An electric consumption data supervision method and system for high-power-consumption industrial equipment

Abstract

This invention discloses a method and system for monitoring power consumption data of high-power-consuming industrial equipment, belonging to the field of power consumption analysis technology. Addressing the problem of artificially high power consumption induced by mechanical transient coupling, this method simultaneously collects current, power, vibration, pressure, and acoustic emission data. It uses a computer-generated transient activation index to locate transient intervals. Empirical mode decomposition is performed on the power data within the transient intervals, and the artificial high-frequency intrinsic mode functions of power consumption are screened by combining the vibration envelope spectrum attenuation characteristics. The transient characteristic parameters of multiple physical quantities are fused with the eigenvalues ​​of the time-delay cross-correlation matrix of vibration and current, and the mutual information values ​​of acoustic emission and pressure, establishing a support vector regression mapping model to output predicted values ​​of the artificially high component. The power consumption data is corrected by subtracting the predicted value of the artificially high component from the original power, and its exponential moving average is calculated to update the power consumption monitoring baseline. An alarm is triggered if the limit is exceeded. This invention effectively filters out artificially high power consumption induced by mechanical coupling, achieving accurate purification and dynamic monitoring of real power consumption under varying operating conditions.

Description

Technical Field

[0001] This invention relates to the field of power consumption analysis technology, specifically to a method and system for monitoring power consumption data of high power-consuming industrial equipment. Background Technology

[0002] Under transient conditions of frequent start-stop or drastic load changes, high-power-consuming industrial equipment inevitably generates strong dynamic torsion and hydrodynamic pulsations in its internal mechanical transmission system. These drastic dynamic responses on the mechanical side not only alter the actual work done at the load end, but also transmit them in reverse to the electromagnetic side of the drive motor through complex electromechanical multi-field coupling paths. This inevitably results in a large amount of unsteady-state distortion induced by mechanical transient excitation being superimposed on the monitoring signal, which should have smoothly represented the actual work consumption.

[0003] Currently, energy consumption monitoring for industrial equipment mainly relies on direct measurement and analysis of single electrical parameters, or conventional time-series filtering and smoothing methods. However, monitoring a single electrical parameter only focuses on the surface of electrical energy and cannot penetrate the electromechanical coupling barrier to trace the source of distorted signals. Conventional filtering methods can only remove steady-state background noise or regular harmonics. When faced with mechanically coupled distortion components with broadband abrupt changes and strong nonlinear characteristics, the lack of cross-verification and decoupling mechanisms using multi-dimensional physical parameters means that existing methods are fundamentally unable to separate the spurious energy consumption components induced by mechanical energy oscillations. This limitation results in severely inflated and distorted energy consumption statistics under transient operating conditions, and distortion spikes easily break through static thresholds, frequently triggering false alarms and severely weakening the accuracy and reliability of the energy consumption monitoring system. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for monitoring power consumption data of high power-consuming industrial equipment, so as to solve the problems mentioned in the background art.

[0005] To address the aforementioned technical problems, this invention provides a method for monitoring power consumption data of high-power-consuming industrial equipment, comprising:

[0006] S100 synchronously acquires current, power, vibration, pressure, and acoustic emission data from industrial equipment. It extracts the first-order difference extrema of the current and acoustic emission data, and uses the time difference between these extrema to calculate the transient activation index of the electrical coupling to locate the transient interval. Specifically, this includes:

[0007] S101. Synchronously acquire the current signal sequence of industrial equipment at a set sampling frequency. Power signal sequence Vibration signal sequence Pressure signal sequence and acoustic emission signal sequence .

[0008] S102. Calculate the acoustic emission signal sequences respectively. First-order difference sequence and current signal sequence First-order difference sequence .

[0009] S103, Extract the first-order difference sequence The first time point set corresponding to the maximum points whose absolute values ​​are greater than the first preset threshold. and first-order difference sequences The second time point set corresponding to the maximum points whose absolute values ​​are greater than the second preset threshold. Specifically:

[0010] Step 1: Process the original acquired acoustic emission signal sequence and current signal sequence Calculate the first-order difference separately. The first-order difference is the magnitude change between two adjacent sampling points. First-order difference sequence. and It reflects the instantaneous rate of change or slope of the signal in the time domain.

[0011] Step 2: Find local maxima in the difference sequence. The maximum point of the difference sequence means that the amplitude of the original signal has undergone the most drastic instantaneous jump at that moment, which usually corresponds to a sudden collision of mechanical parts or an instantaneous loading / unloading of motor load.

[0012] Step 3: To eliminate minor fluctuations or background white noise during normal equipment operation, set a first preset threshold and a second preset threshold. Only retain extreme points whose absolute values ​​are greater than the corresponding thresholds. The timestamps corresponding to these surviving extreme points on the original timeline are aggregated to form the first time point set. Second time point set .

[0013] In the complex operating environment of high-power-consuming equipment, sudden current changes may be caused by grid voltage fluctuations, and sudden acoustic emission changes may be caused by accidental external impacts. These do not belong to the true electromechanical coupling transients.

[0014] Therefore, the mathematical configuration of energy product × Gaussian time window is required:

[0015] On the one hand, through energy product It is mandatory that a sudden change in sufficient energy must occur simultaneously at both the mechanical and electrical ends.

[0016] On the other hand, the abrupt time difference between these two heterogeneous signals is forced through the Gaussian exponential decay term. It must conform to the objective physical delay law of physical stress wave propagation from the mechanical fault point to the stator side.

[0017] Only when both fluctuate violently at the same time and the time difference closely matches the laws of physical propagation will a high activation index be output, thereby achieving a very low misjudgment rate in locating the starting point of the transient interval.

[0018] S104, For the first time point set Second time point set The time interval between adjacent extreme points is checked. If the time interval between two adjacent extreme points is less than the minimum interval threshold, the extreme point with smaller amplitude is removed and the extreme point with larger amplitude is retained to complete the extreme point deburring operation.

[0019] S105, Traversing the Second Time Point Set Each time point in the first time point set The computer electrical coupling transient activation index is used to retrieve the preceding time point with the smallest time difference. The calculation formula is:

[0020] ;

[0021] In the formula, For the first time point set The absolute amplitude of the first-order differential extreme point of the pre-sequence acoustic emission represents the magnitude of the transient elastic energy released on the mechanical side, and is a direct characterization of the occurrence of mechanical failure.

[0022] For the second time point set The absolute amplitude of the first-order differential extreme point of the current represents the intensity of the transient current fluctuation caused by sudden changes in mechanical load torque or rotor motion state on the electrical side.

[0023] The time difference between the two extreme points is the measured time interval between the acoustic emission extreme point and the current extreme point, reflecting the time lag from the occurrence of mechanical disturbance to the generation of an observable response on the electrical side.

[0024] The preset electromechanical physical propagation delay constant represents the average objective physical propagation time from the mechanical vibration source through the rotor shaft system and air gap magnetic field coupling to the stator winding.

[0025] As the expected value of a Gaussian distribution, when equal When the time difference is 1, the Gaussian term takes the value 1, indicating that the time difference perfectly matches the laws of physics.

[0026] The preset time tolerance standard deviation represents the degree of propagation time dispersion caused by differences in equipment installation location and material non-uniformity. It is used to control the width of the Gaussian bell curve and determine the tolerance of time matching. Deviation The further away the distance, the more drastic the decay of the exponential term, thus eliminating coincidental matches due to time misalignment.

[0027] The electromechanical coupling transient activation index is used to accurately identify real-world operational changes coupled from mechanical transient impacts to electrical transients, and to strongly suppress independent noise interference from single physical quantities. A higher value indicates a higher probability of a real electromechanical coupling transient event occurring at the current moment. When the index exceeds a preset threshold, this moment is considered the start of the transient interval.

[0028] By constraining the time difference with a Gaussian attenuation factor and combining the dual energy product of acoustic emission and current, the coupling strength of the physical field is quantified to eliminate misjudgments caused by independent noise of a single physical quantity. Thus, a high activation index is output only when acoustic emission and current fluctuate violently at the same time and the time difference conforms to the physical propagation law.

[0029] S106. When the electromechanical coupling transient activation index is greater than the preset activation threshold, the second time point set... The time point marked in the time frame is the transient start point, and the interval extending backward from the transient start point for a set time length is defined as the transient interval.

[0030] S107. Set the first time point after processing in chronological order. The preceding time point was retrieved and the computer-coupled transient activation index was obtained.

[0031] S200. Perform empirical mode decomposition on the power data within the transient interval to obtain multiple intrinsic mode functions (IMFs). Calculate the kurtosis value of each IMF. Combined with the energy decay characteristics of the envelope spectrum of the vibration data, select the power consumption virtual high-frequency IMFs from the IMFs. Specifically, this includes:

[0032] S201. Extract the power signal sequence within the transient interval. Performing empirical mode decomposition yields N intrinsic mode functions. to .

[0033] S202, Calculate each intrinsic mode function kurtosis value Filter out kurtosis values Greater than the kurtosis threshold The intrinsic mode functions constitute a high-frequency mode set.

[0034] S203, Regarding the vibration signal sequence within the transient interval Bandpass filtering and Hilbert transform are performed to obtain the envelope spectrum. The envelope curve, which shows the energy of the envelope spectrum changing over time, is calculated, and the attenuation coefficient of the fitted envelope curve is determined. .

[0035] S204. Calculate the intrinsic mode function in the high-frequency mode set. center frequency If the center frequency Falling on the attenuation coefficient Within the corresponding mechanical resonance frequency band, the inherent mode function is... It is labeled as the imaginary high-frequency intrinsic mode function of power consumption.

[0036] The mechanical resonance frequency band refers to the inherent frequency range in which the mechanical rotating parts or supporting structures of high-power-consuming industrial equipment undergo a dynamic amplification effect when subjected to excitation force.

[0037] If a certain power The center frequency falls exactly within this frequency band, indicating that the power consumption fluctuation component is coupled at the same frequency with the resonant response on the mechanical side, confirming that the power consumption spike is a virtual high-frequency intrinsic mode function caused by the mechanical resonant energy reinjected into the power grid.

[0038] S300. Construct the time-delay cross-correlation matrix for vibration and current data, calculate the mutual information value between acoustic emission and pressure data, and nonlinearly exponentially fuse the eigenvalues ​​and mutual information values ​​of the time-delay cross-correlation matrix into multi-physical quantity transient characteristic parameters. Use support vector regression to establish a mapping model between the multi-physical quantity transient characteristic parameters and the virtual high-frequency intrinsic mode function of power consumption, outputting the predicted value of the virtual high-frequency component. Specifically, this includes:

[0039] S301, The vibration signal sequence within the transient interval With current signal sequence The system is divided into multiple segments using a sliding window. A cross-correlation function is calculated for each segment, and the time delay corresponding to the peak value of the cross-correlation function is identified. Construct latency values ​​between different sub-segments The time delay covariance matrix is ​​used as the time delay cross-correlation matrix, and the maximum eigenvalue of the time delay cross-correlation matrix is ​​calculated. .

[0040] S302, the acoustic emission signal sequence within the transient interval With pressure signal sequence The signal is divided into equal-length segments. The kernel density estimation method is used to calculate the joint probability density and marginal probability density of the acoustic emission signal sequence and the pressure signal sequence for each segment. The mutual information value is then calculated based on the joint probability density and marginal probability density. .

[0041] Joint probability density and marginal probability density are used to quantify the degree of nonlinear statistical dependence between acoustic emission signals and pressure signals. Wherein:

[0042] The joint probability density describes the probability density of two random variables occurring simultaneously under a specific combination. It reflects the "co-occurrence and interweaving" pattern of acoustic emission and pressure signals in the amplitude space.

[0043] If a specific pulsation value occurs in the pressure, a specific impact value often occurs in the acoustic emission. At the coordinate points of this amplitude combination, the value of the joint probability density will be significantly higher.

[0044] Marginal probability density is the probability density of a single variable obtained by integrating the joint probability density over one of the variables.

[0045] This reflects the distribution law of the amplitude of a single physical quantity without considering the value of another variable. In other words, it is the statistical distribution that acoustic emission and pressure should have when they are completely independent and do not interfere with each other.

[0046] The contributions of vibration / current characteristics and acoustic emission / pressure characteristics vary dynamically under different transient fault modes.

[0047] Traditional linear weighted fusion cannot characterize this game and alternation, so a dual composite structure of dynamic exponential modulation and logarithmic smoothing compensation is required:

[0048] The first term uses the mutual information ratio as an exponent. When the acoustic emission is strongly correlated with pressure, the exponent increases, nonlinearly amplifying the weight of the time delay characteristics.

[0049] The second step introduces a logarithmic smoothing mechanism to prevent... When the value is too large, the fusion value explodes. At the same time, it ensures that when one type of feature is extremely weak, another type of feature can still be effectively preserved in the fusion space, thereby deeply exploring the hidden coupling distortion relationship across physical fields.

[0050] S303, the largest eigenvalue With mutual information value Substitute into the multiphysics coupling distortion feature fusion formula to calculate the fused eigenvalue. , will fuse feature values As a transient characteristic parameter of multiple physical quantities. The formula is:

[0051] ;

[0052] In the formula, This is the preset historical steady-state benchmark mutual information value.

[0053] It is used to fuse feature values; as the sole input vector of the support vector regression model, it is a dimensionless scalar that comprehensively characterizes the intensity and mode characteristics of multi-physics coupling distortion within the current transient interval.

[0054] It represents the maximum eigenvalue of the time-delay cross-correlation matrix; it characterizes the global principal component energy of vibration and current signals in the time-delay dimension, focusing on reflecting the linear topological correlation strength between mechanical rotor dynamics and electromagnetic dynamics. The larger the value, the stronger the topological dependence of the electromechanical coupling.

[0055] It represents the mutual information value between the acoustic emission signal sequence and the pressure signal sequence; it characterizes the degree of nonlinear statistical dependence between the two random variables of acoustics and fluid dynamics, focusing on reflecting the intrinsic correlation between minute mechanical collisions and airflow / hydraulic pulsations. The larger the value, the more intense the fluid-structure interaction or acoustic-pressure correlation.

[0056] It is a preset historical steady-state baseline mutual information value; representing the background level of acoustic emission and pressure interaction information of the equipment under normal operating conditions, and is used as a normalization reference benchmark.

[0057] This reflects the surge multiple of the current transient acoustic-pressure correlation relative to the steady-state baseline. This ratio, as an exponent, directly determines the amplification factor of the time delay characteristics, enabling the fusion process to dynamically adapt to fault modes.

[0058] The multi-physics coupling distortion feature fusion formula is used to calculate the fusion feature value, realizing adaptive nonlinear feature fusion when different physical fields alternate in dominance during transient processes.

[0059] By using the mutual information ratio as an exponent to nonlinearly modulate the time delay characteristics and introducing a logarithmic smoothing mechanism into the mutual information term, we can achieve adaptive representation and deep coupling relationship mining of the alternating dominance of different physical fields in the transient process.

[0060] S304, Fusion feature values ​​extracted from historical transient intervals as input vector The output vector is the amplitude sequence of the power consumption virtual high-frequency intrinsic mode function marked with the corresponding historical transient interval. , construct a training sample set.

[0061] S305. Select the Gaussian kernel function as the kernel function for support vector regression and set the penalty parameter. and insensitive loss parameters Solve the dual optimization problem of support vector regression to obtain the Lagrange multipliers and bias terms, and complete the training of the mapping model.

[0062] S306. Merge the characteristic values ​​of the current transient interval. The input is fed into the trained mapping model, and the regression prediction sequence output by the mapping model is used as the predicted value of the virtual height component corresponding to the current transient interval.

[0063] S400: Subtract the predicted value of the artificially high component from the original power data to obtain the corrected power consumption data. Calculate the exponential moving average of the corrected power consumption data within the steady-state range to update the power consumption monitoring baseline. When the real-time corrected power consumption data deviates from the power consumption monitoring baseline by more than a set threshold, a monitoring alarm signal is output. Specifically, this includes:

[0064] S401. Extract the power signal sequence corresponding to the transient interval. , power signal sequence Subtracting the amplitude of the predicted value of the artificial height component at the same timestamp from each sampling point, the resulting difference sequence is used as the corrected power consumption data. .

[0065] S402, Real-time calculation of vibration signal sequence Pressure signal sequence With current signal sequence When the variances of the three variables are all continuously lower than their respective preset variance thresholds for a set time period, the current interval is determined to be a steady-state interval.

[0066] In industrial settings, even after eliminating transient overstatement, steady-state power consumption data can still fluctuate in a jagged manner due to minor electromagnetic interference and measurement noise.

[0067] Using the original corrected data directly as the baseline will cause the regulatory baseline to fluctuate, which can easily trigger false alarms.

[0068] S403. Extract corrected power consumption data within the steady-state range. Substitute the values ​​into the exponential moving average formula to calculate the exponential moving average:

[0069] ;

[0070] In the formula, For the current moment The exponential moving average is the latest power consumption regulatory baseline value output after smoothing at the current moment.

[0071] The preset smoothing coefficient has a range of values. Used to determine the sensitivity of the smoothing algorithm to the latest data:

[0072] The larger the value, the higher the proportion of the current observation, and the faster the baseline changes with new data, but the weaker the noise resistance.

[0073] The smaller the value, the higher the proportion of the historical baseline, and the smoother and more stable the baseline, but the less sensitive it is to actual power consumption fluctuations. It needs to be set according to the power consumption fluctuation characteristics of the equipment.

[0074] For the current moment The corrected power consumption data, which is the real instantaneous power consumption sample value after removing the artificially high components, is the latest input driving the baseline update.

[0075] For the previous moment The exponential moving average, i.e., the historical baseline value generated at the previous moment, implicitly condenses... Exponentially weighted information of all historical steady-state data up to and including the current time.

[0076] The calculated exponential moving average is updated as the electricity consumption regulatory baseline.

[0077] The exponential moving average is used to filter out high-frequency random noise interference within the steady-state range, extract the smooth core trend that reflects the true steady-state power consumption level of the equipment, and use it as a dynamically updated power consumption monitoring baseline.

[0078] Through a recursive accumulation mechanism, the current baseline value contains both the latest real power consumption drift information and inherits the long-term trend framework of history, thus providing an extremely stable and reliable reference benchmark for subsequent calculation of the relative deviation rate.

[0079] Since this formula is an iterative loop formula that relies on its own historical output as the input for the next step, there is no historical baseline value when the system is cold-started or when the steady-state region is detected for the first time. The formula cannot be calculated directly. In this case, the initial value initialization method must be used, and the specific rules are as follows:

[0080] When the system first enters the steady-state region and the first corrected power consumption data point is collected. At that time, the initial baseline value is forced to be equal to the first observation value, that is, set... .

[0081] This setting makes the first iteration become... .

[0082] Although this hard assignment introduces some truncation error in the initial stage, due to the exponential decay memory characteristic of the EMA formula, the initial outlier weights will decrease with the increase of the number of iterations. Its speed rapidly approaches zero exponentially.

[0083] After a short warm-up period, the impact of the initial artificially assigned values ​​will be completely diluted by the subsequent massive amounts of real steady-state data, and the baseline will naturally converge to the true steady-state smooth center, without substantially interfering with subsequent long-term regulatory alerts.

[0084] S404. Real-time acquisition of current corrected power consumption data within the non-transient interval. Simultaneously read the current power consumption monitoring baseline. .

[0085] S405, Calculate the current corrected power consumption data With electricity consumption regulatory baseline relative deviation rate The relative deviation rate Deviation rate threshold Comparison, if the relative deviation rate Greater than the deviation rate threshold If the duration exceeds the set hysteresis time, a regulatory alarm signal will be generated; among which, .

[0086] This invention also provides a power consumption data monitoring system for high power-consuming industrial equipment, comprising:

[0087] The multi-physical quantity synchronous acquisition module can synchronously acquire current signal sequences, power signal sequences, vibration signal sequences, pressure signal sequences, and acoustic emission signal sequences during the operation of industrial equipment by setting a sampling frequency.

[0088] The transient interval location module calculates the first-order difference extremum points of the signal sequence and quantifies the energy product of acoustic emission and current and the time difference attenuation characteristics according to the electromechanical coupling transient activation exponent formula to identify the transient start point and extract the transient interval.

[0089] The virtual high component extraction module performs empirical mode decomposition on the power data in the transient interval to obtain the intrinsic mode functions, and combines the vibration envelope spectrum decay characteristics to screen out the virtual high frequency intrinsic mode functions of power consumption.

[0090] The feature fusion and mapping module constructs the time-delay cross-correlation matrix of vibration and current and the mutual information value of acoustic emission and pressure. It calculates the fused feature value through the multi-physics field coupling distortion feature fusion formula and outputs the predicted value of the virtual height component based on the support vector regression model.

[0091] The dynamic power consumption monitoring module subtracts the predicted value of the artificially high power from the original power to obtain the corrected power consumption data, calculates the exponential moving average of the corrected power consumption data in the steady-state range to update the monitoring baseline, and outputs a monitoring alarm signal when the real-time corrected power consumption data deviates from the monitoring baseline and exceeds the limit.

[0092] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0093] By utilizing the dual energy product of acoustic emission and electric current and the time difference of physical propagation, the coupling strength is quantified, effectively eliminating misjudgments caused by independent noise of a single physical quantity. The transient interval is located only when multiple physical quantities fluctuate violently at the same time and the time difference conforms to the law of physical propagation, which greatly improves the accuracy of identifying the transient start point of electromechanical coupling.

[0094] By performing empirical mode decomposition on transient power data to achieve frequency band decoupling, and combining the vibration envelope spectrum attenuation characteristics, the virtual high-frequency intrinsic mode function of power consumption induced by mechanical resonance is accurately located, fundamentally isolating the interference of mechanical physical field on power consumption benchmark.

[0095] We construct the time-delay cross-correlation matrix of vibration and current, as well as the mutual information value of acoustic emission and pressure. Through nonlinear exponential fusion, we achieve an adaptive representation of the alternating dominance of different physical fields. Based on the support vector regression model, we accurately predict and reconstruct the transient virtual height component, and deeply explore the hidden coupling distortion relationship across physical fields.

[0096] By subtracting the predicted value of the artificially high component from the original power data, the power consumption is dynamically purified. The power consumption regulatory baseline is dynamically updated using the exponential moving average, which completely filters out the artificially high power consumption under transient operating conditions, keeps the energy consumption benchmark highly stable under complex changing operating conditions, and significantly reduces the regulatory false alarm rate during the start-up and shutdown phase of changing loads. Attached Figure Description

[0097] 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:

[0098] Figure 1 This is a flowchart illustrating a method for monitoring power consumption data of high-power-consuming industrial equipment according to the present invention. Detailed Implementation

[0099] 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.

[0100] Example 1: Please refer to Figure 1 This invention provides a method for monitoring power consumption data of high-power-consuming industrial equipment, comprising:

[0101] S100 synchronously acquires current, power, vibration, pressure, and acoustic emission data from industrial equipment. It extracts the first-order difference extrema of the current and acoustic emission data, and uses the time difference between these extrema to calculate the transient activation index of the electrical coupling to locate the transient interval. Specifically, this includes:

[0102] S101. Synchronously acquire the current signal sequence of industrial equipment at a set sampling frequency. Power signal sequence Vibration signal sequence Pressure signal sequence and acoustic emission signal sequence .

[0103] S102. Calculate the acoustic emission signal sequences respectively. First-order difference sequence and current signal sequence First-order difference sequence .

[0104] S103, Extract the first-order difference sequence The first time point set corresponding to the maximum points whose absolute values ​​are greater than the first preset threshold. and first-order difference sequences The second time point set corresponding to the maximum points whose absolute values ​​are greater than the second preset threshold. .

[0105] In the specific implementation process, it is as follows:

[0106] Step 1 (First-order difference calculation): Calculate the original acquired acoustic emission signal sequence. and current signal sequence Calculate the first-order difference separately. The first-order difference is the magnitude change between two adjacent sampling points. First-order difference sequence. and It reflects the instantaneous rate of change or slope of the signal in the time domain.

[0107] The second step (finding extreme points): Find local maxima in the difference sequence. The maximum point of the difference sequence means that the amplitude of the original signal has undergone the most drastic instantaneous jump at that moment (i.e., the point with the largest slope), which usually corresponds to a sudden collision of mechanical parts (sudden change in acoustic emission) or a sudden loading / unloading of motor load (sudden change in current).

[0108] Step 3 (Threshold Filtering and Set Construction): To eliminate minor fluctuations or background white noise during normal device operation, a first preset threshold (for the absolute value of acoustic emission difference) and a second preset threshold (for the absolute value of current difference) are set. Only extreme points with absolute values ​​greater than the corresponding thresholds are retained. The timestamps corresponding to these surviving extreme points on the original timeline are aggregated to form the first time point set. (Set of moments of dramatic changes in acoustic emission) and second time point set (Set of moments of drastic change in current).

[0109] In the complex operating environment of high-power-consuming equipment, sudden current changes may be caused by grid voltage fluctuations, and sudden acoustic emission changes may be caused by accidental external impacts. These do not belong to the true electromechanical coupling transients.

[0110] In practical implementation, the mathematical configuration of energy product × Gaussian time window needs to be adopted:

[0111] On the one hand, through energy product It is mandatory that a sudden change in sufficient energy must occur simultaneously at both the mechanical end (acoustic emission) and the electrical end (current).

[0112] On the other hand, the abrupt time difference between these two heterogeneous signals is forced through the Gaussian exponential decay term. It must conform to the objective physical delay law of physical stress wave propagation from the mechanical fault point to the stator side.

[0113] Only when both fluctuate violently at the same time and the time difference closely matches the laws of physical propagation will a high activation index be output, thereby achieving a very low misjudgment rate in locating the starting point of the transient interval.

[0114] S104, For the first time point set Second time point set The time interval between adjacent extreme points is checked. If the time interval between two adjacent extreme points is less than the minimum interval threshold, the extreme point with smaller amplitude is removed and the extreme point with larger amplitude is retained to complete the extreme point deburring operation.

[0115] S105, Traversing the Second Time Point Set Each time point in the first time point set The computer electrical coupling transient activation index is used to retrieve the preceding time point with the smallest time difference. The calculation formula is:

[0116] ;

[0117] In the formula, For the first time point set The absolute amplitude of the first-order differential extremum point of the pre-sequence acoustic emission represents the magnitude of the transient elastic energy released by the mechanical side (such as early wear of bearings or rubbing), and is a direct characterization of the occurrence of mechanical failure.

[0118] For the second time point set The absolute amplitude of the first-order differential extreme point of the current represents the intensity of the transient current fluctuation caused by sudden changes in mechanical load torque or rotor motion state on the electrical side.

[0119] The time difference between the two extreme points is the measured time interval between the acoustic emission extreme point and the current extreme point, reflecting the time lag from the occurrence of mechanical disturbance to the generation of an observable response on the electrical side.

[0120] The preset electromechanical physical propagation delay constant represents the average objective physical propagation time from the mechanical vibration source through the rotor shaft system and air gap magnetic field coupling to the stator winding.

[0121] As the expected value (center point) of the Gaussian distribution, when equal When the time difference is 1, the Gaussian term takes the value 1, indicating that the time difference perfectly matches the laws of physics.

[0122] The preset time tolerance standard deviation represents the degree of propagation time dispersion caused by differences in equipment installation location and material non-uniformity. It is used to control the width of the Gaussian bell curve and determine the tolerance of time matching. Deviation The further away the distance, the more drastic the decay of the exponential term, thus eliminating coincidental matches due to time misalignment.

[0123] In practical implementation, the electromechanical coupling transient activation index is used to accurately identify real-world operational changes coupled from mechanical transient impacts to electrical transients, and to strongly suppress independent noise interference from single physical quantities. A higher value indicates a higher probability of a real electromechanical coupling transient event occurring at the current moment. When the index exceeds a preset threshold, this moment is considered the start of the transient interval.

[0124] By constraining the time difference with a Gaussian attenuation factor and combining the dual energy product of acoustic emission and current, the coupling strength of the physical field is quantified to eliminate misjudgments caused by independent noise of a single physical quantity. Thus, a high activation index is output only when acoustic emission and current fluctuate violently at the same time and the time difference conforms to the physical propagation law.

[0125] S106. When the electromechanical coupling transient activation index is greater than the preset activation threshold, the second time point set... The time point marked in the time frame is the transient start point, and the interval extending backward from the transient start point for a set time length is defined as the transient interval.

[0126] S107. Set the first time point after processing in chronological order. The preceding time point was retrieved and the computer-coupled transient activation index was obtained.

[0127] S200. Perform empirical mode decomposition on the power data within the transient interval to obtain multiple intrinsic mode functions (IMFs). Calculate the kurtosis value of each IMF. Combined with the energy decay characteristics of the envelope spectrum of the vibration data, select the power consumption virtual high-frequency IMFs from the IMFs. Specifically, this includes:

[0128] S201. Extract the power signal sequence within the transient interval. Performing empirical mode decomposition yields N intrinsic mode functions. to .

[0129] S202, Calculate each intrinsic mode function kurtosis value Filter out kurtosis values Greater than the kurtosis threshold The intrinsic mode functions constitute a high-frequency mode set.

[0130] S203, Regarding the vibration signal sequence within the transient interval Bandpass filtering and Hilbert transform are performed to obtain the envelope spectrum. The envelope curve, which shows the energy of the envelope spectrum changing over time, is calculated, and the attenuation coefficient of the fitted envelope curve is determined. .

[0131] S204. Calculate the intrinsic mode function in the high-frequency mode set. center frequency If the center frequency Falling on the attenuation coefficient Within the corresponding mechanical resonance frequency band, the inherent mode function is... It is labeled as the imaginary high-frequency intrinsic mode function of power consumption.

[0132] The mechanical resonance frequency band refers to the inherent frequency range in which the mechanical rotating parts or supporting structures of high-power-consuming industrial equipment undergo a dynamic amplification effect when subjected to excitation force.

[0133] In the specific implementation process, if a certain power The center frequency falls exactly within this frequency band, indicating that the power consumption fluctuation component is coupled at the same frequency with the resonant response on the mechanical side, confirming that the power consumption spike is a virtual high-frequency intrinsic mode function caused by the mechanical resonant energy reinjected into the power grid.

[0134] S300. Construct the time-delay cross-correlation matrix for vibration and current data, calculate the mutual information value between acoustic emission and pressure data, and nonlinearly exponentially fuse the eigenvalues ​​and mutual information values ​​of the time-delay cross-correlation matrix into multi-physical quantity transient characteristic parameters. Use support vector regression to establish a mapping model between the multi-physical quantity transient characteristic parameters and the virtual high-frequency intrinsic mode function of power consumption, outputting the predicted value of the virtual high-frequency component. Specifically, this includes:

[0135] S301, The vibration signal sequence within the transient interval With current signal sequence The system is divided into multiple segments using a sliding window. A cross-correlation function is calculated for each segment, and the time delay corresponding to the peak value of the cross-correlation function is identified. Construct latency values ​​between different sub-segments The time delay covariance matrix is ​​used as the time delay cross-correlation matrix, and the maximum eigenvalue of the time delay cross-correlation matrix is ​​calculated. .

[0136] S302, the acoustic emission signal sequence within the transient interval With pressure signal sequence The signal is divided into equal-length segments. The kernel density estimation method is used to calculate the joint probability density and marginal probability density of the acoustic emission signal sequence and the pressure signal sequence for each segment. The mutual information value is then calculated based on the joint probability density and marginal probability density. .

[0137] In practical implementation, the joint probability density and marginal probability density are used to quantify the degree of nonlinear statistical dependence between the acoustic emission signal and the pressure signal. Specifically:

[0138] The joint probability density describes two random variables (acoustic emission amplitude). and pressure amplitude The probability density of simultaneous occurrence of acoustic emission and pressure signals under a specific combination. It reflects the "co-occurrence and interweaving" pattern of acoustic emission and pressure signals in amplitude space.

[0139] If a specific pulsation value occurs in the pressure, a specific impact value often occurs in the acoustic emission. At the coordinate points of this amplitude combination, the value of the joint probability density will be significantly higher (showing a dense distribution).

[0140] Marginal probability density is the probability density of a single variable obtained by integrating (or summing) the joint probability density over one of the variables.

[0141] This reflects the distribution law of the amplitude of a single physical quantity without considering the value of another variable. In other words, it is the statistical distribution that acoustic emission and pressure should have when they are completely independent and do not interfere with each other.

[0142] Under different transient fault modes (e.g., rotor dynamic imbalance tends towards electromechanical coupling, while aerodynamic surge tends towards fluid-structure coupling), the vibration / current characteristics (by...) Characterization) and acoustic emission / pressure characteristics (by The contribution of the representation is dynamic.

[0143] Traditional linear weighted fusion cannot characterize this game and alternation, so a dual composite structure of dynamic exponential modulation and logarithmic smoothing compensation is required:

[0144] The first term uses the mutual information ratio as an exponent. When the acoustic emission is strongly correlated with pressure, the exponent increases, nonlinearly amplifying the weight of the time delay characteristics.

[0145] The second step introduces a logarithmic smoothing mechanism to prevent... When the value is too large, the fusion value explodes. At the same time, it ensures that when one type of feature is extremely weak, another type of feature can still be effectively preserved in the fusion space, thereby deeply exploring the hidden coupling distortion relationship across physical fields.

[0146] S303, the largest eigenvalue With mutual information value Substitute into the multiphysics coupling distortion feature fusion formula to calculate the fused eigenvalue. , will fuse feature values As a transient characteristic parameter of multiple physical quantities. The formula is:

[0147] ;

[0148] In the formula, This is the preset historical steady-state benchmark mutual information value.

[0149] It is used to fuse feature values; as the sole input vector of the support vector regression model, it is a dimensionless scalar that comprehensively characterizes the intensity and mode characteristics of multi-physics coupling distortion within the current transient interval.

[0150] It represents the maximum eigenvalue of the time-delay cross-correlation matrix; it characterizes the global principal component energy of vibration and current signals in the time-delay dimension, focusing on reflecting the linear topological correlation strength between mechanical rotor dynamics and electromagnetic dynamics. The larger the value, the stronger the topological dependence of the electromechanical coupling.

[0151] It represents the mutual information value between the acoustic emission signal sequence and the pressure signal sequence; it characterizes the degree of nonlinear statistical dependence between the two random variables of acoustics and fluid dynamics, focusing on reflecting the intrinsic correlation between minor mechanical collisions (such as the initial stage of blade fracture) and airflow / hydraulic pulsation. The larger the value, the more intense the fluid-structure interaction or acoustic-pressure correlation.

[0152] It is a preset historical steady-state baseline mutual information value; representing the background level of acoustic emission and pressure interaction information of the equipment under normal operating conditions, and is used as a normalization reference benchmark.

[0153] This reflects the surge multiple of the current transient acoustic-pressure correlation relative to the steady-state baseline. This ratio, as an exponent, directly determines the amplification factor of the time delay characteristics, enabling the fusion process to dynamically adapt to fault modes.

[0154] In the specific implementation process, the multi-physics field coupling distortion feature fusion formula is used to calculate the fusion feature value, so as to realize the adaptive nonlinear feature fusion when different physical fields alternate in dominance during the transient process.

[0155] By using the mutual information ratio as an exponent to nonlinearly modulate the time delay characteristics and introducing a logarithmic smoothing mechanism into the mutual information term, we can achieve adaptive representation and deep coupling relationship mining of the alternating dominance of different physical fields in the transient process.

[0156] S304, Fusion feature values ​​extracted from historical transient intervals as input vector The output vector is the amplitude sequence of the power consumption virtual high-frequency intrinsic mode function marked with the corresponding historical transient interval. , construct a training sample set.

[0157] S305. Select the Gaussian kernel function as the kernel function for support vector regression and set the penalty parameter. and insensitive loss parameters Solve the dual optimization problem of support vector regression to obtain the Lagrange multipliers and bias terms, and complete the training of the mapping model.

[0158] S306. Merge the characteristic values ​​of the current transient interval. The input is fed into the trained mapping model, and the regression prediction sequence output by the mapping model is used as the predicted value of the virtual height component corresponding to the current transient interval.

[0159] S400: Subtract the predicted value of the artificially high component from the original power data to obtain the corrected power consumption data. Calculate the exponential moving average of the corrected power consumption data within the steady-state range to update the power consumption monitoring baseline. When the real-time corrected power consumption data deviates from the power consumption monitoring baseline by more than a set threshold, a monitoring alarm signal is output. Specifically, this includes:

[0160] S401. Extract the power signal sequence corresponding to the transient interval. , power signal sequence Subtracting the amplitude of the predicted value of the artificial height component at the same timestamp from each sampling point, the resulting difference sequence is used as the corrected power consumption data. .

[0161] S402, Real-time calculation of vibration signal sequence Pressure signal sequence With current signal sequence When the variances of the three variables are all continuously lower than their respective preset variance thresholds for a set time period, the current interval is determined to be a steady-state interval.

[0162] In practice, even after eliminating transient high values ​​in industrial settings, steady-state power consumption data can still fluctuate due to minor electromagnetic interference and measurement noise.

[0163] Using the original corrected data directly as the baseline will cause the regulatory baseline to fluctuate, which can easily trigger false alarms.

[0164] S403. Extract corrected power consumption data within the steady-state range. Substitute the values ​​into the exponential moving average formula to calculate the exponential moving average:

[0165] ;

[0166] In the formula, For the current moment The exponential moving average is the latest power consumption regulatory baseline value output after smoothing at the current moment.

[0167] The preset smoothing coefficient has a range of values. Used to determine the sensitivity of the smoothing algorithm to the latest data:

[0168] The larger the value, the higher the proportion of the current observation, and the faster the baseline changes with new data, but the weaker the noise resistance.

[0169] The smaller the value, the higher the proportion of the historical baseline, and the smoother and more stable the baseline, but the less sensitive it is to actual power consumption fluctuations. It needs to be set according to the power consumption fluctuation characteristics of the equipment.

[0170] For the current moment The corrected power consumption data, which is the real instantaneous power consumption sample value after removing the artificially high components, is the latest input driving the baseline update.

[0171] For the previous moment The exponential moving average, i.e., the historical baseline value generated at the previous moment, implicitly condenses... Exponentially weighted information of all historical steady-state data up to and including the current time.

[0172] The calculated exponential moving average is updated as the electricity consumption regulatory baseline.

[0173] In practice, the exponential moving average is used to filter out high-frequency random noise interference within the steady-state range, extract the smooth core trend that reflects the true steady-state power consumption level of the equipment, and use this as a dynamically updated power consumption monitoring baseline.

[0174] By using a recursive accumulation mechanism, the current baseline value contains both the latest real power consumption drift information (ensuring tracking) and inherits the long-term trend framework of history (ensuring smoothness), thus providing an extremely stable and reliable reference benchmark for subsequent calculation of the relative deviation rate.

[0175] Because this formula is an iterative loop formula that relies on its own historical output as the input for the next step, it is particularly relevant during system cold starts (i.e., ...). When the first sampling point (or the first detection of a steady-state interval) is reached, there is no historical baseline value. The formula cannot be calculated directly. In this case, the initial value initialization method must be used, and the specific rules are as follows:

[0176] When the system first enters the steady-state region and the first corrected power consumption data point is collected. At that time, the initial baseline value is forced to be equal to the first observation value, that is, set... .

[0177] This setting makes the first iteration become... .

[0178] Although this hard assignment introduces some truncation error in the initial stage (because the first data point may contain random noise), due to the exponential decay memory property of the EMA formula, the initial outlier weights will decrease with the number of iterations. Its speed rapidly approaches zero exponentially.

[0179] After a short warm-up period (tens to hundreds of sampling points), the impact of the initial artificially assigned values ​​will be completely diluted by the subsequent massive amount of real steady-state data, and the baseline will naturally converge to the real steady-state smooth center, without causing substantial interference to subsequent long-term regulatory alarms.

[0180] S404. Real-time acquisition of current corrected power consumption data within the non-transient interval. Simultaneously read the current power consumption monitoring baseline. .

[0181] S405, Calculate the current corrected power consumption data With electricity consumption regulatory baseline relative deviation rate The relative deviation rate Deviation rate threshold Comparison, if the relative deviation rate Greater than the deviation rate threshold If the duration exceeds the set hysteresis time, a regulatory alarm signal will be generated; among which, .

[0182] Example 2: The present invention also provides a power consumption data monitoring system for high power-consuming industrial equipment, comprising:

[0183] The multi-physical quantity synchronous acquisition module can synchronously acquire current signal sequences, power signal sequences, vibration signal sequences, pressure signal sequences, and acoustic emission signal sequences during the operation of industrial equipment by setting a sampling frequency.

[0184] The transient interval location module calculates the first-order difference extremum points of the signal sequence and quantifies the energy product of acoustic emission and current and the time difference attenuation characteristics according to the electromechanical coupling transient activation exponent formula to identify the transient start point and extract the transient interval.

[0185] The virtual high component extraction module performs empirical mode decomposition on the power data in the transient interval to obtain the intrinsic mode functions, and combines the vibration envelope spectrum decay characteristics to screen out the virtual high frequency intrinsic mode functions of power consumption.

[0186] The feature fusion and mapping module constructs the time-delay cross-correlation matrix of vibration and current and the mutual information value of acoustic emission and pressure. It calculates the fused feature value through the multi-physics field coupling distortion feature fusion formula and outputs the predicted value of the virtual height component based on the support vector regression model.

[0187] The dynamic power consumption monitoring module subtracts the predicted value of the artificially high power from the original power to obtain the corrected power consumption data, calculates the exponential moving average of the corrected power consumption data in the steady-state range to update the monitoring baseline, and outputs a monitoring alarm signal when the real-time corrected power consumption data deviates from the monitoring baseline and exceeds the limit.

[0188] 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.

[0189] 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 power consumption data monitoring method for high-power-consuming industrial equipment, characterized in that: The method includes: S100 synchronously collects current data, power data, vibration data, pressure data and acoustic emission data of industrial equipment, extracts the first-order difference extreme points of current data and acoustic emission data, and calculates the electrical coupling transient activation index based on the time difference of extreme points to locate the transient interval. S200. Perform empirical mode decomposition on the power data in the transient interval to obtain multiple intrinsic mode functions, calculate the kurtosis value of each intrinsic mode function, and combine the energy decay characteristics of the envelope spectrum of the vibration data to screen out the power consumption virtual high frequency intrinsic mode function from the intrinsic mode functions. S300. Construct the time delay cross-correlation matrix of vibration data and current data, calculate the mutual information value of acoustic emission data and pressure data, and nonlinearly exponentially fuse the eigenvalues ​​and mutual information values ​​of the time delay cross-correlation matrix into multi-physical quantity transient characteristic parameters. Use the support vector regression algorithm to establish a mapping model between the multi-physical quantity transient characteristic parameters and the power consumption virtual high frequency intrinsic mode function, and output the virtual high component prediction value. S400: Subtract the predicted value of the artificially high component from the original power data to obtain the corrected power consumption data. Calculate the exponential moving average of the corrected power consumption data within the steady-state range to update the power consumption monitoring baseline. When the real-time corrected power consumption data deviates from the power consumption monitoring baseline by more than a set threshold, output a monitoring alarm signal.

2. The power consumption data supervision method for high-power-consuming industrial equipment according to claim 1, characterized in that: S100 includes: S101, synchronously collect a current signal sequence of the industrial equipment at a set sampling frequency , a power signal sequence , a vibration signal sequence , a pressure signal sequence , and an acoustic emission signal sequence ; S102, respectively calculate the first-order difference sequence of the acoustic emission signal sequence , the first-order difference sequence of the current signal sequence , and the first-order difference sequence of the voltage signal sequence . ; S103, extracting a first-order difference sequence a first time point set corresponding to the maximum value points with absolute values greater than a first preset threshold , and the first-order difference sequence a second time point set corresponding to the maximum value points with absolute values greater than a second preset threshold ; S104, performing interval check on adjacent extreme points in the first time point set and the second time point set If the interval between two adjacent extreme points is less than the minimum interval threshold, the extreme point with smaller amplitude is removed and the extreme point with larger amplitude is retained. S105, traversing each time point in the second time point set , retrieving the previous time point with the smallest time difference in the first time point set , and calculating the computer electrically coupled transient activation index. S106. When the electromechanical coupling transient activation index is greater than the preset activation threshold, the second time point set... The time point marked in the time frame is the transient start point, and the interval extending backward from the transient start point for a set time length is defined as the transient interval. S107. Set the first time point after processing in chronological order. The preceding time point was retrieved and the computer-coupled transient activation index was obtained.

3. The method for monitoring power consumption data of high-power-consuming industrial equipment according to claim 2, characterized in that: The electromechanical coupling transient activation index The calculation formula is: ； In the formula, For the first time point set The absolute amplitude of the first-order difference extremum point of the pre-sequence acoustic emission; For the second time point set The absolute amplitude of the first-order difference extreme point of the current in the middle; The time difference between the two extreme points; The preset electromechanical propagation delay constant; This is the preset time tolerance standard deviation.

4. The method for monitoring power consumption data of high-power-consuming industrial equipment according to claim 2, characterized in that: S200 includes: S201. Extract the power signal sequence within the transient interval. Performing empirical mode decomposition yields N intrinsic mode functions. to ; S202, Calculate each intrinsic mode function kurtosis value Filter out kurtosis values Greater than the kurtosis threshold The intrinsic mode functions constitute a high-frequency mode set; S203, Regarding the vibration signal sequence within the transient interval Bandpass filtering and Hilbert transform are performed to obtain the envelope spectrum. The envelope curve, which shows the energy of the envelope spectrum changing over time, is calculated, and the attenuation coefficient of the fitted envelope curve is determined. ; S204. Calculate the intrinsic mode function in the high-frequency mode set. center frequency If the center frequency Falling on the attenuation coefficient Within the corresponding mechanical resonance frequency band, the inherent mode function is... It is labeled as the imaginary high-frequency intrinsic mode function of power consumption.

5. A method for monitoring power consumption data of high-power-consuming industrial equipment according to claim 4, characterized in that: The S300 includes: S301, The vibration signal sequence within the transient interval With current signal sequence The system is divided into multiple segments using a sliding window. A cross-correlation function is calculated for each segment, and the time delay corresponding to the peak value of the cross-correlation function is identified. Construct latency values ​​between different sub-segments The time delay covariance matrix is ​​used as the time delay cross-correlation matrix, and the maximum eigenvalue of the time delay cross-correlation matrix is ​​calculated. ; S302, the acoustic emission signal sequence within the transient interval With pressure signal sequence The signal is divided into equal-length segments. The joint probability density and marginal probability density of the acoustic emission signal sequence and the pressure signal sequence for each segment are calculated using the kernel density estimation method. The mutual information value is then calculated based on the joint probability density and marginal probability density. ; S303, the largest eigenvalue With mutual information value Substitute into the multiphysics coupling distortion feature fusion formula to calculate the fused eigenvalue. , will fuse feature values As a transient characteristic parameter of multiple physical quantities.

6. A method for monitoring power consumption data of high-power-consuming industrial equipment according to claim 5, characterized in that: The formula for fusing multiphysics field coupling distortion features is as follows: ； In the formula, This is the preset historical steady-state benchmark mutual information value.

7. A method for monitoring power consumption data of high-power-consuming industrial equipment according to claim 5, characterized in that: The S300 also includes: S304, Fusion feature values ​​extracted from historical transient intervals as input vector The output vector is the amplitude sequence of the power consumption virtual high-frequency intrinsic mode function marked with the corresponding historical transient interval. Construct a training sample set; S305. Select the Gaussian kernel function as the kernel function for support vector regression and set the penalty parameter. and insensitive loss parameters Solve the dual optimization problem of support vector regression to obtain the Lagrange multipliers and bias terms, and complete the training of the mapping model; S306. Merge the characteristic values ​​of the current transient interval. The input is fed into the trained mapping model, and the regression prediction sequence output by the mapping model is used as the predicted value of the virtual height component corresponding to the current transient interval.

8. A method for monitoring power consumption data of high-power-consuming industrial equipment according to claim 7, characterized in that: The S400 includes: S401. Extract the power signal sequence corresponding to the transient interval. , power signal sequence Subtracting the amplitude of the predicted value of the artificial height component at the same timestamp from each sampling point, the resulting difference sequence is used as the corrected power consumption data. ; S402, Real-time calculation of vibration signal sequence Pressure signal sequence With current signal sequence When the variances of the three variables are all continuously lower than their respective preset variance thresholds for a set time period, the current interval is determined to be a steady-state interval. S403. Extract corrected power consumption data within the steady-state range. Substitute the values ​​into the exponential moving average formula to calculate the exponential moving average: ； In the formula, For the current moment The exponential moving average; The preset smoothing coefficient has a range of values. ; For the current moment Corrected power consumption data; For the previous moment The calculated exponential moving average is then updated to the electricity consumption regulatory baseline. S404. Real-time acquisition of current corrected power consumption data within the non-transient interval. Simultaneously read the current power consumption monitoring baseline. ; S405, Calculate the current corrected power consumption data With power consumption regulatory baseline relative deviation rate The relative deviation rate Deviation rate threshold Comparison, if the relative deviation rate Greater than the deviation rate threshold If the duration exceeds the set hysteresis time, a regulatory alarm signal will be generated; among which, .

9. A power consumption data monitoring system for high power-consuming industrial equipment, characterized in that: A method for monitoring power consumption data of high-power-consuming industrial equipment as described in any one of claims 1 to 8, the system comprising: The multi-physical quantity synchronous acquisition module can synchronously acquire current signal sequences, power signal sequences, vibration signal sequences, pressure signal sequences, and acoustic emission signal sequences during the operation of industrial equipment by setting a sampling frequency; The transient interval location module calculates the first-order difference extreme points of the signal sequence and quantifies the energy product of acoustic emission and current and the time difference attenuation characteristics according to the electromechanical coupling transient activation exponent formula to identify the transient start point and extract the transient interval. The virtual high component extraction module performs empirical mode decomposition on the power data in the transient interval to obtain the intrinsic mode functions, and combines the vibration envelope spectrum decay characteristics to screen out the virtual high frequency intrinsic mode functions of power consumption; The feature fusion and mapping module constructs the time delay cross-correlation matrix of vibration and current and the mutual information value of acoustic emission and pressure. It calculates the fused feature value through the multi-physics field coupled distortion feature fusion formula and outputs the predicted value of the virtual height component based on the support vector regression model. The dynamic power consumption monitoring module subtracts the predicted value of the artificially high power from the original power to obtain the corrected power consumption data, calculates the exponential moving average of the corrected power consumption data in the steady-state range to update the monitoring baseline, and outputs a monitoring alarm signal when the real-time corrected power consumption data deviates from the monitoring baseline and exceeds the limit.

Images