Intelligent frequency conversion control method, device and system based on dynamic data acquisition

Abstract

The invention discloses an intelligent frequency conversion control method, device and system based on dynamic data acquisition, and the method comprises the steps: obtaining the electromagnetic parameter waveform of a motor in real time, carrying out the deep analysis of features through a machine learning algorithm, and achieving the recognition of the abnormal recognition precision; synchronously acquiring a rotor position and an evaluation parameter set, executing an exception analysis strategy, generating a risk evaluation coefficient, and accurately locking an exception moment; in the process, through a synergistic effect of quantizing angle deviation, current abnormity and sampling delay, a mathematical framework is provided for fault prediction, and risks are reduced in the acquisition process; and under the condition that the risk assessment coefficient is received, the risk assessment coefficient and a preset assessment threshold are compared and analyzed, and whether a corresponding frequency conversion control instruction is executed or not is selected according to a comparison result, so that high-precision and high-efficiency intelligent frequency conversion control on the motor is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of frequency converter control, and in particular to an intelligent frequency conversion control method, device and system based on dynamic data acquisition. Background Art

[0002] With the rapid development of industrial automation and intelligent technology, frequency conversion control technology has been widely used in the fields of motor speed regulation, energy saving and precise control. By adjusting the working frequency of the motor, the speed of the motor can be effectively controlled to achieve efficient operation under different working conditions. On the one hand, traditional variable frequency control systems often collect real-time data through sensors, but the single data collection method, limited data processing capabilities and fixed control strategies result in the inability to accurately perform variable frequency actions during subsequent analysis due to noise in the data; however, as the working state of the motor becomes more and more complex, the use of big data and artificial intelligence technologies can predict the state of the motor and optimize the control strategy. However, the current state prediction of the motor mainly relies on a single data source or a simple model, which makes it difficult to fully and accurately reflect the complex state of the motor; On the other hand, in the actual operation of the motor and the frequency conversion controller, the dynamic coupling effect between the direct axis (d-axis) and the quadrature axis (q-axis) and the nonlinear characteristics of the electromagnetic parameters often make it difficult for traditional models to accurately characterize the motor behavior under complex working conditions; for example, traditional methods often ignore the mutual influence of the direct and quadrature axis parameters under different load conditions. At the same time, the response processing of electromagnetic parameter characteristics (such as inductance and magnetic flux saturation characteristics) and injection signals (such as high-frequency excitation signals) has the problem of insufficient accuracy, which makes the system insufficiently adaptable and flexible. To a certain extent, it is impossible to analyze and feedback the mechanism in a timely manner, which reduces the intelligence of the frequency conversion control system and makes it difficult to cope with dynamic and complex working environments and load changes. Summary of the invention

[0003] 1. Technical issues to be resolved In view of the deficiencies in the prior art, the present invention provides an intelligent frequency conversion control method, device and system based on dynamic data acquisition, which deeply analyzes the first operating information of the motor, forms a first graph structure through the first operating data, forms a second graph structure based on the first attribute data, and also associates the first operating data and the first attribute data with each other to create a third graph structure, which can comprehensively capture various factors in the operation of the motor, improve the accuracy and intelligence level of the motor state prediction, and output the health state of the first motor; when an abnormal state is identified, the abnormal evaluation coefficient is generated to execute the first frequency conversion action, and when it is determined that a fluctuation deviation occurs, an optimized frequency conversion strategy is executed to achieve high-precision and high-efficiency intelligent frequency conversion control of the motor, thereby solving the problems raised in the background technology.

[0004] (II) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: In a first aspect, the present invention provides an intelligent frequency conversion control method based on dynamic data acquisition, comprising: Obtain the electromagnetic parameters of the target motor. When the monitored electromagnetic parameters generate waveform signals, execute the acquisition adjustment plan, analyze and process the waveform signals in combination with machine learning algorithms, and identify and classify the acquisition adjustment types. Under the condition of receiving the acquisition adjustment type, the rotor position and the evaluation parameter set of the motor are obtained within the first time period, the rotor position is compared and analyzed with the preset standard rotor position, the time corresponding to the deviation state is obtained, and the abnormal analysis strategy is executed in combination with the preprocessed evaluation parameter set to generate a risk assessment coefficient; Upon receiving the risk assessment coefficient, the risk assessment coefficient is compared and analyzed with the preset assessment threshold, and whether to execute the corresponding frequency conversion control instruction is selected according to the comparison result; The target motor operating parameters are obtained and a topological space is constructed. The risk assessment coefficient is embedded as a feature to form a graph structure, and an updated model of the graph structure is obtained to obtain a state prediction model. When the variable frequency control instruction is triggered, the model optimization strategy is executed within a preset second time period.

[0005] Furthermore, the electromagnetic parameters include stator resistance, quadrature-axis inductance, direct-axis inductance and permanent magnet flux; the waveform signal includes square wave, trapezoidal wave, triangle wave and sine wave; and the step of analyzing and processing the waveform characteristics in combination with a machine learning algorithm includes: Signal processing: filter the captured waveform signal to extract the characteristics of the waveform signal, including at least amplitude, frequency, phase, envelope, and harmonics; Abnormal identification: When a square wave or trapezoidal wave is identified, a local constant signal is injected; when a triangle wave or sine wave is identified, a global time-varying signal is injected; at the same time, a decision tree algorithm is used to train the extracted features and the injected signals; during the training process, a parameter matrix is ​​constructed based on electromagnetic parameters, and a weight factor is obtained based on the numerical change characteristics of the electromagnetic parameters, and a score is assigned to the weight factor, and the weight factor and the score are multiplied to obtain the recognition accuracy coefficient; the numerical change characteristics include mean value, kurtosis, kurtosis, quartile deviation and variance; Adjustment identification: In the real-time monitoring process, the direct and quadrature axis current equations are combined as constraints and mapped to electromagnetic parameters to obtain a corresponding set of training samples; the identification accuracy coefficient corresponding to each set of training samples is used as the sample label, and the classification adjustment model is obtained through training with the training samples to output the amplitude and frequency of the injection signal corresponding to the current waveform signal.

[0006] Furthermore, the loss function of the hidden layer of the classification adjustment model is as follows: ; Where L represents the loss function of the hidden layer of the classification adjustment model, α1 and α2 are weight loss coefficients, N represents the number of training samples in each group, N represents the number of samples in any group of training samples, Ca represents the parameter matrix, α0 represents the adjustment factor, Indicates the first The prediction accuracy coefficient of samples is Indicates the first The identification accuracy coefficient of samples is represents the recognition accuracy coefficient of the jth sample in a group, represents the constraint function, y i Indicates the first The constraint term of samples, y i represents the constraint item of the jth sample in a group; ; In the formula, y represents the constraint condition, f dq (·) represents the nonlinear function under the d-axis or q-axis, wave represents the waveform signal, and R s Indicates stator resistance, L d Indicates the quadrature-axis inductance, L q represents the direct-axis inductance, Ψ f represents the permanent magnet flux, and ω represents the weight correction coefficient.

[0007] Furthermore, the evaluation parameter set includes a sampling delay rate, a current component, and a motor electrical angle deviation; and the step of determining the deviation state includes: Obtain the difference between the rotor position and the preset standard rotor position, and set the gradient range. When the difference is within the preset gradient range, it is determined that there is no deviation, otherwise, it is determined to be a deviation state; At the same time, an evaluation parameter set is obtained, and a set parameter set is obtained from a database, including a sampling delay rate threshold, a direct-axis current threshold, a quadrature-axis current threshold, and an angle deviation threshold; After dimensionless processing and weighted calculation, the deviation factor is generated; The formula is: ; In the formula, fault represents the deviation factor, late represents the sampling delay rate, lates represents the sampling delay rate threshold, b0 represents the weight correction coefficient, which is used to strengthen the marginal risk of high delay, and its value is generally 2-3. Dev1 represents the current component of the direct axis, Devs1 represents the DC current threshold, Dev2 represents the current component of the quadrature axis, Devs2 represents the quadrature axis current threshold, Ang represents the motor electrical angle deviation, and Angs represents the angle deviation threshold. B1, B2, and B3 are all weight coefficients.

[0008] Furthermore, the steps of executing the anomaly analysis strategy include: Collect the moments judged as deviation states, randomly extract two time series, obtain abnormal time periods, sort the abnormal time periods in time series and mark them as T1, T2, ..., Tn; where the time series is a dynamic change, and n represents the number of times it is judged as a deviation state; divide the abnormal time period into equal parts and mark it as r, and when it has a remainder, it is r+1, where r is a positive integer and a dynamic value; combined with the deviation factor, calculate the risk assessment coefficient of the corresponding abnormal time period under r or r+1.

[0009] Furthermore, the risk assessment coefficient is compared and analyzed with the preset assessment threshold: Calculate the ratio between the risk assessment coefficient and the risk assessment threshold, and send different frequency conversion adjustment signals according to the ratio; where the ratio = risk assessment coefficient / risk assessment threshold; When the ratio is less than 0.8, no response operation; When 0.8≤ratio≤1.2, a type 1 adjustment signal is issued and the first frequency conversion strategy is executed; When the ratio is greater than 1.2, a second type of adjustment signal is issued and the second frequency conversion strategy is executed.

[0010] Furthermore, the first frequency conversion strategy algorithm model is: ; Where V new Indicates the corrected voltage, V base Indicates the current voltage, f newIndicates the corrected motor operating frequency, f base Indicates the current motor operating frequency, K V Indicates the voltage regulation coefficient; acc new Indicates the corrected adjustment speed, acc base Indicates the current adjustment speed; t new represents the time control period of the correction, t base Indicates the current control cycle; Get the current operating frequency. When it exceeds the preset frequency threshold, it is marked as a high frequency band; otherwise, it is marked as a low frequency band and the voltage output is corrected proportionally. Under low frequency conditions: ; Under high frequency conditions: ; ; Where V new Indicates the corrected voltage, V base Indicates the current voltage, f new Indicates the corrected motor operating frequency, f base Indicates the current motor operating frequency, acc new Indicates the corrected adjustment speed, acc base Indicates the current adjustment speed; t new represents the time control period of the correction, t base Indicates the current control cycle.

[0011] Furthermore, the steps of executing the model optimization strategy include: Obtain the duration and frequency of the deviation state within the corresponding time period to calculate the time complexity; Obtain the number of responses to the variable frequency adjustment signal and the variable frequency adjustment ratio within the corresponding time period to calculate the action complexity; Based on the time complexity and action complexity, the target optimization function min(fz1, fz2) is constructed, and the target optimization function is used as the connection function of the state prediction model. The number of nodes in the last hidden layer in the state prediction model is consistent with the number of parameters involved in the impedance formula of the target optimization function. The weights of the last layer of the state prediction model are initialized, and the gradient descent algorithm is used for learning. The hyperparameters are adjusted through the training set, and the performance of the model is evaluated through the validation set and the test set to retrain the entire network model to achieve model optimization.

[0012] In a second aspect, the present invention provides an intelligent frequency conversion control device based on dynamic data collection, comprising: The waveform monitoring management module obtains the electromagnetic parameters of the target motor. When the monitored electromagnetic parameters generate waveform signals, the acquisition adjustment plan is executed. The waveform signals are analyzed and processed in combination with machine learning algorithms to identify and classify acquisition adjustment types. The abnormality assessment module, under the condition of receiving the acquisition adjustment type, obtains the rotor position and the evaluation parameter set of the motor within the first time period, compares and analyzes the rotor position with the preset standard rotor position, obtains the time corresponding to the deviation state, combines the preprocessed evaluation parameter set, executes the abnormality analysis strategy, and generates the risk assessment coefficient; The feedback adjustment module, upon receiving the risk assessment coefficient, compares and analyzes the risk assessment coefficient with a preset assessment threshold, and selects whether to execute the corresponding frequency conversion control instruction according to the comparison result; A predictive optimization model is developed to obtain the target motor operating parameters and construct a topological space. The risk assessment coefficient is embedded as a feature to form a graph structure, and an updated model of the graph structure is obtained to obtain a state prediction model. When a variable frequency control instruction is triggered, the model optimization strategy is executed within a preset second time period.

[0013] In a third aspect, the present invention provides an intelligent variable frequency control system based on dynamic data acquisition, comprising: A frequency converter, used for outputting a frequency conversion adjustment signal; At least two DC power supplies, used to supply power to the frequency converter; A variable frequency control device obtains the electromagnetic parameters of a target motor. When the monitored electromagnetic parameters generate a waveform signal, an acquisition adjustment plan is executed, the waveform signal is analyzed and processed in combination with a machine learning algorithm, and the acquisition adjustment type is identified and classified; under the condition of receiving the acquisition adjustment type, the rotor position and the evaluation parameter set of the motor are obtained within a first time period, the rotor position is compared and analyzed with a preset standard rotor position, the moment corresponding to the deviation state is obtained, and the abnormal analysis strategy is executed in combination with the pre-processed evaluation parameter set to generate a risk assessment coefficient; under the condition of receiving the risk assessment coefficient, the risk assessment coefficient is compared and analyzed with a preset evaluation threshold, and whether to execute the corresponding variable frequency control instruction is selected according to the comparison result; the operating parameters of the target motor are obtained and a topological space is constructed, the risk assessment coefficient is embedded as a feature to form a graph structure, and an updated model of the graph structure is obtained to obtain a state prediction model; under the triggering of the variable frequency control instruction, a model optimization strategy is executed within a preset second time period.

[0014] (III) Beneficial effects The present invention provides an intelligent frequency conversion control method, device and system based on dynamic data acquisition, which has the following beneficial effects: 1. The present invention embeds the physical laws of direct and quadrature axes into the data-driven model through the designed loss function, the dual design of single sample precision enhancement and inter-sample relationship constraints, solves the limitations of traditional methods under dynamic coupling conditions, and provides an optimization framework for motor control that takes into account accuracy, robustness and safety; 2. The present invention analyzes the deviation of the rotor position and quantifies the synergistic effect of angle deviation, current anomaly and sampling delay, which not only provides a mathematical framework for fault prediction but also reduces the risk in the acquisition process. In the analysis of the rotor position deviation, the moment determined to be a deviation state is collected, the anomaly analysis strategy is executed, the risk assessment coefficient is generated, the ratio between the risk assessment coefficient and the risk assessment threshold is calculated, and different risk signals are divided. Under high-risk or medium-risk signals, the first frequency conversion strategy or the second frequency conversion strategy is executed accordingly, and the mechanical energy consumption of the motor is changed through the frequency conversion controller to keep the motor in a normal state at all times. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a flow chart of an intelligent frequency conversion control method according to an exemplary embodiment; Figure 2 is a schematic diagram of steps of an intelligent frequency conversion control method according to an exemplary embodiment; Figure 3 It is a module schematic diagram of an intelligent frequency conversion control device according to an exemplary embodiment. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example 1

[0017] The embodiment of the present invention provides an intelligent frequency conversion control method based on dynamic data acquisition; Figure 1 is a flow chart of an intelligent frequency conversion control method according to an exemplary embodiment; Figure 2 is a schematic diagram of the steps of an intelligent frequency conversion control method according to an exemplary embodiment; please refer to Figure 1 - Figure 2 , the method comprising: The waveform monitoring management module obtains the electromagnetic parameters of the target motor. When the monitored electromagnetic parameters generate waveform signals, the acquisition adjustment plan is executed, and the waveform signals are analyzed and processed in combination with machine learning algorithms to identify and classify the acquisition adjustment types. Among them, the electromagnetic parameters include stator resistance, quadrature-axis inductance, direct-axis inductance, and permanent magnet flux; the waveform signals include square waves, trapezoidal waves, triangle waves, and sine waves. The steps of analyzing and processing waveform features in combination with machine learning algorithms include: Signal processing: Filter the captured waveform signal to extract the characteristics of the waveform signal, including at least amplitude, frequency, phase, envelope, and harmonics; transform domain analysis technology (such as Fourier transform, wavelet transform, etc.) can also be used to extract the characteristics of the signal in the transform domain; Abnormal identification: When a square wave or trapezoidal wave is identified, a local constant signal is injected; when a triangle wave or sine wave is identified, a global time-varying signal is injected; at the same time, a decision tree algorithm is used to train the extracted features and the injected signals; during the training process, a parameter matrix is ​​constructed based on electromagnetic parameters, and a weight factor is obtained based on the numerical change characteristics of the electromagnetic parameters, and a score is assigned to the weight factor, and the weight factor and the score are multiplied to obtain the recognition accuracy coefficient; the numerical change characteristics include mean value, kurtosis, kurtosis, quartile deviation and variance; The step of constructing a parameter matrix based on electromagnetic parameters includes: Statistical analysis: Based on the time sequence, the electromagnetic parameters with the same order are grouped together to construct a parameter matrix, which is represented as follows: ; Where Ca represents the parameter matrix, each column vector represents the data value of a data point at each time, and each row vector represents the electromagnetic parameters at a certain time, including stator resistance, quadrature-axis inductance, direct-axis inductance, and permanent magnet flux; C 1,1 , C 1,2 , C 1,3 and C 1,4 is the electromagnetic parameter at the first moment, C 2,1 , C 2,2 , C 2,3 and C 2,4 is the electromagnetic parameter at the second moment, C m,1 , C m,2 , C m,3 and C m,4 is the electromagnetic parameter at the mth moment; φ1, φ2, φ3 and φ4 are all weight factors, and φ1+φ2+φ3+φ4=1; Formula explanation: The numerical variation characteristics can reflect the distribution, volatility, peak degree, skewness and dispersion of electromagnetic parameters. For example, determine the influence of each feature on the identification accuracy coefficient. For example, a large variance may indicate a large fluctuation of the parameter, which has a greater impact on the accuracy, so the weight is higher; kurtosis and kurtosis may reflect the abnormality of the parameter. A high kurtosis means that there are more extreme values ​​in the data set, which may affect the accuracy; the interquartile range reflects the dispersion of the data, which is similar to the variance, but more robust. In the process of analyzing the weight factor, the statistical characteristics of each parameter can be weighted and summed, and then the weighted values ​​of all parameters are integrated to obtain the overall value, which is defined as the weight factor; each statistical characteristic can be used as a dimension, and the contribution of each parameter can be calculated through a certain function (such as linear combination, product, etc.), and then the results of all parameters are integrated and defined as the weight factor; in addition, under variable working conditions, the contribution to each parameter can be adjusted in real time according to the speed / load (such as focusing on variance at high speed and focusing on kurtosis at low speed); the weight factor of each parameter can also be determined by the particle swarm optimization algorithm. The particle swarm optimization algorithm (PSO) is a global optimization algorithm that simulates the foraging behavior of bird flocks and is often used for feature selection and optimization problems; The specific process includes: Randomly generate the initial population and set the number of iterations to t max , the population is composed of M1 particles, and the position and speed of each particle are defined. The position of each particle can represent a possible feature weight combination (i.e., the matrix vector formed by the weight factors of each target data sequence, represented as [φ1, φ2, φ3, φ4] in this embodiment), and the speed of each particle determines the search direction in the feature space; the binary bidding selection strategy is used to calculate and obtain the mixed population, and mutation and crossover operations are introduced to avoid the algorithm from falling into the local optimum in the process of increasing the diversity of the particle swarm; After each iteration, the fitness function is calculated, and the ideal point Fmin of the mixed population is obtained based on the fitness. Each objective function value at the ideal point Fmin takes the minimum value of the mixed population on the objective to obtain the optimal position and optimal speed of the particle, and the particle is defined as the optimal particle, so as to select an optimal particle from each sub-population as the elite and put it into the offspring population, and repeat the cycle until the iteration is stopped; the acquisition of the weight factor is set according to the design situation, and the specific process will not be described in detail; The assigned score is consistent with the weight factor. The score can be the same as the weight factor, or the score can be a multiple of the weight factor. Adjustment identification: In the real-time monitoring process, the DC and AC axis current equations are combined as constraints and mapped to electromagnetic parameters to obtain a corresponding set of training samples. The identification accuracy coefficient corresponding to each set of training samples is used as the sample label. The classification adjustment model is trained through the training samples and the amplitude and frequency of the injection signal corresponding to the current waveform signal are output, that is, the acquisition adjustment type corresponding to the current waveform signal is obtained, which is used for the subsequent accurate acquisition of relevant parameters of the motor (for example: motor operating parameters); Among them, the loss function of the hidden layer of the classification adjustment model is as follows: ; Where L represents the loss function of the hidden layer of the classification adjustment model, α1 and α2 are weight loss coefficients, N represents the number of training samples in each group, N represents the number of samples in any group of training samples, i represents the index of N, j represents the index of N, Ca represents the parameter matrix, α0 represents the adjustment factor, and α0 is greater than 0, which is used to dynamically adjust the error sensitivity. Indicates the first The prediction accuracy coefficient of samples is Indicates the first The identification accuracy coefficient of samples is represents the recognition accuracy coefficient of the jth sample in a group, represents the constraint function, y i Indicates the first The constraint term of samples, y i represents the constraint term of the jth sample in a group; exp(·) is an exponential function, for example: at high temperature or magnetic field saturation, the error Large, using exponential function to suppress gradient explosion and avoid overfitting transient noise of the model; in, ; In the formula, y represents the constraint condition, f dq (·) represents the nonlinear function under the d-axis or q-axis, wave represents the waveform signal, and R s Indicates stator resistance, L d Indicates the quadrature-axis inductance, L q represents the direct-axis inductance, Ψ f represents permanent magnet flux linkage, ω represents weight correction coefficient, and different waveform signals correspond to different weight correction coefficients. According to a large amount of data analysis, the weight correction coefficient corresponding to the square wave is set to be greater than the weight correction coefficient corresponding to the trapezoidal wave, greater than the weight correction coefficient corresponding to the sine wave, and greater than the weight correction coefficient corresponding to the triangle wave; Specifically, this embodiment adopts: Constraints: In the actual operation of the motor and the frequency converter, the dynamic coupling effect between the direct axis (d-axis) and the quadrature axis (q-axis) and the nonlinear characteristics of the electromagnetic parameters often make it difficult for traditional models to accurately describe the motor behavior under complex working conditions. Therefore, the response processing of electromagnetic parameter characteristics (such as inductance and flux saturation) and injected signals (such as local constant signals) is insufficiently accurate. To this end, by introducing d-axis and q-axis constraints, the physical boundary conditions for parameter identification are constructed, and the identification accuracy coefficient is adjusted based on the constraints to quantitatively evaluate the effectiveness of feature extraction, thereby explaining the reasons for model failure under complex working conditions. For example, in weakly constrained conditions (such as steady-state operation), the electromagnetic parameter feature extraction is optimized first to ensure basic identification accuracy. In strongly constrained conditions (such as load mutations), the accuracy requirements are relaxed, and the robustness is enhanced by injecting signals to suppress the influence of parameter perturbations. Signal injection: By injecting a local constant signal or a global time-varying signal into the motor, the response analysis enhances the sensitivity to parameter changes, making the identification accuracy coefficient more accurate. It should be noted that the local constant signal or the global time-varying signal can be expressed as a signal with a specific frequency and amplitude (for example: a high-frequency voltage or current signal). Usually, the value of the local constant signal is greater than the global time-varying signal; Double summation logic: prevents the model from having parameter estimation polarization under extreme conditions (such as stalling and regenerative braking); In summary, through constraints, signal injection, and dual summation logic, the physical laws of direct and quadrature axes are embedded in the data-driven model through the dual design of single sample precision enhancement and inter-sample relationship constraints, which solves the limitations of traditional methods under dynamic coupling conditions and provides an optimization framework for motor control that takes into account accuracy, robustness, and safety. The abnormality prediction module, under the condition of receiving the acquisition adjustment type, obtains the rotor position and the evaluation parameter set of the motor within the first time period, compares and analyzes the rotor position with the preset standard rotor position, obtains the time corresponding to the deviation state, and executes the abnormality analysis strategy in combination with the preprocessed evaluation parameter set to generate a risk assessment coefficient; wherein the evaluation parameter set includes the sampling delay rate, the current component and the motor electrical angle deviation; The step of determining the deviation state includes: Obtain the difference between the rotor position and the preset standard rotor position, and set the gradient range. When the difference is within the preset gradient range, it is determined that there is no deviation, otherwise, it is determined to be a deviation state; Among them, the preprocessing includes denoising and adaptive smoothing; Acquire the sampling delay rate, current component and motor electrical angle at the same time, and obtain the set parameter set from the database, including the sampling delay rate threshold, the direct axis current threshold, the quadrature axis current threshold and the angle deviation threshold; After dimensionless processing and weighted calculation, the deviation factor is generated; The formula is: ; In the formula, fault represents the deviation factor, late represents the sampling delay rate, lates represents the sampling delay rate threshold, b0 represents the weight correction coefficient, which is used to strengthen the marginal risk of high delay, and is generally taken as 2-3, dev1 represents the current component of the direct axis, devs1 represents the DC current threshold, dev2 represents the current component of the quadrature axis, devs2 represents the quadrature axis current threshold, and the product form captures the coordinated abnormality of the current component, ang represents the motor electrical angle deviation, angs represents the angle deviation threshold, and the square term amplifies the impact of significant deviations, reflecting the nonlinear risk accumulation characteristics; b1, b2, and b3 are all weight coefficients, and b1, b2, and b3 are all greater than 0, and b1+b2+b3=1; Example: Data 1: Assume that at a certain moment, set late / lates=0.8, take b0=2, dev1 / devs1=1.1, dev2 / devs2=0.9, ang / angs=3 / 5, b1=0.2, b2=0.3, b3=0.5; Then fault=0.2*exp(2*0.8)+0.3*(1.1*0.9)+0.5*(3 / 5) 2 =1.467; Data 2: Assume that at a certain moment, set late / lates=0.9, take b0=2, dev1 / devs1=1.1, dev2 / devs2=0.9, ang / angs=3 / 5, b1=0.2, b2=0.3, b3=0.5; Then fault=0.2*exp(2*0.9)+0.3*(1.1*0.9)+0.5*(3 / 5) 2 =1.6869; Comparing data 1 and data 2, 1.6869>1.467, indicating that the risk level at the corresponding moment of data 1 is higher than the risk level at the corresponding moment of data 2; The steps to execute the anomaly analysis strategy include: Collect the moments determined as deviation states, randomly extract two time series, obtain abnormal time periods, sort the abnormal time periods in time series and mark them as T1, T2, ..., Tn; where the time series is a dynamic change, and n represents the number of times it is determined as a deviation state; divide the abnormal time period into equal parts and mark it as r, and when there is a remainder, it is r+1, and r is a positive integer and a dynamic value; Combined with the deviation factor, the risk assessment coefficient corresponding to the abnormal time period under r or r+1 is calculated respectively; When it is the abnormal time period corresponding to r, ; When it is the abnormal time period corresponding to (r+1), ; In the formula, risk represents the risk assessment coefficient, T risk Indicates the abnormal time period, which is calculated as the duration between the start and end time of the abnormal time period, and the values ​​are T1, T2, ..., Tn; It represents the average value of the deviation factor obtained in any abnormal time period, ε represents the control parameter, which is used to control the speed of the rotor position deviation state, and the control parameter is the dynamic change, μ0 and μ1 are weight coefficients, and both μ0 and μ1 are greater than 0; It should be noted that the rotor position deviation has many effects on the motor, including reducing efficiency, increasing vibration and noise, causing overheating, affecting the control system, increasing mechanical load, electromagnetic interference and triggering protection mechanisms. Especially in permanent magnet synchronous motors, their drive control usually relies on rotor position feedback (such as encoders or Hall sensors). If the rotor position deviates, the motor control system cannot accurately obtain the rotor position, resulting in errors in the motor control signal. If it is determined to be a deviation state, the stator will generate an asymmetric electromagnetic field, which will cause the motor torque to fluctuate and the operating state to be uneven, resulting in current imbalance, which will directly affect the distribution of current components and the stability of electrical angle deviation; thus, the motor will not run normally, which will affect the adjustment of speed, torque and other parameters, and may even cause motor damage or system failure; In summary, by quantifying the synergistic effects of angle deviation, current anomaly, and sampling delay, we not only provide a mathematical framework for fault prediction, but also reduce risks during the acquisition process; The feedback adjustment module, upon receiving the risk assessment coefficient, compares and analyzes the risk assessment coefficient with a preset assessment threshold, and selects whether to execute the corresponding frequency-controlled control instruction according to the comparison result; Among them, the risk assessment coefficient is compared and analyzed with the preset assessment threshold: Calculate the ratio between the risk assessment coefficient and the risk assessment threshold, and send different frequency conversion adjustment signals according to the ratio; where the ratio = risk assessment coefficient / risk assessment threshold; When the ratio is less than 0.8, the motor operation process is judged to be in a low-risk environment and no operation is responded to; When 0.8≤ratio≤1.2, the motor operation process is judged to be in a medium-risk environment, and a type-one adjustment signal is issued to execute the first frequency conversion strategy; When the ratio is greater than 1.2, it is determined that the motor operation process is in a high-risk environment, and a second-class adjustment signal is issued to execute the second frequency conversion strategy; The first frequency conversion strategy algorithm model is: ; Where V new Indicates the corrected voltage, V base Indicates the current voltage, f new Indicates the corrected motor operating frequency, f base Indicates the current motor operating frequency, K V Indicates the voltage regulation coefficient; acc new Indicates the corrected adjustment speed, acc base Indicates the current adjustment speed; t new represents the time control period of the correction, t base Indicates the current control cycle; The algorithm model of the second frequency conversion strategy is: Get the current operating frequency. When it exceeds the preset frequency threshold, it is marked as a high frequency band. Otherwise, it is marked as a low frequency band and the voltage output is corrected proportionally. It should be noted that the preset frequency threshold is usually 30Hz. Under low frequency conditions: ; Under high frequency conditions: ; ; Where V new Indicates the corrected voltage, V base Indicates the current voltage, f new Indicates the corrected motor operating frequency, f base Indicates the current motor operating frequency, acc new Indicates the corrected adjustment speed, acc base Indicates the current adjustment speed; t new represents the time control period of the correction, t base Indicates the current control cycle; Specifically, this embodiment adopts: Frequency conversion adjustment: After receiving different frequency conversion adjustment signals, it will make adaptive adjustments according to the severity of the abnormality, and change the mechanical energy consumption of the motor through the frequency conversion controller, so that the motor is always in a normal state (the rotor position deviation is within the error range); in addition, in the actual operation process, in order to ensure the safety of the frequency conversion motor, the power size of its frequency conversion adjustment will be restricted, otherwise it will affect the service life of the frequency conversion motor. Therefore, the power adjustment amount of the motor is limited within the set range; In summary, after receiving the risk assessment coefficient, the motor frequency, voltage and time parameters are dynamically adjusted according to the risk ratio to achieve closed-loop control; Predictive optimization model, obtain the operating parameters of the target motor and construct a topological space, embed the risk assessment coefficient as a feature to form a graph structure, and obtain an updated model of the graph structure to obtain a state prediction model; under the triggering of the variable frequency control instruction, execute the model optimization strategy within the second time period; wherein the motor operating parameters include current, voltage, temperature, vibration, frequency, speed, torque, and power; Specifically, the state prediction model is based on a graph convolutional neural network structure, with the physical components of the motor (stator, rotor, bearing, winding, cooling system, etc.) as a node, and the motor operating parameters such as current, voltage, temperature, vibration, etc. as the dynamic attributes of the node; by embedding the risk assessment coefficient as a feature, the feature is integrated with the original feature of the node (for example, temperature, vibration amplitude), that is, it is directly used as part of the node feature to form a graph structure, and the data source is updated in real time to obtain a state prediction model; the specific implementation method belongs to common knowledge and will not be described in detail here; The steps to implement the model optimization strategy include: Obtain the duration and frequency of the deviation state within the corresponding time period to calculate the time complexity; The formula is: ; Where fz1 represents time complexity, con represents duration, fre represents occurrence frequency, α1 and α2 are weight correction coefficients, and α3 and α4 are both greater than 0; k represents the occurrence time of the deviation state, and k=1, 2...nd; where nd represents the total number of deviation states that occur; Obtain the number of responses to the variable frequency adjustment signal and the variable frequency adjustment ratio within the corresponding time period to calculate the action complexity; The formula is: ; Where, fz2 represents the complexity of the action, num represents the number of times the frequency conversion adjustment signal is responded to, pro represents the frequency conversion adjustment ratio of each response to the frequency conversion adjustment signal, σ1 and σ2 represent weight correction coefficients, and σ1 and σ2 are both greater than 0; Frequency conversion adjustment ratio: indicates the difference between the corrected motor operating frequency and the current motor operating frequency in the process of responding to the frequency conversion adjustment signal; Based on the time complexity and action complexity, a target optimization function min(fz1, fz2) is constructed, and the target optimization function is used as the connection function of the state prediction model. The number of nodes in the last hidden layer in the state prediction model is consistent with the number of parameters involved in the impedance formula of the target optimization function. The weights of the last layer of the state prediction model are initialized, and the gradient descent algorithm is used for learning. The hyperparameters are adjusted through the training set, and the performance of the model is evaluated through the validation set and the test set to retrain the entire network model to achieve model optimization. In the subsequent monitoring process, the optimized state prediction model is used to predict the state of the target motor. If it is under medium or high risk conditions, the frequency conversion adjustment signal is responded again to trigger the frequency conversion control instruction. In addition, the optimization form can be set to rolling optimization, which means that the state prediction model is updated in a limited time domain and optimized based on the latest data and information to adapt to the new environment or data. Example 2

[0018] The embodiment of the present invention provides an intelligent frequency conversion control device based on dynamic data acquisition; Figure 3 is a schematic diagram of a module of an intelligent frequency conversion control device according to an exemplary embodiment; Figure 3 ,include: The waveform monitoring management module obtains the electromagnetic parameters of the target motor. When the monitored electromagnetic parameters generate waveform signals, the acquisition adjustment plan is executed. The waveform signals are analyzed and processed in combination with machine learning algorithms to identify and classify acquisition adjustment types. The abnormality assessment module, under the condition of receiving the acquisition adjustment type, obtains the rotor position and the evaluation parameter set of the motor within the first time period, compares and analyzes the rotor position with the preset standard rotor position, obtains the time corresponding to the deviation state, combines the preprocessed evaluation parameter set, executes the abnormality analysis strategy, and generates the risk assessment coefficient; The feedback adjustment module, upon receiving the risk assessment coefficient, compares and analyzes the risk assessment coefficient with a preset assessment threshold, and selects whether to execute the corresponding frequency conversion control instruction according to the comparison result; A predictive optimization model is developed to obtain the target motor operating parameters and construct a topological space. The risk assessment coefficient is embedded as a feature to form a graph structure, and an updated model of the graph structure is obtained to obtain a state prediction model. When a variable frequency control instruction is triggered, the model optimization strategy is executed within a preset second time period. Example 3

[0019] The embodiment of the present invention provides an intelligent variable frequency control system based on dynamic data acquisition, comprising: A frequency converter, used for outputting a frequency conversion adjustment signal; At least two DC power supplies, used to supply power to the frequency converter; A variable frequency control device obtains the electromagnetic parameters of a target motor. When the monitored electromagnetic parameters generate a waveform signal, an acquisition adjustment plan is executed, the waveform signal is analyzed and processed in combination with a machine learning algorithm, and the acquisition adjustment type is identified and classified; under the condition of receiving the acquisition adjustment type, the rotor position and the evaluation parameter set of the motor are obtained within a first time period, the rotor position is compared and analyzed with a preset standard rotor position, the moment corresponding to the deviation state is obtained, and the abnormal analysis strategy is executed in combination with the pre-processed evaluation parameter set to generate a risk assessment coefficient; under the condition of receiving the risk assessment coefficient, the risk assessment coefficient is compared and analyzed with a preset evaluation threshold, and whether to execute the corresponding variable frequency control instruction is selected according to the comparison result; the operating parameters of the target motor are obtained and a topological space is constructed, the risk assessment coefficient is embedded as a feature to form a graph structure, and an updated model of the graph structure is obtained to obtain a state prediction model; under the triggering of the variable frequency control instruction, a model optimization strategy is executed within a preset second time period.

[0020] In the present application, the weight correction coefficient involved is determined by the coefficient of variation method, which is a method of weighting each indicator according to the degree of variation between the current value of each evaluation indicator and the target value; if the numerical difference of an indicator is large, it can clearly distinguish the evaluated objects, indicating that the indicator has rich discrimination information, and thus the indicator should be given a larger weight; conversely, if the numerical difference of each evaluated object on a certain indicator is small, then the ability of this indicator to distinguish the evaluation objects is weak, and thus the indicator should be given a smaller weight; this method directly uses the information contained in each indicator to obtain the weight of the indicator by calculation, and therefore is objective.

[0021] In the application, the several formulas involved are all calculated by removing the dimensions and taking their numerical values. The formula is a formula obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The formula is set by technical personnel in this field according to actual conditions.

[0022] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0023] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0024] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. An intelligent frequency conversion control method based on dynamic data acquisition, characterized in that: The method comprises: Obtain the electromagnetic parameters of the target motor. When the monitored electromagnetic parameters generate waveform signals, execute the acquisition adjustment plan, analyze and process the waveform signals in combination with machine learning algorithms, and identify and classify the acquisition adjustment types. Under the condition of receiving the acquisition adjustment type, the rotor position and the evaluation parameter set of the motor are obtained within the first time period, the rotor position is compared and analyzed with the preset standard rotor position, the time corresponding to the deviation state is obtained, and the abnormal analysis strategy is executed in combination with the preprocessed evaluation parameter set to generate a risk assessment coefficient; Upon receiving the risk assessment coefficient, the risk assessment coefficient is compared and analyzed with the preset assessment threshold, and whether to execute the corresponding frequency conversion control instruction is selected according to the comparison result; The target motor operating parameters are obtained and a topological space is constructed. The risk assessment coefficient is embedded as a feature to form a graph structure, and an updated model of the graph structure is obtained to obtain a state prediction model. When the variable frequency control instruction is triggered, the model optimization strategy is executed within a preset second time period.

2. The intelligent frequency conversion control method based on dynamic data acquisition according to claim 1 is characterized in that: The electromagnetic parameters include stator resistance, quadrature-axis inductance, direct-axis inductance, and permanent magnet flux; the waveform signal includes square wave, trapezoidal wave, triangle wave, and sine wave; the step of analyzing and processing the waveform characteristics in combination with a machine learning algorithm includes: Signal processing: filter the captured waveform signal to extract the characteristics of the waveform signal, including at least amplitude, frequency, phase, envelope, and harmonics; Abnormal identification: When a square wave or trapezoidal wave is identified, a local constant signal is injected; when a triangle wave or sine wave is identified, a global time-varying signal is injected; at the same time, a decision tree algorithm is used to train the extracted features and the injected signals; during the training process, a parameter matrix is ​​constructed based on electromagnetic parameters, and a weight factor is obtained based on the numerical change characteristics of the electromagnetic parameters, and a score is assigned to the weight factor, and the weight factor and the score are multiplied to obtain the recognition accuracy coefficient; the numerical change characteristics include mean value, kurtosis, kurtosis, quartile deviation and variance; Adjustment identification: In the real-time monitoring process, the direct and quadrature axis current equations are combined as constraints and mapped to electromagnetic parameters to obtain a corresponding set of training samples; the identification accuracy coefficient corresponding to each set of training samples is used as the sample label, and the classification adjustment model is obtained through training with the training samples to output the amplitude and frequency of the injection signal corresponding to the current waveform signal.

3. The intelligent frequency conversion control method based on dynamic data acquisition according to claim 1 is characterized in that: The loss function of the hidden layer of the classification adjustment model is as follows: ； Where L represents the loss function of the hidden layer of the classification adjustment model, α1 and α2 are weight loss coefficients, N represents the number of training samples in each group, N represents the number of samples in any group of training samples, Ca represents the parameter matrix, α0 represents the adjustment factor, Indicates the first The prediction accuracy coefficient of samples is Indicates the first The identification accuracy coefficient of samples is represents the recognition accuracy coefficient of the jth sample in a group, represents the constraint function, y i Indicates the first The constraint term of samples, y i Represents the constraint item of the jth sample in a group; in, ; In the formula, y represents the constraint condition, f dq (·) represents the nonlinear function under the d-axis or q-axis, wave represents the waveform signal, and R s Indicates stator resistance, L d Indicates the quadrature-axis inductance, L q represents the direct-axis inductance, Ψ f represents the permanent magnet flux, and ω represents the weight correction coefficient.

4. The intelligent frequency conversion control method based on dynamic data acquisition according to claim 1 is characterized in that: The evaluation parameter set includes a sampling delay rate, a current component, and a motor electrical angle deviation; The steps to determine the deviation state include: Obtain the difference between the rotor position and the preset standard rotor position, and set the gradient range. When the difference is within the preset gradient range, it is determined that there is no deviation, otherwise, it is determined to be a deviation state; At the same time, an evaluation parameter set is obtained, and a set parameter set is obtained from a database, including a sampling delay rate threshold, a direct-axis current threshold, a quadrature-axis current threshold, and an angle deviation threshold; After dimensionless processing and weighted calculation, the deviation factor is generated; The formula is: ； In the formula, fault represents the deviation factor, late represents the sampling delay rate, lates represents the sampling delay rate threshold, b0 represents the weight correction coefficient, which is used to strengthen the marginal risk of high delay, and its value is generally 2-3. Dev1 represents the current component of the direct axis, Devs1 represents the DC current threshold, Dev2 represents the current component of the quadrature axis, Devs2 represents the quadrature axis current threshold, Ang represents the motor electrical angle deviation, and Angs represents the angle deviation threshold. B1, B2, and B3 are all weight coefficients.

5. The intelligent frequency conversion control method based on dynamic data acquisition according to claim 1 is characterized in that: The steps to execute an anomaly analysis strategy include: Collect the moments judged as deviation states, randomly extract two time series, obtain abnormal time periods, sort the abnormal time periods in time series and mark them as T1, T2, ..., Tn; where the time series is a dynamic change, and n represents the number of times it is judged as a deviation state; divide the abnormal time period into equal parts and mark it as r, and when it has a remainder, it is r+1, where r is a positive integer and a dynamic value; combined with the deviation factor, calculate the risk assessment coefficient of the corresponding abnormal time period under r or r+1.

6. The intelligent frequency conversion control method based on dynamic data acquisition according to claim 1 is characterized in that: Compare and analyze the risk assessment coefficient with the preset assessment threshold: Calculate the ratio between the risk assessment coefficient and the risk assessment threshold, and send different frequency conversion adjustment signals according to the ratio; where the ratio = risk assessment coefficient / risk assessment threshold; When the ratio is less than 0.8, no response operation; When 0.8≤ratio≤1.2, a type 1 adjustment signal is issued and the first frequency conversion strategy is executed; When the ratio is greater than 1.2, a second type of adjustment signal is issued and the second frequency conversion strategy is executed.

7. The intelligent frequency conversion control method based on dynamic data acquisition according to claim 1 is characterized in that: The first frequency conversion strategy algorithm model is: ； Where V new Indicates the corrected voltage, V base Indicates the current voltage, f new Indicates the corrected motor operating frequency, f base Indicates the current motor operating frequency, K V Indicates the voltage regulation coefficient; acc new Indicates the corrected adjustment speed, acc base Indicates the current adjustment speed; t new represents the time control period of the correction, t base Indicates the current control cycle; Get the current operating frequency. When it exceeds the preset frequency threshold, it is marked as a high frequency band. On the contrary, it is marked as low frequency band and the voltage output is corrected proportionally; Under low frequency conditions: ; Under high frequency conditions: ; ； Where V new Indicates the corrected voltage, V base Indicates the current voltage, f new Indicates the corrected motor operating frequency, f base Indicates the current motor operating frequency, acc new Indicates the corrected adjustment speed, acc base Indicates the current adjustment speed; t new represents the time control period of the correction, t base Indicates the current control cycle.

8. The intelligent frequency conversion control method based on dynamic data acquisition according to claim 1 is characterized in that: The steps to implement the model optimization strategy include: Obtain the duration and frequency of the deviation state within the corresponding time period to calculate the time complexity; Obtain the number of responses to the variable frequency adjustment signal and the variable frequency adjustment ratio within the corresponding time period to calculate the action complexity; Based on the time complexity and action complexity, the target optimization function min(fz1, fz2) is constructed, and the target optimization function is used as the connection function of the state prediction model. The number of nodes in the last hidden layer in the state prediction model is consistent with the number of parameters involved in the impedance formula of the target optimization function. The weights of the last layer of the state prediction model are initialized, and the gradient descent algorithm is used for learning. The hyperparameters are adjusted through the training set, and the performance of the model is evaluated through the validation set and the test set to retrain the entire network model to achieve model optimization.

9. An intelligent frequency conversion control device based on dynamic data acquisition, characterized in that: include: The waveform monitoring management module obtains the electromagnetic parameters of the target motor. When the monitored electromagnetic parameters generate waveform signals, the acquisition adjustment plan is executed. The waveform signals are analyzed and processed in combination with machine learning algorithms to identify and classify acquisition adjustment types. The abnormality assessment module, under the condition of receiving the acquisition adjustment type, obtains the rotor position and the evaluation parameter set of the motor within the first time period, compares and analyzes the rotor position with the preset standard rotor position, obtains the time corresponding to the deviation state, combines the preprocessed evaluation parameter set, executes the abnormality analysis strategy, and generates the risk assessment coefficient; The feedback adjustment module, upon receiving the risk assessment coefficient, compares and analyzes the risk assessment coefficient with a preset assessment threshold, and selects whether to execute the corresponding frequency-controlled control instruction according to the comparison result; A predictive optimization model is developed to obtain the target motor operating parameters and construct a topological space. The risk assessment coefficient is embedded as a feature to form a graph structure, and an updated model of the graph structure is obtained to obtain a state prediction model. When a variable frequency control instruction is triggered, the model optimization strategy is executed within a preset second time period.

10. An intelligent frequency conversion control system based on dynamic data acquisition, characterized in that: The system comprises: A frequency converter, used for outputting a frequency conversion adjustment signal; At least two DC power supplies, used to supply power to the frequency converter; A variable frequency control device obtains the electromagnetic parameters of a target motor. When the monitored electromagnetic parameters generate a waveform signal, an acquisition adjustment plan is executed, the waveform signal is analyzed and processed in combination with a machine learning algorithm, and the acquisition adjustment type is identified and classified; under the condition of receiving the acquisition adjustment type, the rotor position and the evaluation parameter set of the motor are obtained within a first time period, the rotor position is compared and analyzed with a preset standard rotor position, the moment corresponding to the deviation state is obtained, and the abnormal analysis strategy is executed in combination with the pre-processed evaluation parameter set to generate a risk assessment coefficient; under the condition of receiving the risk assessment coefficient, the risk assessment coefficient is compared and analyzed with a preset evaluation threshold, and whether to execute the corresponding variable frequency control instruction is selected according to the comparison result; the operating parameters of the target motor are obtained and a topological space is constructed, the risk assessment coefficient is embedded as a feature to form a graph structure, and an updated model of the graph structure is obtained to obtain a state prediction model; under the triggering of the variable frequency control instruction, a model optimization strategy is executed within a preset second time period.

Images