications, motors of the same model exhibit significant individual differences due to manufacturing process errors, variations in installation conditions, and different degrees of aging. For example, some motors have vibration baselines higher than the general standard at the time of manufacture, yet they can operate stably for a long time; after the performance of older motors degrades, the general threshold can easily lead to false shutdowns; and customized motors lack standard references, making it difficult for existing fusion logic to adapt. These problems result in multimodal data fusion based on general standards having poor adaptability and a high false judgment rate, failing to meet the personalized and precise control needs of motors in complex industrial scenarios. Therefore, a fusion solution that can adapt to the individual characteristics of motors is urgently needed. Summary of the Invention

[0004] The purpose of this invention is to provide a motor drive intelligent control system and method based on multimodal data fusion to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: The intelligent control method for motor drives based on multimodal data fusion includes the following steps: S1. Collect multimodal operation data of the target motor during the historical operation stages of the entire life cycle, analyze the multimodal operation data to extract the features of each modal data and the correlation between different modal features; based on the above features and correlations, establish a multimodal benchmark feature set of the target motor; S2. During the operation of the target motor, real-time multimodal operation data is collected. The real-time multimodal operation data is analyzed in accordance with the analysis method of multimodal operation data in S1, so as to extract the corresponding real-time features and real-time modal correlations. S3. Compare the real-time features with the features corresponding to the multimodal baseline feature set to generate the dynamic change trajectory of the real-time features, identify abnormal feature points in the dynamic change trajectory and mark them as drift features; for drift features, calculate the deviation between the corresponding real-time modal correlation and the correlation in the multimodal baseline feature set. If the deviation is within a preset range, it is determined to be a single modal occasional fluctuation; if the deviation exceeds the preset range, it is determined to be a multimodal collaborative anomaly. S4. When the fluctuation is determined to be a single-mode intermittent fluctuation, maintain the current drive parameters of the motor; when the abnormality is determined to be a multi-mode coordinated anomaly, output the corresponding drive parameter adjustment command to correct the abnormal state according to the mode type involved in the anomaly; periodically collect multi-mode operation data when the target motor is running normally, update the multi-mode reference feature set, and if the motor is repaired or the operating conditions are changed, re-execute S1 to construct a new multi-mode reference feature set.

[0006] S1 includes the following: Collect multimodal operating data of the target motor during the historical operating stages of the entire life cycle. Specifically, collect multimodal operating data of the target motor under typical operating conditions, including no-load, rated load, overload and start-stop process. The multimodal operating data includes at least current, vibration, temperature and speed data. The sampling frequency of each mode data is consistent, and the collection time covers the stable operating stage of each typical operating condition. The collected multimodal operation data is preprocessed, including denoising, data interpolation, and time base alignment. Temporal and frequency domain features are extracted from the preprocessed multimodal operation data. The temporal features are then processed according to formula F. t =f t (x t ) Calculate, x t For time-domain data in multimodal operational data, f t These are time-domain characteristic functions (such as RMS value, mean, rate of change, etc.); where the frequency-domain characteristics are expressed by formula F. p =f p (X p ) Calculate, X p f is the frequency domain transformation result of multimodal operation data. p For frequency domain characteristic functions (such as the proportion of harmonic components, the amplitude of the dominant frequency, etc.); Select feature pairs with physical coupling relationships from the time-domain and frequency-domain features, denoted as F1 and F2 respectively; calculate the dynamic correlation coefficient C1 between the two features, and C1 = cov(F1,F2) / [σ(F1)×σ(F2)], where cov(F1,F2) is the covariance of F1 and F2, σ(F1) is the standard deviation of F1, and σ(F2) is the standard deviation of F2; calculate the time delay correlation coefficient C2 between the two features, and C2 = ∑ t[F1'(t)×F2'(t+τ)], where F1'(t) is the F1 eigenvalue of the normalized time t, F2'(t+τ) is the F2 eigenvalue of the normalized time t+τ, and τ is the response delay time determined by cross-correlation analysis; A multimodal baseline feature set is established, which includes the normal fluctuation range of each modal feature determined by [μ±k×σ], where μ is the mean, σ is the standard deviation, and k is the interval coefficient; as well as the normal range of dynamic correlation coefficient [C1_min,C1_max] and the judgment range of time delay correlation value [C2_min,C2_max]. The above ranges are all determined by the normal operation data of the target motor in the historical operation stage of the entire life cycle in the database.

[0007] S2 includes the following: During the operation of the target motor, real-time multimodal operation data of the same type as in S1 is collected and continuously acquired using the same sampling frequency as in S1. The real-time multimodal operation data is processed according to the preprocessing method of S1. Real-time features are extracted from the preprocessed real-time multimodal operation data according to the feature extraction method of S1, and the real-time time-domain features are expressed according to formula F. t_real =f t (x t_real Calculate x t_real For real-time time-domain data, f t The time-domain characteristic function selected in S1; the frequency-domain characteristic is expressed by formula F. p_real =f p (X p_real ) Calculate, X p_real f is the result obtained by transforming real-time multimodal operating data into the same frequency domain in S1. p The frequency domain characteristic function selected in S1; Referring to the physical coupling feature pairs (F1, F2) selected in S1, the real-time modal correlation is calculated, and the formula for calculating the real-time dynamic correlation coefficient C1 is: C1_real=cov(F1_real,F2_real) / [σ(F1_real)×σ(F2_real)], where cov(F1_real,F1_real) is the covariance of the real-time feature pair, σ(F1_real) is the standard deviation of real-time feature F1, and σ(F2_real) is the standard deviation of real-time feature F2; the formula for calculating the real-time delay correlation coefficient C2_real is: C2_real=∑ t[F1'_real(t)×F2'_real(t+τ)], where F1'_real(t) is the eigenvalue of the real-time feature F1 at time t after normalization, F2'_real(t+τ) is the eigenvalue of the real-time feature F2 at time t+τ after normalization, and τ is the response delay time determined in S1; thus, real-time features and real-time modal correlations consistent with the features and correlation dimensions in S1 are obtained.

[0008] S3 includes the following: The real-time features are mapped and compared with the features corresponding to the multimodal benchmark feature set to generate the real-time feature dynamic change trajectory with time as the axis. The real-time feature dynamic change trajectory is divided into several segments corresponding to the typical working conditions in S1, so that the real-time features are compared with the benchmark features under the same working conditions. For each segment of the real-time feature dynamic change trajectory, calculate the feature change slope K_real within the sliding window, and K_real = [F real (t)-F real (tT)] / T, where T represents the sliding window duration; if K_real exceeds 1.2 times the slope range [K_min,K_max] of the most recent same working condition in S1, it is marked as a trend anomaly; for the segment of the real-time feature dynamic change trajectory marked as a trend anomaly, the instantaneous mutation value ΔF is calculated for each corresponding real-time feature point, and ΔF=|F real (t)-F real (t-1)|, if the instantaneous mutation value ΔF>α×σ, it is marked as a mutation outlier, where α represents the mutation coefficient and σ represents the standard deviation of the corresponding feature in S1; for mutation outliers, check whether other modal features at the corresponding time are consistent with the trend change of the mutation outlier. If they are inconsistent, they are judged as isolated anomalies, and if they are consistent, they are marked as drift features. For drift features, the deviation between the corresponding real-time modal correlation and the correlation between the multimodal baseline feature set is calculated, specifically including: The median of the normal range [C1_min, C1_max] of the most recent dynamic correlation coefficient under the same working condition in the multimodal benchmark feature set is obtained as the benchmark reference value C1_ref, and C1_ref=(C1_min+C1_max) / 2. The relative deviation ΔC1 between the real-time dynamic correlation coefficient and the benchmark reference value is calculated, and ΔC1=|C1_real-C1_ref| / C1_ref. Similarly, the median of the time delay correlation value judgment range [C2_min, C2_max] of the most recent time delay correlation value under the same working condition in the multimodal benchmark feature set is obtained as the benchmark reference value C2_ref, and C2_ref=(C2_min+C2_max) / 2. The relative deviation ΔC2 between the real-time time delay correlation coefficient and the benchmark reference value is calculated, and ΔC2=|C2_real-C2_ref| / C2_ref. A preset dynamic correlation deviation threshold γ1 and a time delay correlation deviation threshold γ2 are established. Both γ1 and γ2 are determined based on the fluctuation range of the correlation relationship of historical normal data under the same operating conditions in the multimodal reference feature set. The values ​​of γ1 and γ2 are dynamically adjusted according to the motor operation stage (γ1 and γ2 are 0.1-0.15 in the initial stage, 0.15-0.2 in the middle stage, and 0.2-0.25 in the later stage). If ΔC1≤γ1 and ΔC2≤γ2, it is judged as a single-mode occasional fluctuation. If ΔC1>γ1 or ΔC2>γ2, it is judged as a multimodal cooperative anomaly.

[0009] S4 includes the following: When the anomaly is determined to be a single-mode intermittent fluctuation, the current drive parameters of the target motor are kept unchanged. When the anomaly is determined to be a multi-mode coordinated anomaly, the corresponding drive parameter adjustment command is output according to the type of physical coupling characteristics involved in the anomaly. The excess deviation ΔE of the anomaly exceeding the threshold is calculated, and ΔE=max(ΔC1-γ1,ΔC2-γ2), and the excess deviation ΔE is the basis for drive parameter adjustment. The drive parameter adjustment amount ΔP is calculated, and ΔP=β×ΔE, where β is the adjustment coefficient corresponding to the drive parameter, and the corresponding value range needs to be determined by relevant personnel based on the physical coupling strength and the safe operation boundary of the motor. Based on the adjustment amount ΔP of the drive parameters, the drive parameters are adjusted to P_new = P_current + ΔP, where P_current is the current drive parameter value. After adjustment, the motor operating status is monitored in real time, and real-time features and real-time modal correlations are continuously extracted according to the analysis method of S2. If ΔC1≤γ1 and ΔC2≤γ2 within 3 consecutive sampling periods, the drive parameters are restored to the current value P_current before adjustment. If it does not return to the normal range, ΔE' and ΔP' are recalculated based on the new real-time correlation deviation, and ΔE' = max(ΔC1'-γ1, ΔC2'-γ2), ΔP' = β × ΔE', and the adjustment process is repeated. When the cumulative number of adjustments reaches the preset upper limit and the abnormality is not eliminated, or the real-time feature after adjustment exceeds 1.5 times the normal fluctuation range of the corresponding feature in S1, the motor protection shutdown command is triggered, and the multimodal data at the time of the abnormality and the adjustment process log are recorded. Regularly collect multimodal operation data of the target motor during continuous normal operation, and update the multimodal benchmark feature set. The mean of the new benchmark feature is μ_new=ω×μ_old+(1-ω)×μ_add, where ω is the historical benchmark weight, with a value of 0.6-0.8. The standard deviation of the new benchmark feature is σ_new={[ω×σ_old²+(1-ω)×σ_add²]}^(1 / 2), where μ_add and σ_add are the mean and standard deviation of the newly added normal data. The normal interval of the dynamic correlation coefficient and the judgment range of the time delay correlation value are updated synchronously. The boundary of the new interval is the weighted average of the boundary of the historical interval and the statistical boundary of the newly added data. When the target motor is repaired (such as replacing core components) or the operating conditions change abruptly (load rate change ≥30%, ambient temperature change ≥20℃), S1 is re-executed to construct a new multimodal benchmark feature set. The original benchmark feature set is archived and stored for full life cycle comparative analysis.

[0010] The intelligent control system for motor drive based on multimodal data fusion includes: a reference feature construction module, a real-time data processing module, an anomaly identification and judgment module, and an intelligent control and update module; The benchmark feature construction module collects multimodal operation data from the historical operation stages of the target motor throughout its entire life cycle, analyzes the multimodal operation data to extract the features of each modal data and the correlation between different modal features; based on the above features and correlations, a multimodal benchmark feature set for the target motor is established. During the operation of the target motor, the real-time data processing module collects real-time multimodal operation data and analyzes the real-time multimodal operation data in accordance with the analysis method of multimodal operation data, thereby extracting the corresponding real-time features and real-time modal correlations. The anomaly identification and judgment module compares the real-time features with the features corresponding to the multimodal benchmark feature set, generates the dynamic change trajectory of the real-time features, identifies abnormal feature points in the dynamic change trajectory and marks them as drift features; for drift features, it calculates the deviation between the corresponding real-time modal correlation and the correlation in the multimodal benchmark feature set. If the deviation is within a preset range, it is judged as a single modal occasional fluctuation; if the deviation exceeds the preset range, it is judged as a multimodal collaborative anomaly. When the intelligent control and update module determines that the fluctuation is a single-mode intermittent fluctuation, it maintains the current drive parameters of the motor; when it determines that the multi-mode coordinated anomaly is, it outputs the corresponding drive parameter adjustment command to correct the abnormal state according to the mode type involved in the anomaly; it periodically collects multi-mode operation data when the target motor is running normally, updates the multi-mode reference feature set, and reconstructs a new multi-mode reference feature set if the motor is repaired or the operating conditions are changed.

[0011] The benchmark feature construction module includes a multimodal data acquisition unit, a feature extraction and correlation analysis unit, and a benchmark feature set generation unit; The multimodal data acquisition unit is used to collect multimodal operation data under typical operating conditions throughout the entire life cycle of the target motor, with consistent sampling frequency for each mode and acquisition duration covering the stable operating phase. The feature extraction and correlation analysis unit is used to preprocess the collected multimodal operation data, extract time-domain and frequency-domain features, and calculate the dynamic correlation coefficient and time delay correlation coefficient of feature pairs with physical coupling relationship. The benchmark feature set generation unit is used to determine the normal fluctuation range of each modal feature, the normal range of dynamic correlation coefficient, and the judgment range of time delay correlation value based on historical normal operation data, and establish a multimodal benchmark feature set.

[0012] The real-time data processing module includes a real-time multimodal data acquisition unit, a real-time feature extraction unit, and a real-time correlation calculation unit; The real-time multimodal data acquisition unit is used to continuously acquire real-time multimodal operation data of the same type as the benchmark feature construction module during the operation of the target motor, and the sampling frequency is consistent with the data of the benchmark feature construction module. The real-time feature extraction unit is used to process the real-time multimodal operation data and extract the corresponding real-time time domain features and real-time frequency domain features by referring to the preprocessing and feature extraction methods of the benchmark feature construction module. The real-time correlation calculation unit is used to calculate the real-time dynamic correlation coefficient and the real-time delay correlation coefficient by referring to the physical coupling feature pairs selected in the benchmark feature construction module, and obtain the real-time correlation consistent with the benchmark dimension.

[0013] The anomaly identification and judgment module includes a feature trajectory comparison unit, an anomaly point verification and marking unit, and an anomaly type judgment unit; The feature trajectory comparison unit is used to map and compare real-time features with corresponding features in the benchmark feature set, generate real-time feature dynamic change trajectories and divide them into segments corresponding to typical working conditions, and calculate the feature change slope within the sliding window to identify trend anomalies; the anomaly point verification and marking unit is used to calculate instantaneous mutation values ​​for trend anomalies to screen mutation anomalies, exclude isolated anomalies by verifying the trend consistency of multimodal features, and mark drift features; the anomaly type determination unit is used to calculate the deviation between the real-time correlation corresponding to the drift feature and the benchmark correlation, and combine the dynamically adjusted deviation threshold to determine single-modal occasional fluctuations or multimodal collaborative anomalies.

[0014] The intelligent control and update module includes a drive parameter adjustment unit, a closed-loop control unit, and a reference dynamic update unit; The drive parameter adjustment unit is used to maintain the current drive parameters when the fluctuation is determined to be a single-mode intermittent fluctuation, and to output the corresponding drive parameter adjustment command according to the mode type involved in the anomaly when the anomaly is determined to be a multi-mode coordinated anomaly. The closed-loop control unit is used to monitor the motor operating status after adjustment. If the anomaly is eliminated, the original parameters are restored. If the cumulative adjustment exceeds the limit, the protection shutdown is triggered. The benchmark dynamic update unit is used to periodically collect normal operation data to update the multi-mode benchmark feature set. When the motor is under maintenance or the operating condition changes suddenly, the benchmark feature construction process is re-executed and the historical benchmark feature set is archived.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention establishes a unique multimodal benchmark feature set by collecting historical data of the target motor throughout its entire life cycle, and dynamically adjusts the judgment threshold in conjunction with its operating stages, ensuring that the monitoring and control logic is fully adapted to individual characteristics, thus solving problems such as "high misjudgment rate of general standards" and "no reference benchmark for customized motors." This invention employs a three-layer progressive logic of "trend anomaly → sudden anomaly → correlation verification," combined with the physical coupling relationship of multimodal features and dynamic and time-delay correlation coefficient deviation analysis, to accurately distinguish between "single-modal occasional fluctuations" and "multimodal collaborative anomalies," significantly reducing the misjudgment rate, especially showing greater sensitivity to the identification of early-stage latent faults in motors. For multimodal collaborative anomalies, this invention outputs differentiated adjustment commands based on the excess deviation ΔE and physical coupling type, and continuously optimizes through closed-loop feedback, achieving intelligent control of "precise correction rather than simple shutdown." Simultaneously, by periodically updating the benchmark feature set, the system can adapt to the long-term degradation of motor performance, extending the effective monitoring cycle. This invention constructs a complete closed loop of "benchmark establishment → real-time monitoring → anomaly identification → dynamic control and benchmark update", covering the entire stage of motor from break-in period to aging period; even after maintenance or sudden changes in operating conditions, the monitoring and control logic can be reconstructed to ensure that it always matches the current state of the motor, thus providing a guarantee for efficient and safe operation throughout the entire life cycle. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the intelligent control method for motor drive based on multimodal data fusion according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 The present invention provides the following technical solution: The intelligent control method for motor drives based on multimodal data fusion includes the following steps: S1. Collect multimodal operation data of the target motor during the historical operation stages of the entire life cycle, analyze the multimodal operation data to extract the features of each modal data and the correlation between different modal features; based on the above features and correlations, establish a multimodal benchmark feature set of the target motor; S2. During the operation of the target motor, real-time multimodal operation data is collected. The real-time multimodal operation data is analyzed in accordance with the analysis method of multimodal operation data in S1, so as to extract the corresponding real-time features and real-time modal correlations. S3. Compare the real-time features with the features corresponding to the multimodal baseline feature set to generate the dynamic change trajectory of the real-time features, identify abnormal feature points in the dynamic change trajectory and mark them as drift features; for drift features, calculate the deviation between the corresponding real-time modal correlation and the correlation in the multimodal baseline feature set. If the deviation is within a preset range, it is determined to be a single modal occasional fluctuation; if the deviation exceeds the preset range, it is determined to be a multimodal collaborative anomaly. S4. When the fluctuation is determined to be a single-mode intermittent fluctuation, maintain the current drive parameters of the motor; when the abnormality is determined to be a multi-mode coordinated anomaly, output the corresponding drive parameter adjustment command to correct the abnormal state according to the mode type involved in the anomaly; periodically collect multi-mode operation data when the target motor is running normally, update the multi-mode reference feature set, and if the motor is repaired or the operating conditions are changed, re-execute S1 to construct a new multi-mode reference feature set.

[0019] S1 includes the following: Collect multimodal operating data of the target motor throughout its entire life cycle and historical operating phases. Specifically, collect multimodal operating data of the target motor under typical operating conditions, including no-load, rated load, overload and start-stop processes. The multimodal operating data should include at least current, vibration, temperature and speed data. The sampling frequency of each mode data should be consistent and the collection duration should cover the stable operating phase of each typical operating condition. The collected multimodal operation data is preprocessed, including denoising, data interpolation, and time base alignment. Temporal and frequency domain features are extracted from the preprocessed multimodal operation data. The temporal features are then processed according to formula F. t =f t (x t ) Calculate, x t For time-domain data in multimodal operational data, f t These are time-domain characteristic functions (such as RMS value, mean, rate of change, etc.); where the frequency-domain characteristics are expressed by formula F. p =f p (X p ) Calculate, X p f is the frequency domain transformation result of multimodal operation data. p For frequency domain characteristic functions (such as the proportion of harmonic components, the amplitude of the dominant frequency, etc.); Select feature pairs with physical coupling relationships from the time-domain and frequency-domain features, denoted as F1 and F2 respectively; calculate the dynamic correlation coefficient C1 between the two features, and C1 = cov(F1,F2) / [σ(F1)×σ(F2)], where cov(F1,F2) is the covariance of F1 and F2, σ(F1) is the standard deviation of F1, and σ(F2) is the standard deviation of F2; calculate the time delay correlation coefficient C2 between the two features, and C2 = ∑ t [F1'(t)×F2'(t+τ)], where F1'(t) is the F1 eigenvalue of the normalized time t, F2'(t+τ) is the F2 eigenvalue of the normalized time t+τ, and τ is the response delay time determined by cross-correlation analysis; A multimodal baseline feature set is established, which includes the normal fluctuation range of each modal feature determined by [μ±k×σ], where μ is the mean, σ is the standard deviation, and k is the interval coefficient; as well as the normal range of dynamic correlation coefficient [C1_min,C1_max] and the judgment range of time delay correlation value [C2_min,C2_max]. The above ranges are all determined by the normal operation data of the target motor in the historical operation stage of the entire life cycle in the database.

[0020] In this embodiment, taking a 3.7kW asynchronous motor (rated speed 1440r / min) as an example, the implementation process of S1 is as follows: Data on current, vibration, temperature, and speed were collected under no-load and rated load conditions (sampling frequency 10kHz, 2 hours per condition, covering the motor break-in period, stable operation period, and early aging period). DB4 wavelet denoising was used, with 6 wavelet decomposition levels and a soft thresholding function. Missing data were filled using current-temperature correlation interpolation, with a linear regression model of T = 0.8I + 25 (T is temperature, I is current). When temperature data was missing, the missing value was calculated by substituting the real-time current value into the model. The rising edge of the speed pulse was used as the synchronous trigger signal, and the time difference between the sampling time of each modal data and the trigger signal was corrected using a linear calibration algorithm to ensure a delay ≤ 0.5ms after time base alignment. The effective current value F was extracted. t1 Temperature change rate F t2 Isochronous time domain characteristics, and the proportion of the third harmonic of the current F p1 Vibration amplitude F at 24Hz p2 Equal frequency domain characteristics; Assuming that relevant personnel select an electromagnetic-mechanical coupling pair (F) based on the physical coupling relationship. p1 With F p2 ) and electro-thermal coupling pairs (F t1 With F t2 ), calculate to get cov(F p1 ,F p2 )=0.32, σ(F p1 )=0.45, σ(F p2 =0.71, according to the C1 calculation formula, C1 = 0.32 / (0.45×0.71) = 1.0; for the electro-thermal coupling pair (F t1 With F t2 Cross-correlation analysis was performed, with a sliding window length of 10s, a correlation threshold of 0.8, and a response delay time τ = 8s. For F... t1 and F t2 After performing min-max normalization, C2 is calculated as 4.2 within the summation interval of 30s using the C2 formula. A multimodal baseline feature set was established, based on statistical assumptions made from 50 sets of normal operation data covering different stages of the entire life cycle of the motor: F t1 The normal range is [6.9, 8.7]A (μ=7.8, σ=0.3, k=3); the C1 range is [-1.0, 1.0], and the C2 range is [3.5, 5.0].

[0021] S2 includes the following: During the operation of the target motor, real-time multimodal operation data of the same type as in S1 is collected and continuously acquired using the same sampling frequency as in S1. The real-time multimodal operation data is processed according to the preprocessing method of S1. Real-time features are extracted from the preprocessed real-time multimodal operation data according to the feature extraction method of S1, and the real-time time-domain features are expressed according to formula F. t_real =f t (x t_real Calculate x t_real For real-time time-domain data, f t The time-domain characteristic function selected in S1; the frequency-domain characteristic is expressed by formula F. p_real =f p (X p_real ) Calculate, X p_real f is the result obtained by transforming real-time multimodal operating data into the same frequency domain in S1. p The frequency domain characteristic function selected in S1; Referring to the physical coupling feature pairs (F1, F2) selected in S1, the real-time modal correlation is calculated, and the formula for calculating the real-time dynamic correlation coefficient C1 is: C1_real=cov(F1_real,F2_real) / [σ(F1_real)×σ(F2_real)], where cov(F1_real,F1_real) is the covariance of the real-time feature pair, σ(F1_real) is the standard deviation of real-time feature F1, and σ(F2_real) is the standard deviation of real-time feature F2; the formula for calculating the real-time delay correlation coefficient C2_real is: C2_real=∑ t [F1'_real(t)×F2'_real(t+τ)], where F1'_real(t) is the eigenvalue of the real-time feature F1 at time t after normalization, F2'_real(t+τ) is the eigenvalue of the real-time feature F2 at time t+τ after normalization, and τ is the response delay time determined in S1; thus, real-time features and real-time modal correlations consistent with the features and correlation dimensions in S1 are obtained.

[0022] S3 includes the following: The real-time features are mapped and compared with the features corresponding to the multimodal benchmark feature set to generate the real-time feature dynamic change trajectory with time as the axis. The real-time feature dynamic change trajectory is divided into several segments corresponding to the typical working conditions in S1, so that the real-time features are compared with the benchmark features under the same working conditions. For each segment of the real-time feature dynamic change trajectory, calculate the feature change slope K_real within the sliding window, and K_real = [F real (t)-F real(tT)] / T, where T represents the sliding window duration; if K_real exceeds 1.2 times the slope range [K_min,K_max] of the most recent same working condition in S1, it is marked as a trend anomaly; for the segment of the real-time feature dynamic change trajectory marked as a trend anomaly, the instantaneous mutation value ΔF is calculated for each corresponding real-time feature point, and ΔF=|F real (t)-F real (t-1)|, if the instantaneous mutation value ΔF>α×σ, it is marked as a mutation outlier, where α represents the mutation coefficient and σ represents the standard deviation of the corresponding feature in S1; for mutation outliers, check whether other modal features at the corresponding time are consistent with the trend change of the mutation outlier. If they are inconsistent, they are judged as isolated anomalies, and if they are consistent, they are marked as drift features. For drift features, the deviation between the corresponding real-time modal correlation and the correlation between the multimodal baseline feature set is calculated, specifically including: The median of the normal range [C1_min, C1_max] of the most recent dynamic correlation coefficient under the same working condition in the multimodal benchmark feature set is obtained as the benchmark reference value C1_ref, and C1_ref=(C1_min+C1_max) / 2. The relative deviation ΔC1 between the real-time dynamic correlation coefficient and the benchmark reference value is calculated, and ΔC1=|C1_real-C1_ref| / C1_ref. Similarly, the median of the time delay correlation value judgment range [C2_min, C2_max] of the most recent time delay correlation value under the same working condition in the multimodal benchmark feature set is obtained as the benchmark reference value C2_ref, and C2_ref=(C2_min+C2_max) / 2. The relative deviation ΔC2 between the real-time time delay correlation coefficient and the benchmark reference value is calculated, and ΔC2=|C2_real-C2_ref| / C2_ref. A preset dynamic correlation deviation threshold γ1 and a time delay correlation deviation threshold γ2 are established. Both γ1 and γ2 are determined based on the fluctuation range of the correlation relationship of historical normal data under the same operating conditions in the multimodal reference feature set. The values ​​of γ1 and γ2 are dynamically adjusted according to the motor operation stage (γ1 and γ2 are 0.1-0.15 in the initial stage, 0.15-0.2 in the middle stage, and 0.2-0.25 in the later stage). If ΔC1≤γ1 and ΔC2≤γ2, it is judged as a single-mode occasional fluctuation. If ΔC1>γ1 or ΔC2>γ2, it is judged as a multimodal cooperative anomaly.

[0023] S4 includes the following: When the anomaly is determined to be a single-mode intermittent fluctuation, the current drive parameters of the target motor are kept unchanged. When the anomaly is determined to be a multi-mode coordinated anomaly, the corresponding drive parameter adjustment command is output according to the type of physical coupling characteristics involved in the anomaly. The excess deviation ΔE of the anomaly exceeding the threshold is calculated, and ΔE=max(ΔC1-γ1,ΔC2-γ2), and the excess deviation ΔE is the basis for drive parameter adjustment. The drive parameter adjustment amount ΔP is calculated, and ΔP=β×ΔE, where β is the adjustment coefficient corresponding to the drive parameter, and the corresponding value range needs to be determined by relevant personnel based on the physical coupling strength and the safe operation boundary of the motor. Based on the adjustment amount ΔP of the drive parameters, the drive parameters are adjusted to P_new = P_current + ΔP, where P_current is the current drive parameter value. After adjustment, the motor operating status is monitored in real time, and real-time features and real-time modal correlations are continuously extracted according to the analysis method of S2. If ΔC1≤γ1 and ΔC2≤γ2 within 3 consecutive sampling periods, the drive parameters are restored to the current value P_current before adjustment. If it does not return to the normal range, ΔE' and ΔP' are recalculated based on the new real-time correlation deviation, and ΔE' = max(ΔC1'-γ1, ΔC2'-γ2), ΔP' = β × ΔE', and the adjustment process is repeated. When the cumulative number of adjustments reaches the preset upper limit and the abnormality is not eliminated, or the real-time feature after adjustment exceeds 1.5 times the normal fluctuation range of the corresponding feature in S1, the motor protection shutdown command is triggered, and the multimodal data at the time of the abnormality and the adjustment process log are recorded. Regularly collect multimodal operation data of the target motor during continuous normal operation, and update the multimodal benchmark feature set. The mean of the new benchmark feature is μ_new=ω×μ_old+(1-ω)×μ_add, where ω is the historical benchmark weight, with a value of 0.6-0.8. The standard deviation of the new benchmark feature is σ_new={[ω×σ_old²+(1-ω)×σ_add²]}^(1 / 2), where μ_add and σ_add are the mean and standard deviation of the newly added normal data. The normal interval of the dynamic correlation coefficient and the judgment range of the time delay correlation value are updated synchronously. The boundary of the new interval is the weighted average of the boundary of the historical interval and the statistical boundary of the newly added data. When the target motor is repaired (such as replacing core components) or the operating conditions change abruptly (load rate change ≥30%, ambient temperature change ≥20℃), S1 is re-executed to construct a new multimodal benchmark feature set. The original benchmark feature set is archived and stored for full life cycle comparative analysis.

[0024] In this embodiment, when a multimodal coordination anomaly is determined, the corresponding driving parameter adjustment instruction is output according to the type of physical coupling characteristics involved in the anomaly: If the anomaly involves electromagnetic-mechanical coupling characteristics (such as current frequency domain characteristics and vibration frequency domain characteristics), adjust the driving frequency according to the formula Δf=k_f×max(ΔC1-γ1,ΔC2-γ2), where k_f is the frequency adjustment coefficient (valued at 0.5-2Hz / unit deviation), and the absolute value of Δf shall not exceed 5% of the rated frequency. If the anomaly involves an electro-thermal coupling characteristic pair (such as current time-domain characteristics and temperature time-domain characteristics), adjust the cooling system power according to the formula ΔP_cool=k_c×max(ΔC1-γ1,ΔC2-γ2), where k_c is the cooling power coefficient (value 0.1-0.3kW / unit deviation), ensuring that the winding temperature change rate after adjustment is ≤ the upper limit of the temperature change rate under the same operating condition in S1; After adjustment, real-time data is continuously collected. If ΔC1≤γ1 and ΔC2≤γ2 within 10 sampling periods, the parameters are restored to the baseline before adjustment; otherwise, the adjustment is repeated until the anomaly is eliminated or the protection shutdown is triggered.

[0025] The intelligent control system for motor drive based on multimodal data fusion includes: a reference feature construction module, a real-time data processing module, an anomaly identification and judgment module, and an intelligent control and update module; The benchmark feature construction module collects multimodal operation data from the historical operation stages of the target motor throughout its entire life cycle, analyzes the multimodal operation data to extract the features of each modal data and the correlation between different modal features; based on the above features and correlations, a multimodal benchmark feature set for the target motor is established. During the operation of the target motor, the real-time data processing module collects real-time multimodal operation data and analyzes the real-time multimodal operation data in accordance with the analysis method of multimodal operation data, thereby extracting the corresponding real-time features and real-time modal correlations. The anomaly identification and judgment module compares the real-time features with the features corresponding to the multimodal benchmark feature set, generates the dynamic change trajectory of the real-time features, identifies abnormal feature points in the dynamic change trajectory and marks them as drift features; for drift features, it calculates the deviation between the corresponding real-time modal correlation and the correlation in the multimodal benchmark feature set. If the deviation is within a preset range, it is judged as a single modal occasional fluctuation; if the deviation exceeds the preset range, it is judged as a multimodal collaborative anomaly. When the intelligent control and update module determines that the fluctuation is a single-mode intermittent fluctuation, it maintains the current drive parameters of the motor; when it determines that the multi-mode coordinated anomaly is, it outputs the corresponding drive parameter adjustment command to correct the abnormal state according to the mode type involved in the anomaly; it periodically collects multi-mode operation data when the target motor is running normally, updates the multi-mode reference feature set, and reconstructs a new multi-mode reference feature set if the motor is repaired or the operating conditions are changed.

[0026] The benchmark feature construction module includes a multimodal data acquisition unit, a feature extraction and correlation analysis unit, and a benchmark feature set generation unit; The multimodal data acquisition unit is used to collect multimodal operation data under typical operating conditions throughout the entire life cycle of the target motor, with consistent sampling frequency for each mode and acquisition duration covering the stable operating phase. The feature extraction and correlation analysis unit is used to preprocess the collected multimodal operation data, extract time-domain and frequency-domain features, and calculate the dynamic correlation coefficient and time delay correlation coefficient of feature pairs with physical coupling relationship. The benchmark feature set generation unit is used to determine the normal fluctuation range of each modal feature, the normal range of dynamic correlation coefficient, and the judgment range of time delay correlation value based on historical normal operation data, and establish a multimodal benchmark feature set.

[0027] The real-time data processing module includes a real-time multimodal data acquisition unit, a real-time feature extraction unit, and a real-time correlation calculation unit; The real-time multimodal data acquisition unit is used to continuously acquire real-time multimodal operation data of the same type as the benchmark feature construction module during the operation of the target motor, and the sampling frequency is consistent with the data of the benchmark feature construction module. The real-time feature extraction unit is used to process the real-time multimodal operation data and extract the corresponding real-time time domain features and real-time frequency domain features by referring to the preprocessing and feature extraction methods of the benchmark feature construction module. The real-time correlation calculation unit is used to calculate the real-time dynamic correlation coefficient and the real-time delay correlation coefficient by referring to the physical coupling feature pairs selected in the benchmark feature construction module, and obtain the real-time correlation consistent with the benchmark dimension.

[0028] The anomaly identification and judgment module includes a feature trajectory comparison unit, an anomaly point verification and marking unit, and an anomaly type judgment unit; The feature trajectory comparison unit is used to map and compare real-time features with corresponding features in the benchmark feature set, generate real-time feature dynamic change trajectories and divide them into segments corresponding to typical working conditions, and calculate the feature change slope within the sliding window to identify trend anomalies; the anomaly point verification and marking unit is used to calculate instantaneous mutation values ​​for trend anomalies to screen mutation anomalies, exclude isolated anomalies by verifying the trend consistency of multimodal features, and mark drift features; the anomaly type determination unit is used to calculate the deviation between the real-time correlation corresponding to the drift feature and the benchmark correlation, and combine the dynamically adjusted deviation threshold to determine single-modal occasional fluctuations or multimodal collaborative anomalies.

[0029] The intelligent control and update module includes a drive parameter adjustment unit, a closed-loop control unit, and a reference dynamic update unit; The drive parameter adjustment unit is used to maintain the current drive parameters when the fluctuation is determined to be a single-mode intermittent fluctuation, and to output the corresponding drive parameter adjustment command according to the mode type involved in the anomaly when the anomaly is determined to be a multi-mode coordinated anomaly. The closed-loop control unit is used to monitor the motor operating status after adjustment. If the anomaly is eliminated, the original parameters are restored. If the cumulative adjustment exceeds the limit, the protection shutdown is triggered. The benchmark dynamic update unit is used to periodically collect normal operation data to update the multi-mode benchmark feature set. When the motor is under maintenance or the operating condition changes suddenly, the benchmark feature construction process is re-executed and the historical benchmark feature set is archived.

[0030] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0031] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A motor drive intelligent control method based on multi-modal data fusion, characterized in that: The method comprises the following steps: S1. Collecting multi-modal operation data in the historical operation stage of the whole life cycle of the target motor, analyzing the multi-modal operation data to extract the characteristics of each modal data and the correlation between different modal characteristics, and establishing a multi-modal reference feature set of the target motor based on the above characteristics and correlation; S2. During the operation of the target motor, real-time multi-modal operation data is collected, and the real-time multi-modal operation data is analyzed by referring to the analysis method of the multi-modal operation data in S1, so as to extract corresponding real-time characteristics and real-time modal correlation; S3. Comparing the real-time characteristics with the corresponding characteristics of the multi-modal reference feature set, generating a dynamic change trajectory of the real-time characteristics, identifying abnormal feature points in the dynamic change trajectory and marking them as drift features; for the drift features, the deviation of the corresponding real-time modal correlation and the correlation in the multi-modal reference feature set is calculated, if the deviation is within the preset range, it is determined as single modal occasional fluctuation; if the deviation exceeds the preset range, it is determined as multi-modal coordinated anomaly; S4. When it is determined as single modal occasional fluctuation, the current driving parameters of the motor are maintained; when it is determined as multi-modal coordinated anomaly, according to the modal type involved in the anomaly, the corresponding driving parameter adjustment instruction is output to correct the abnormal state; periodically collecting multi-modal operation data of the target motor in normal operation, updating the multi-modal reference feature set, if the motor is repaired or the working condition is changed, a new multi-modal reference feature set is constructed by executing S1 again.

2. The intelligent control method of motor drive based on multi-modal data fusion according to claim 1, characterized in that: The S1 comprises the following contents: Collecting multi-modal operation data in the historical operation stage of the whole life cycle of the target motor, specifically collecting multi-modal operation data of the target motor under typical working conditions, the typical working conditions including no load, rated load, overload and start-stop process, the multi-modal operation data at least including current, vibration, temperature and speed data, the sampling frequency of each modal data is consistent, and the collection time covers the stable operation stage of each typical working condition; The collected multi-modal operation data is preprocessed, wherein the preprocessing includes denoising, data interpolation and time reference alignment; The preprocessed multi-modal operation data is extracted in time domain and frequency domain respectively; The multi-modal reference feature set is established, including the normal fluctuation interval of each modal characteristic determined by [μ±k×σ], wherein μ is the mean, σ is the standard deviation, and k is the interval coefficient; and the dynamic correlation coefficient normal interval [C1_min, C1_max] and the time delay correlation value determination range [C2_min, C2_max], and the above interval range is determined by the normal operation data in the historical operation stage of the whole life cycle of the target motor. The time domain feature is calculated according to a formula F t = f t (x t ), x t is time domain data in the multi-modal operation data, and f t is a time domain feature function; wherein The frequency domain feature is calculated according to the formula F p =f p (X p ) wherein X p is the frequency domain conversion result of the multi-modal operation data, and f p is a frequency domain feature function; Selecting a pair of features with physical coupling relationship from time domain features and frequency domain features, denoted as F1 and F2 respectively; calculating dynamic correlation coefficient C1 of the two features, and C1 = cov(F1, F2) / [σ(F1) × σ(F2)], wherein cov(F1, F2) is the covariance of F1 and F2, σ(F1) is the standard deviation of F1, and σ(F2) is the standard deviation of F2; calculating time delay correlation coefficient C2 of the two features, and C2 = ∑ t [F1’(t) × F2’(t + τ)], F1’(t) is the normalized F1 feature value at time t, F2’(t + τ) is the normalized F2 feature value at time t + τ, and τ is the response delay time determined through cross-correlation analysis; The S2 comprises the following contents: 3.The intelligent motor drive control method based on multi-modal data fusion of claim 2, wherein: During the operation of the target motor, real-time multi-modal operation data of the same type as in S1 is collected, and the same sampling frequency as in S1 is used for continuous collection; The real-time multi-modal operation data is processed by referring to the preprocessing method of S1; The S3 comprises the following contents: For the pre-processed real-time multi-modal operation data, real-time features are extracted by referring to the feature extraction method of S1, and the real-time time domain features are calculated according to the formula F t_real =f t (x t_real ), x t_real is real-time time domain data, f t is the time domain feature function selected in S1; the frequency domain Features are calculated according to the formula F p_real = f p (X p_real ) where X p_real is the result of the same frequency domain conversion of real-time multi-modal operating data in S1, and f p is the selected frequency domain feature function in S1. Referring to the selected physical coupling feature pairs (F1, F2) in S1, the real-time modal correlation relationship is calculated, and the calculation formula of the real-time dynamic correlation coefficient C1 is: C1_real=cov(F1_real, F2_real) / [σ(F1_real)×σ(F2_real)], cov(F1_real, F1_real) is the covariance of the real-time feature pair, σ(F1_real) is the standard deviation of the real-time feature F1, and σ(F2_real) is the standard deviation of the real-time feature F2; the calculation formula of the real-time time delay correlation coefficient C2_real is: C2_real=∑ t [F1’_real(t)×F2’_real(t+τ)], F1’_real(t) is the normalized feature value of the real-time feature F1 at time t, F2’_real(t+τ) is the normalized feature value of the real-time feature F2 at time t+τ, and τ is the response delay time determined in S1; thereby obtaining the real-time features and the real-time modal correlation relationship consistent with the features and the correlation relationship dimensions in S1.

4. The intelligent control method of motor drive based on multi-modal data fusion according to claim 3, characterized in that: ​ Map and compare the real-time features corresponding to the multi-modal benchmark feature set, generate a real-time feature dynamic change trajectory with time as the axis, and divide the real-time feature dynamic change trajectory into several segments corresponding to the typical working conditions in S1, so as to compare the real-time features with the benchmark features under the same working conditions; For each segment of the dynamic changing trajectory of each real-time feature, calculate the feature changing slope K_real in the sliding window, and K_real = [F real (t) - F real (t - T)] / T, where T represents the sliding window duration; if K_real exceeds 1.2 times the slope range [K_min, K_max] of the same working condition in S1, it is marked as a trend anomaly; For the segment of the real-time feature dynamic change trajectory marked as trend anomaly, the instantaneous mutation value ΔF is calculated for each real-time feature point corresponding to the segment, and ΔF = |F real (t)-F real (t-1)|, and if the instantaneous mutation value ΔF > α × σ, it is marked as a mutation anomaly point, where α represents a mutation coefficient and σ represents a standard deviation of the corresponding feature in S1; for the mutation anomaly point, it is checked whether other modal features at the corresponding time are consistent with the trend change of the mutation anomaly point, and if not, it is determined as an isolated anomaly, and if so, it is marked as a drift feature; For the drift features, calculate the deviation of the corresponding real-time modal correlation and the correlation in the multi-modal benchmark feature set, specifically including: Obtain the median of the dynamic correlation normal interval [C1_min, C1_max] under the same working condition in the multi-modal benchmark feature set as the benchmark reference value C1_ref, and C1_ref=(C1_min+C1_max) / 2, calculate the relative deviation ΔC1 of the real-time dynamic correlation and the benchmark reference value, and ΔC1=|C1_real-C1_ref| / C1_ref; Similarly, obtain the median of the time delay correlation value judgment range [C2_min, C2_max] under the same working condition in the multi-modal benchmark feature set as the benchmark reference value C2_ref, and C2_ref=(C2_min+C2_max) / 2, calculate the relative deviation ΔC2 of the real-time time delay correlation and the benchmark reference value, and ΔC2=|C2_real-C2_ref| / C2_ref; Pre-set dynamic correlation deviation threshold γ1 and time delay correlation deviation threshold γ2, wherein γ1 and γ2 are determined according to the correlation fluctuation range of the same working condition historical normal data in the multi-modal benchmark feature set, and the values of γ1 and γ2 are dynamically adjusted according to the motor running stage; If ΔC1≤γ1 and ΔC2≤γ2, it is determined that it is a single modal occasional fluctuation; If ΔC1>γ1 or ΔC2>γ2, it is determined that it is a multi-modal coordinated anomaly.

5. The intelligent control method of the motor drive based on multi-modal data fusion according to claim 4, characterized in that: S4 includes the following contents: When it is determined that it is a single modal occasional fluctuation, the current driving parameters of the target motor are maintained unchanged; When it is determined that it is a multi-modal coordinated anomaly, output the corresponding driving parameter adjustment instruction according to the type of physical coupling features involved in the anomaly; Calculate the excess deviation amount ΔE of the anomaly exceeding the threshold, and ΔE=max(ΔC1-γ1, ΔC2-γ2), and the excess deviation amount ΔE is the basis for driving parameter adjustment; Calculate the driving parameter adjustment amount ΔP, and ΔP=β×ΔE, wherein β is the adjustment coefficient corresponding to the driving parameter; According to the driving parameter adjustment amount ΔP, the driving parameter is adjusted to P_new=P_current+ΔP, wherein P_current is the current driving parameter value, the motor operating state is monitored in real time after adjustment, and the real-time features and real-time modal correlation are continuously extracted according to the analysis mode of S2; if ΔC1≤γ1 and ΔC2≤γ2 in the next 3 sampling periods, the driving parameter is restored to the current value P_current before adjustment; if it does not return to the normal range, ΔE' and ΔP' are recalculated based on the new real-time correlation deviation, ΔE'=max(ΔC1'-γ1,ΔC2'-γ2), ΔP'=β×ΔE', and the adjustment process is repeated; when the cumulative adjustment number reaches the preset upper limit and the abnormality is still not eliminated, or the real-time features after adjustment exceed 1.5 times of the normal fluctuation interval of the corresponding features in S1, a motor protection shutdown instruction is triggered, and the multi-modal data at the time of abnormality and the adjustment process log are recorded; Periodically collect multi-modal operating data of the target motor during continuous normal operation, and update the multi-modal reference feature set; the new reference feature mean μ_new=ω×μ_old+(1-ω)×μ_add, wherein ω is the historical reference weight, the new reference feature standard deviation σ_new={[ω×σ_old²+(1-ω)×σ_add²]}^(1 / 2), wherein μ_add and σ_add are the mean and standard deviation of the new normal data; the dynamic correlation coefficient normal interval and the time delay correlation value determination range are updated synchronously, and the new interval boundary is the weighted average of the historical interval boundary and the new data statistical boundary; when the target motor is repaired, a new multi-modal reference feature set is constructed by re-executing S1, and the original reference feature set is archived for life cycle comparison and analysis.

6. The motor drive intelligent control system based on multi-modal data fusion, applied to the motor drive intelligent control method based on multi-modal data fusion in any one of claims 1-5, characterized in that: The system comprises a reference feature construction module, a real-time data processing module, an abnormality identification and determination module, and an intelligent control and updating module; The reference feature construction module collects multi-modal operating data in the historical operating stage of the target motor during the whole life cycle, analyzes the multi-modal operating data, extracts features of each modal data, and the correlation between different modal features; based on the above features and correlation, a multi-modal reference feature set of the target motor is established; The real-time data processing module collects real-time multi-modal operating data during the operation of the target motor, analyzes the real-time multi-modal operating data by referring to the analysis mode of the multi-modal operating data, and extracts corresponding real-time features and real-time modal correlation; The abnormality identification and determination module compares the real-time features with the corresponding features in the multi-modal reference feature set, generates a dynamic change trajectory of the real-time features, identifies abnormal feature points in the dynamic change trajectory and marks them as drift features; for the drift features, the deviation of the corresponding real-time modal correlation and the correlation in the multi-modal reference feature set is calculated, if the deviation is within the preset range, it is determined as a single modal occasional fluctuation; if the deviation exceeds the preset range, it is determined as a multi-modal coordinated abnormality. The intelligent control and update module maintains the current driving parameters of the motor when determining that it is a single modal sporadic fluctuation, and outputs corresponding driving parameter adjustment instructions to correct the abnormal state according to the modal type involved in the abnormality when determining that it is a multi-modal coordinated abnormality; periodically collects multi-modal operation data of the target motor during normal operation, and updates the multi-modal reference feature set, and if the motor is repaired or the working condition is changed, a new multi-modal reference feature set is re-constructed.

7. The multi-modal data fusion based intelligent control system for motor drive according to claim 6, wherein: The reference feature construction module comprises a multi-modal data acquisition unit, a feature extraction and correlation analysis unit, and a reference feature set generation unit; The multi-modal data acquisition unit is configured to acquire multi-modal operation data of the target motor under typical working conditions in the whole life cycle, and the sampling frequencies of the modal data are consistent, and the collection time covers the stable stage of the working condition; The feature extraction and correlation analysis unit is configured to preprocess the collected multi-modal operation data, extract time domain features and frequency domain features, and calculate dynamic correlation coefficients and time delay correlation coefficients of feature pairs having a physical coupling relationship; and the reference feature set generation unit is configured to determine normal fluctuation intervals, dynamic correlation coefficient normal intervals, and time delay correlation value determination ranges of the modal features based on historical normal operation data, and establish a multi-modal reference feature set.

8. The multi-modal data fusion based intelligent control system for motor drive according to claim 6, wherein: The real-time data processing module comprises a real-time multi-modal data acquisition unit, a real-time feature extraction unit, and a real-time correlation relationship calculation unit; The real-time multi-modal data acquisition unit is configured to continuously acquire real-time multi-modal operation data consistent with the type of the reference feature construction module during the operation of the target motor, and the sampling frequency is consistent with that of the reference feature construction module; the real-time feature extraction unit is configured to process the real-time multi-modal operation data and extract corresponding real-time time domain features and real-time frequency domain features by referring to the preprocessing and feature extraction manner of the reference feature construction module; The real-time correlation relationship calculation unit is configured to calculate real-time dynamic correlation coefficients and real-time time delay correlation coefficients by referring to the selected physical coupling feature pairs in the reference feature construction module, and obtain real-time correlation relationships consistent with the reference dimension.

9. The multi-modal data fusion based intelligent control system for motor drive according to claim 6, wherein: The abnormality recognition and determination module comprises a feature trajectory comparison unit, an abnormal point verification and marking unit, and an abnormal type determination unit; The feature trajectory comparison unit is configured to map and compare the real-time features with the corresponding features in the reference feature set, generate real-time feature dynamic change trajectories, divide the trajectories into segments corresponding to typical working conditions, and calculate feature change slopes in a sliding window to identify trend abnormalities; the abnormal point verification and marking unit is configured to calculate instantaneous mutation values of the trend abnormal points to screen out mutation abnormal points, exclude isolated abnormalities by verifying the consistency of multi-modal feature trends, and mark the drift features; and the abnormal type determination unit is configured to calculate the deviation of the real-time correlation relationship corresponding to the drift features from the reference correlation relationship, and determine single modal sporadic fluctuation or multi-modal coordinated abnormality combined with a dynamically adjusted deviation threshold.

10. The multi-modal data fusion based intelligent control system for motor drive according to claim 6, wherein: The intelligent control and update module comprises a driving parameter adjustment unit, a closed-loop control unit, and a reference dynamic update unit; The driving parameter adjustment unit is configured to maintain the current driving parameter when determining that the single mode is sporadic fluctuation, and output a corresponding driving parameter adjustment instruction according to the mode type involved in the abnormality when determining that the multi-mode is collaborative abnormality; the closed-loop control unit is configured to monitor the running state of the adjusted motor, and restore the original parameter if the abnormality is eliminated, or trigger a protection shutdown if the cumulative adjustment exceeds the limit; The reference dynamic updating unit is configured to periodically collect normal operation data to update the multi-mode reference feature set, and re-execute the reference feature construction process when the motor is repaired or the working condition is suddenly changed, and archive the historical reference feature set.

Images