A remote management system for frequency converters based on the Internet of Things

Abstract

The present invention relates to the field of inverter management technology, and solves the inherent defects of the existing technology that it cannot adapt to the dynamic frequency modulation of the inverter, is prone to the disadvantage of local minimum values, and has low processing efficiency. In particular, it relates to an inverter remote management system based on the Internet of Things, which includes: a preprocessing module for collecting original data of the inverter, calculating an interference threshold for removing interference data from the original data, and obtaining preprocessed data based on the interference threshold. The present invention can greatly improve the selection efficiency of weights and thresholds by utilizing group optimization, and can improve the classification accuracy of the inverter operating status. By connecting the competition layer of unsupervised learning and the neural network of supervised learning in series, a new neural network model is formed, which avoids the disadvantage of local minimum values ​​in the neural network, greatly improves the stability of the neural network, and improves learning efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of frequency converter management, and in particular to a frequency converter remote management system based on the Internet of Things. Background Art

[0002] A frequency converter is a power electronic device that controls the speed of a motor by adjusting the power supply frequency and voltage. The IoT-based frequency converter remote management system collects operating parameters such as voltage, current, and temperature in real time, and combines edge computing with cloud-based analysis to achieve intelligent monitoring of device status, energy efficiency optimization, and abnormality warnings. Existing technologies usually use traditional fixed filters to implement frequency converter data collection, which cannot adapt to the inherent defects of frequency converter dynamic frequency modulation. In addition, existing technologies use unsupervised neural networks to analyze frequency converter data, which is prone to local minima. At the same time, the amount of learning data is relatively large, and the processing speed is low. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention provides a frequency converter remote management system based on the Internet of Things, which solves the inherent defects of the existing technology that it cannot adapt to the dynamic frequency modulation of the frequency converter and is prone to the disadvantage of local minimum values, and at the same time solves the technical problem of low processing efficiency. It achieves the goal of accurately identifying burst interference pulses through adaptive thresholds and avoiding the disadvantage of local minimum values. At the same time, it can also reduce the amount of learning data, greatly improve the stability of the neural network, and improve the learning efficiency.

[0004] To solve the above technical problems, the present invention provides the following technical solutions: a remote management system for inverters based on the Internet of Things, the system comprising:

[0005] The preprocessing module is used to collect the original data Z(y) of the frequency converter, calculate the interference threshold Y for removing interference data from the original data Z(y), and obtain the preprocessing data C based on the interference threshold Y. y ;

[0006] Alignment module, used to process the data C y Calculate the optimization coefficient γ after iteration yh , according to the optimization coefficient γ yh Calculate the alignment tensor D i ;

[0007] The mixed feature module is used to align the tensor D i Calculate the hybrid characteristic H used to monitor the dynamic characteristics of the inverter t ;

[0008] The victory meta-module is used to calculate the value of the mixed feature H t Calculate the victory element S after iteration L ;

[0009] Status monitoring module, used to monitor the status of the L Calculate dynamic filter value Do s , based on dynamic filter value Do s Obtain the optimized inverter status monitoring model;

[0010] Remote management module for the hybrid feature H t The inverter status detection result is input into the inverter status monitoring model, and the inverter status detection result is sent to the remote management center.

[0011] Furthermore, the preprocessing module includes:

[0012] The carrier frequency filtering module is used to calculate the carrier frequency filtering value P for filtering based on the original data Z(y) zp ;

[0013] Interference threshold module, used to establish a sliding window H with a window length of C and calculate the carrier frequency filter value P within the window zp Interference threshold Y;

[0014] Elimination module, used to filter the carrier frequency value P according to the interference threshold Y zp Eliminate the interference data in

[0015] If P zp >Y, then the carrier frequency filtering value P zp To interfere with the data and remove it;

[0016] If P zp ≤Y, then the carrier frequency filtering value P zp It is normal data and is retained;

[0017] Filling module, used to obtain adjacent data L adjacent to the removed interference data k+1 and L k-1 , calculate the filling data T used to supplement the interference data b ;

[0018] Merge module, used to combine normal data and padding data T b Merge into preprocessed data C in chronological order y .

[0019] Furthermore, the alignment module includes:

[0020] Interconnection module, used to pre-process data C y Arbitrarily select two different preprocessed data C a and C b , and calculate the preprocessed data C y The mutual information L(Ca ,C b ;▽C a ,▽C b );

[0021] Balance module, used for mutual information L(C a ,C b ;▽C a ,▽C b ) Calculate the balance coefficient γ p ;

[0022] Gradient module, used to calculate the equilibrium coefficient γ p Calculate the gradient value T f ;

[0023] Iteration module, used to calculate the gradient value T f Calculate the iteration coefficient γ of the p+1th generation p+1 ;

[0024] Iteration threshold module, used to calculate the iteration threshold D based on the quantity A a .

[0025] Optimization coefficient module, used to calculate the optimal coefficient according to the iteration threshold D a Determine whether the iteration is finished;

[0026] If p+1<D a , then the optimization coefficient γ is repeatedly optimized yh Calculation of

[0027] If p+1≥D a , then the iteration ends and the final optimization coefficient γ is obtained yh ;

[0028] Optimization alignment module, used to optimize the coefficient γ yh Calculate the alignment tensor D i .

[0029] Furthermore, the hybrid feature module includes:

[0030] Scalar value tensor module for aligning tensors D i The scalar value tensor B is obtained by local normalization of the sliding window h ;

[0031] The time domain feature module is used to calculate the time domain feature according to the scale value tensor B. h The time domain feature S is obtained by sliding differential method y ;

[0032] Band energy module, used to calculate the energy of the frequency band according to the scalar value tensor B h Calculate the frequency band energy characteristic P used to evaluate the frequency band energy fluctuation of the inverter signal d ;

[0033] The cross-metric module is used to select a pair of sensors with physical association (a m ,b n ), the sensor pair (a m ,b n ) of the scalar-valued tensor B m and B n And calculate the cross metric J c ;

[0034] The splicing module is used to calculate the time domain feature S y , frequency band energy characteristics P d and cross metric J c Calculate the mixed feature H t .

[0035] Furthermore, the victory meta module includes:

[0036] Preparation module, used to combine multiple mixed features H t As the input of the input layer, and define the initial learning rate Cs x , weight w O and neighborhood radius R l ;

[0037] Euclidean distance module, used to calculate multiple mixed features H t The Euclidean distance Jl o ;

[0038] Competition module, used to calculate the o Calculate the victory element S of the competition layer L ;

[0039] Update module, used to update the learning rate Cs x and neighborhood radius R l Calculate the updated learning rate Cs x+1 and update the neighborhood radius R l+1 ;

[0040] Control module, used to update the learning rate Cs x+1 Control Victory Element S L iterations;

[0041] If Cs x+1 ≤0.001, the iteration ends and the current victory element S is output L ;

[0042] If Cs x+1 >0.001, then recalculate the victory yuan S L .

[0043] Furthermore, the status monitoring module includes:

[0044] Implicit module, used to convert multiple victory elements S L As input, the implicit value Y is calculated through the implicit function of the hidden layer h ;

[0045] Output module, used to calculate the implicit value Y h Calculate the output value Sc of the output layer i ;

[0046] The mean square error module is used to define the expected value Q according to the model w , calculate the output value Sc i and the expected value Q w The mean square error value W f ;

[0047] Optimize model module for multiple basic weights w C , intercept value Y J and the intercept threshold y z Calculate the optimized weight Q y , iterative intercept YC u and iterative cutoff threshold J d , based on the optimized weight Q y , iterative intercept YC u and iterative cutoff threshold J d Construct optimization model YH;

[0048] Dynamic screening module, used to filter the mean square error value W f Calculate dynamic filter value Do s ;

[0049] Model building module, used to dynamically filter the value Do s Complete the establishment of the inverter status monitoring model;

[0050] If W f <Do s , then the end and the inverter status monitoring model is obtained;

[0051] If W f ≥Do s , then re-train the model.

[0052] Furthermore, the expression of the optimization model YH is:

[0053]

[0054] Among them, Yh q represents the qth optimization variable value, and p represents the victory element S L The number of

[0055] Furthermore, the optimization model module (54) includes:

[0056] Initialization module, used to convert the basic weight w C , intercept value Y J and the intercept threshold y z Converted into multiple particles Li through vector encoding z , based on particle Li z The encoded value defines the initial position C of each particle w , speed v c , individual optimal position G z and the group optimal position Q z ;

[0057] Fitness module, used to construct the fitness Sy used to guide particles to migrate to a better area d (Z);

[0058] Update the speed module to adjust the speed according to the speed v c Calculate the updated velocity value v of the cth particle of the Dth generation c (D+1);

[0059] Update position module, used to update the speed value v c (D+1) Calculate the updated position value Gw x (D+1);

[0060] The optimal position module is used to update the position value Gw x (D+1) calculate the optimal position value Zy a ;

[0061] The judgment module is used to obtain the current number of iterations U and determine whether the iteration should be stopped according to the current number of iterations;

[0062] If U < 1000, continue to iterate;

[0063] If U≥1000, stop the iteration and set the optimal position value Zy at this time a Perform reverse encoding to obtain the optimized weight Q y , iterative intercept YC u and iterative cutoff threshold J d .

[0064] Furthermore, the optimal position value Zy a The calculation formula is:

[0065]

[0066] Among them, Zy a Indicates the ath optimal position value, Sy d[Gw x (D+1)] represents the fitness Sy d The independent variable of (Z) is the updated position value Gw x (D+1), Sy d (Q z ) represents the fitness Sy d The independent variable of (Z) is the group optimal position Q z .

[0067] By means of the above technical solution, the present invention provides a remote management system for inverters based on the Internet of Things, which has at least the following beneficial effects:

[0068] 1. The present invention solves the noise interference problem in traditional inverter signal acquisition in the data preprocessing stage. Through PWM carrier frequency adaptive filtering technology, combined with real-time tracking of the actual switching frequency of the inverter, the digital filter parameters are dynamically adjusted, and the switching noise band centered on the PWM carrier frequency is accurately locked and eliminated. This solves the inherent defect of traditional fixed filters that cannot adapt to the dynamic frequency modulation of the inverter. The invention innovatively introduces a differential pulse detection algorithm, calculates the differential characteristics of the current signal in real time, and combines adaptive thresholds to accurately identify sudden interference pulses caused by grid transients or power device switching, and eliminates and fills sudden interference pulses, thereby improving the fluency and integrity of the data.

[0069] 2. The present invention can enhance multiple features of the inverter operation data by acquiring hybrid features, including time domain, frequency domain and cross-features. Through these features, the analysis and classification of the inverter operation status can be completed more accurately, thereby improving the accuracy of inverter analysis and management.

[0070] 3. The present invention forms a new neural network model by connecting a competitive layer with unsupervised learning and a neural network with supervised learning in series, which avoids the shortcomings of local minima in neural networks in the existing technology. At the same time, it can also reduce the amount of learning data, so that the new neural network can successfully learn and classify and identify data without a huge amount of data, greatly improving the stability of the neural network and improving learning efficiency.

[0071] 4. The present invention can greatly improve the selection efficiency of weights and thresholds by utilizing the method of group optimization, and has smaller test errors and better nonlinear fitting capabilities, which can improve the classification accuracy of the inverter operating status. By connecting the competition layer with unsupervised learning and the neural network with supervised learning in series, a new neural network model is constructed, which avoids the shortcomings of local minima in neural networks in the existing technology. At the same time, it can also reduce the amount of learning data, so that the new neural network can successfully learn and classify and identify data without a huge amount of data, which greatly improves the stability of the neural network and improves learning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0073] Figure 1 This is a structural block diagram of a frequency converter remote management system based on the Internet of Things of the present invention;

[0074] Figure 2 This is a structural block diagram of the preprocessing module of the present invention;

[0075] Figure 3 This is a structural block diagram of the alignment module of the present invention;

[0076] Figure 4 It is a structural block diagram of the hybrid feature module of the present invention;

[0077] Figure 5 This is a structural block diagram of the victory unit module of the present invention;

[0078] Figure 6 This is a structural block diagram of the status monitoring module of the present invention;

[0079] Figure 7 This is a structural diagram of the optimization model module of the present invention.

[0080] In the figure: 1. Preprocessing module; 11. Carrier frequency filtering module; 12. Interference threshold module; 13. Removal module; 14. Filling module; 15. Merging module; 2. Alignment module; 21. Interaction module; 22. Balance module; 23. Gradient module; 24. Iteration module; 25. Iteration threshold module; 26. Optimization coefficient module; 27. Optimization alignment module; 3. Hybrid feature module; 31. Scaling tensor module; 32. Time domain feature module; 33. Band energy module; 34. Cross-metric module; 35. Splicing module; 4. Victory element module; 41. Preparation module; 42. Euclidean distance module; 43. Competition module; 44. Update module; 45. Control module; 5. State monitoring module; 51. Implicit module; 52. Output module; 53. Mean square error module; 54. Optimization model module; 541. Initialization module; 542. Fitness module; 543. Update speed module; 544. Update position module; 545. Optimal position module; 546. Judgment module; 55. Dynamic screening module; 56. Model building module; 6. Remote management module. DETAILED DESCRIPTION

[0081] To make the above-mentioned objectives, features, and advantages of the present invention more clearly understood, the present invention is further described below in detail with reference to the accompanying drawings and specific embodiments. This will enable a full understanding of how this application uses technical means to solve technical problems and achieve technical effects, and to implement the invention accordingly.

[0082] Since the existing technology cannot adapt to the inherent defects of the dynamic frequency modulation of the inverter, and is prone to the disadvantage of local minimum value, and has the technical problem of low processing efficiency, this embodiment proposes a remote management system for inverters based on the Internet of Things, such as Figure 1-7 As shown, the system can accurately identify burst interference pulses through adaptive thresholds and avoid the disadvantages of local minimum values. At the same time, it can also reduce the amount of learning data, greatly improve the stability of the neural network, and improve learning efficiency. The system includes:

[0083] The preprocessing module 1 is used to collect the original data Z(y) of the frequency converter, calculate the interference threshold Y for removing interference data from the original data Z(y), and obtain the preprocessing data C based on the interference threshold Y. y ; Preprocessing module 1 includes:

[0084] The carrier frequency filtering module 11 is used to calculate the carrier frequency filtering value P for filtering based on the original data Z(y) zp , the calculation formula is:

[0085]

[0086] Among them, a0 and b0 represent the adaptive filter coefficients, f PWM Indicates the PWM real-time switching frequency, T s Indicates the sampling period;

[0087] Interference threshold module 12 is used to establish a sliding window H with a window length of C and calculate the carrier frequency filtering value P within the window zp The interference threshold Y is calculated as follows:

[0088] Y=L no ·σ

[0089] Among them, L no represents the empirical coefficient, σ represents the sliding estimate of the current standard deviation; in practical applications, the empirical coefficient L no It is usually set to -3, and the current standard deviation sliding estimate σ can be obtained through the circuit in the acquisition circuit.

[0090] Elimination module 13, used for filtering the carrier frequency value P according to the interference threshold Y zp Eliminate the interference data in

[0091] If P zp>Y, then the carrier frequency filtering value P zp To interfere with the data and remove it;

[0092] If P zp ≤Y, then the carrier frequency filtering value P zp It is normal data and is retained;

[0093] Filling module 14 is used to obtain adjacent data L adjacent to the removed interference data k+1 and L k-1 , calculate the filling data T used to supplement the interference data b , the calculation formula is:

[0094]

[0095] Among them, T b Indicates the bth filling data;

[0096] The merging module 15 is used to combine normal data and padding data T b Merge into preprocessed data C in chronological order y . The present invention has achieved three key technological innovations in the data preprocessing stage, solving the problem of noise interference in traditional inverter signal acquisition. First, through PWM carrier frequency adaptive filtering technology, we track the actual switching frequency of the inverter in real time, dynamically adjust the digital filter parameters, accurately lock and eliminate the switching noise band centered on the PWM carrier frequency, and solve the inherent defect that the traditional fixed filter cannot adapt to the dynamic frequency modulation of the inverter. Secondly, the differential pulse detection algorithm is innovatively introduced to calculate the differential characteristics of the current signal in real time, combined with the adaptive threshold to accurately identify the sudden interference pulses caused by grid transients or power device switching, and mark and eliminate them. It has been verified by actual measurements that this preprocessing scheme improves the system signal-to-noise ratio from 18dB of the traditional scheme to 49dB, with a relative improvement of 171%.

[0097] Alignment module 2 is used to process the data C y Calculate the optimization coefficient γ after iteration yh , according to the optimization coefficient γ yh Calculate the alignment tensor D i ; Preprocessed data C y Contains a variety of data features. To further analyze the operating status of the inverter, it is necessary to obtain these data features in order to perform status analysis more accurately. To solve this problem, the detailed implementation steps are as follows:

[0098] Interconnection module 21, used for pre-processing data C y Arbitrarily select two different preprocessed data C a and C b , and calculate the preprocessed data Cy Mutual information The calculation formula is:

[0099]

[0100] in, Represents the preprocessed data C y The information entropy of A and B represents the preprocessed data Jz za and Jz zb The number of x a and y b Represents the preprocessed data C a and C b Corresponding data values; mutual information refers to the amount of information one variable contains about another variable. Information entropy is a common method for calculating uncertain information and will not be described in detail here.

[0101] Balance module 22, for Calculate the balance coefficient γ p , the calculation formula is:

[0102]

[0103] Among them, γ p represents the pth balance coefficient, and γ p ∈[0,1];

[0104] Gradient module 23, for calculating the equilibrium coefficient γ p Calculate the gradient value T f , the calculation formula is:

[0105]

[0106] Among them, T f represents the fth gradient value;

[0107] Iteration module 24 is used to calculate the gradient value T f Calculate the iteration coefficient γ of the p+1th generation p+1 , the calculation formula is:

[0108] γ p+1 =γ p +η·T f |γ p

[0109] Where η represents the learning rate, T f |γ p Represents the gradient value T f In the balance coefficient γ p The value of

[0110] Iteration threshold module 25, used to calculate the iteration threshold D according to the quantity A a , the calculation formula is:

[0111] D a =θ·A

[0112] Among them, θ represents a random coefficient; the random coefficient θ can be randomly generated by a random generator. The random generator is a common method for obtaining random numbers and will not be described in detail here.

[0113] The optimization coefficient module 26 is used to optimize the coefficient according to the iteration threshold D a Determine whether the iteration is finished;

[0114] If p+1<D a , then return to the interconnection module 21 to repeat the optimization coefficient γ yh Calculation of

[0115] If p+1≥D a , then the iteration ends and the final optimization coefficient γ is obtained yh ;

[0116] The optimization alignment module 27 is used to optimize the coefficient γ yh Calculate the alignment tensor D i , the calculation formula is:

[0117]

[0118] Among them, D i Represents the tth aligned tensor, by preprocessing the data C y The preprocessing can be done based on the preprocessed data C y The aligned tensor D i , aligned tensor D i It can more accurately reflect the working status of the inverter, thereby helping subsequent steps to analyze and manage the inverter status more accurately and quickly, and improve the efficiency and accuracy of data processing.

[0119] Mixed feature module 3 is used to align the tensor D i Calculate the hybrid characteristic H used to monitor the dynamic characteristics of the inverter t ; Align tensor D i After optimizing the accuracy and efficiency of the data, it is also necessary to align the tensor D i The features are decomposed and mixed features H are formed t In order to solve this problem, the specific methods adopted are as follows:

[0120] The scalar value tensor module 31 is used to align the tensor D i The scalar value tensor B is obtained by local normalization of the sliding window h; Local normalization is a method of normalizing a tensor in a local area through a sliding window, which can enhance local contrast or reduce the inconsistency of local features. Local normalization is a common tensor processing method and will not be described in detail here.

[0121] The time domain feature module 32 is used to calculate the time domain feature according to the scale value tensor B. h The time domain feature S is obtained by sliding differential method y ; The sliding differentiation method is often used in time domain feature extraction to smooth signals, enhance local contrast or normalize features. When used in conjunction with the local normalization method, the differentiation results can be further processed to enhance features. The sliding differentiation method is a common tensor processing method and will not be described in detail here.

[0122] The frequency band energy module 33 is used to calculate the frequency band energy according to the scale value tensor B. h Calculate the frequency band energy characteristic P used to evaluate the frequency band energy fluctuation of the inverter signal d , the calculation formula is:

[0123]

[0124] Where j represents the imaginary unit, f represents the frequency component, and T represents the alignment tensor D i The length of the corresponding alignment time period, t represents the time corresponding to the frequency f;

[0125] The cross-metric module 34 is used to select a pair of sensors (a) having physical association among the sensors of the frequency converter collecting the raw data Z(y). m ,b n ), the sensor pair (a m ,b n ) of the scalar-valued tensor B m and B n , and calculate the cross metric J c , the calculation formula is:

[0126]

[0127] Among them, J c represents the c-th cross metric, Indicates the calculation of the scalar value tensor B m and B n The outer product of

[0128] The splicing module 35 is used to y , frequency band energy characteristics P d and cross metric J c Calculate the mixed feature H t , the calculation formula is:

[0129] Ht =[S y ||P d ||J c ]

[0130] Among them, H t represents the t-th mixed feature, S y ||P d ||J c Represents the time domain feature S y , frequency band energy characteristics P d and cross metric J c splicing, by mixing features H t The acquisition of multiple features of the inverter operation data, including time domain, frequency domain and cross-features, can improve the analysis and classification of the inverter operation status more accurately, thereby improving the accuracy of inverter analysis and management.

[0131] Victory element module 4 is used to calculate the value of the mixed feature H t Calculate the victory element S after iteration L Since the neural network in the prior art is prone to falling into local minima and the network performance is unstable, which affects the normal monitoring and analysis, in order to solve this problem, this embodiment proposes a more detailed implementation method as follows:

[0132] Preparation module 41, for combining multiple mixed features H t As the input of the input layer, and define the initial learning rate Cs x , weight w O and neighborhood radius R l ; Here we first build the input layer of the model and define the key model parameters based on practical experience.

[0133] Euclidean distance module 42, used to calculate multiple mixed features H t The Euclidean distance Jl o , the calculation formula is:

[0134] Jl o =||H t -w C ||

[0135]

[0136] w C =w o +K h ·Cs x w o

[0137] Among them, Jl o represents the oth Euclidean distance, K hrepresents the domain function, w C represents the basic weight;

[0138] Competition module 43, for calculating the distance J1 according to the Euclidean distance o Calculate the victory element S of the competition layer L , the expression is:

[0139] S L =argmin(Jl o )

[0140] Among them, S L Indicates the Lth victory unit;

[0141] Update module 44 is used to update the learning rate Cs x and neighborhood radius R l Calculate the updated learning rate Cs x+1 and update the neighborhood radius R l+1 , the calculation formula is:

[0142]

[0143] Among them, INT represents the rounding function, U represents the current number of learning times, and l represents the neighborhood radius R of the lth generation l ;

[0144] Control module 45, used to update the learning rate Cs x+1 Control Victory Element S L iterations;

[0145] If Cs x+1 ≤0.001, the iteration ends and the current victory element S is output L ;

[0146] If Cs x+1 >0.001, then return to the preparation module 41 and recalculate the victory element S L Since the learning rate is a continuously decaying quantity, after multiple iterations, the learning rate is used to determine whether the iteration has ended. By connecting a competitive layer with unsupervised learning and a neural network with supervised learning in series, a new neural network model is formed. This avoids the shortcomings of local minima in neural networks in existing technologies, while also reducing the amount of learning data. This allows the new neural network to successfully learn and classify data without requiring a large amount of data, greatly improving the stability of the neural network and improving learning efficiency.

[0147] Status monitoring module 5, used to monitor the status of the victory element S L Calculate dynamic filter value Do s , based on dynamic filter value Do sGet the optimized inverter status monitoring model; Based on the previous step, it is necessary to convert the victory element S L For further processing, the detailed implementation steps are as follows:

[0148] Implicit module 51, used to convert multiple victory elements S L As input, the implicit value Y is calculated through the implicit function of the hidden layer h , the calculation formula is:

[0149]

[0150] Among them, P represents victory yuan S L The number of C represents the basic weight, Y J Represents the intercept value; the intercept value can be obtained by implicit value Y h The vertical intercept of is obtained, and x represents the independent variable in the function g(x).

[0151] Output module 52, used to output the implicit value Y h Calculate the output value Sc of the output layer i , the calculation formula is:

[0152]

[0153] Where N represents the implicit value Y h The number of y z Indicates the intercept threshold; intercept threshold y z It is obtained by peak detection method, which is a commonly used method to obtain the intercept threshold y z The method is not described here.

[0154] The mean square error module 53 is used to define the expected value Q according to the model w , calculate the output value Sc i and the expected value Q w The mean square error value W f , the calculation formula is:

[0155]

[0156] Among them, W f represents the fth mean square error value; expected value Q w It can be set as the data under the normal state of the inverter, can be set according to the experience of the staff, or can be set according to the average of the normal values ​​of the inverter's historical data.

[0157] The optimization model module 54 is used to optimize the model according to the multiple basic weights w C , intercept value Y J and the intercept threshold y zCalculate the optimized weight Q y , iterative intercept YC u and iterative cutoff threshold J d , based on the optimized weight Q y , iterative intercept YC u and iterative cutoff threshold J d Construct the optimization model YH, the expression is:

[0158]

[0159] Among them, Yh q represents the qth optimization variable value, and p represents the victory element S L the number of

[0160] Initialization module 541 is used to set the basic weight w C , intercept value Y J and the intercept threshold y z Converted into multiple particles Li through vector encoding z , based on particle Li z The encoded value defines the initial position C of each particle w , speed v c , individual optimal position G z and the group optimal position Q z ;

[0161] Fitness module 542, used to construct the fitness Sy for guiding particles to migrate to a better area d (Z), the expression is:

[0162]

[0163] Among them, Li a He Li b They represent the actual position of the particle and the expected position of the particle in the particle swarm, respectively, where z≥a,b, R represents the logarithm of the two particle positions, and Z represents the fitness Sy d (Z) independent variable;

[0164] Update speed module 543, for updating speed v c Calculate the updated velocity value v of the cth particle of the Dth generation c (D+1), calculated as:

[0165] v c (D+1)=α·v c +β1δ1(G z -C w )+β2δ2(Q z -C w )

[0166] Among them, α represents the inertia weight, β1 and β2 represent the acceleration coefficients, δ1 and δ2 represent random numbers, and δ1, δ2∈[0,1];

[0167] The position update module 544 is used to update the speed value v c (D+1) Calculate the updated position value Gw x (D+1), calculated as:

[0168] Gw x (D+1)=C w +v c (D+1)

[0169] Among them, Gw x (D+1) represents the xth updated position value;

[0170] The optimal position module 545 is used to update the position value Gw according to x (D+1) calculate the optimal position value Zy a , the calculation formula is:

[0171]

[0172] Among them, Zy a Indicates the ath optimal position value, Sy d [Gw x (D+1)] represents the fitness Sy d The independent variable of (Z) is the updated position value Gw x (D+1), Sy d (Q z ) represents the fitness Sy d The independent variable of (Z) is the group optimal position Q z ;

[0173] A judgment module 546 is used to obtain the current number of iterations U and determine whether the iteration should be stopped according to the current number of iterations;

[0174] If U<1000, return to the fitness module 542 to continue iteration;

[0175] If U≥1000, stop the iteration and set the optimal position value Zy at this time a Perform reverse encoding to obtain the optimized weight Q y , iterative intercept YC u and iterative cutoff threshold J d The optimal position value Zy in the particle swarm algorithm is often used in reverse coding. a Reversely obtain the optimized weight Q y , iterative intercept YC u and iterative cutoff threshold J dThe method is not described here.

[0176] Dynamic screening module 55, for filtering the f Calculate dynamic filter value Do s , the calculation formula is:

[0177]

[0178] Among them, μ and θ represent weight coefficients, Represents the mean square error value W f The mean of Wf Represents the mean square error value W f The standard deviation of

[0179] Model building module 56, used to dynamically filter the value Do s Complete the establishment of the inverter status monitoring model;

[0180] If W f <Do s , then the end and the inverter status monitoring model is obtained;

[0181] If W f ≥Do s , then return to module 53 to retrain the model, and solve the problem of selecting connection weights and thresholds of neural networks in the prior art by utilizing the group optimization method. The prior art sets them according to experimental experience, and then adjusts the connection weights and thresholds during the training process. This method requires repeated adjustments to the connection weights and thresholds, which seriously affects the learning efficiency of the neural network. The present invention is based on a group optimization method, which can greatly improve the selection efficiency of weights and thresholds, and has smaller test errors and better nonlinear fitting capabilities, and can improve the classification accuracy of the inverter operating status. By connecting a competitive layer with unsupervised learning and a neural network with supervised learning in series, a new neural network model is formed, which avoids the shortcomings of local minima in neural networks in the prior art, and can also reduce the amount of learning data, so that the new neural network does not need a huge amount of data to successfully learn and classify and identify data, greatly improving the stability of the neural network and improving learning efficiency.

[0182] Remote management module 6, used to set the hybrid characteristic H tInput into the inverter status monitoring model yields inverter status detection results, including normal operation, abnormalities, and shutdowns. These results are then sent to the remote management center. The remote management center uses these detection results to remotely manage the inverter, triggering alerts that are automatically sent to the operator's mobile phone or industrial control system. This shortens fault response time, predicts component aging, enables proactive replacement of spare parts, avoids unexpected downtime, and extends equipment life. The remote management center can modify inverter parameters remotely without requiring on-site operation.

[0183] Those skilled in the art will appreciate that all or part of the steps in the above-described embodiment methods can be accomplished by instructing related hardware through a program. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0184] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to in detail. For the above embodiments, since they are basically similar to the method embodiments, the description is relatively simple. For relevant parts, please refer to the partial description of the method embodiments.

[0185] The above embodiments provide a detailed introduction to the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A remote management system for frequency converters based on the Internet of Things, characterized in that: The system includes: The preprocessing module (1) is used to collect the original data Z(y) of the frequency converter, calculate the interference threshold Y for removing interference data from the original data Z(y), and obtain the preprocessing data C based on the interference threshold Y. y ; Alignment module (2) is used to align the preprocessed data C y Calculate the optimization coefficient γ after iteration yh , according to the optimization coefficient γ yh Calculate the alignment tensor D i ; The hybrid feature module (3) is used to i Calculate the hybrid characteristic H used to monitor the dynamic characteristics of the inverter t ; Victory element module (4), used to calculate the value of the mixed feature H t Calculate the victory element S after iteration L ; Status monitoring module (5), used for L Calculate dynamic filter value Do s , based on dynamic filter value Do s Obtain the optimized inverter status monitoring model; Remote management module (6) is used to set the hybrid characteristic H t Input into the inverter status monitoring model to obtain the inverter status detection result, and send the inverter status detection result to the remote management center; The pre-processing module (1) comprises: The carrier frequency filtering module (11) is used to calculate the carrier frequency filtering value P for filtering based on the original data Z(y) zp ; Interference threshold module (12), used to establish a sliding window H with a window length of C, and calculate the carrier frequency filtering value P inside the window zp Interference threshold Y; Elimination module (13), used for filtering the carrier frequency value P according to the interference threshold Y zp Eliminate the interference data in If P zp >Y, then the carrier frequency filtering value P zp To interfere with the data and remove it; If P zp ≤Y, then the carrier frequency filtering value P zp It is normal data and is retained; Filling module (14), used to obtain adjacent data L adjacent to the removed interference data k+1 and L k-1 , calculate the filling data T used to supplement the interference data b ; Merging module (15) is used to merge normal data and padding data T b Merge into preprocessed data C in chronological order y .

2. The inverter remote management system based on the Internet of Things according to claim 1, characterized in that: The alignment module (2) comprises: Interconnection module (21) is used to pre-process data C y Arbitrarily select two different preprocessed data C a and C b , and calculate the preprocessed data C y The mutual information L(C a ,C b ;▽C a ,▽C b ); The balancing module (22) is used to balance the mutual information L(C a ,C b ;▽C a ,▽C b ) Calculate the balance coefficient γ p ; Gradient module (23) is used to adjust the equilibrium coefficient γ p Calculate the gradient value T f ; Iteration module (24) is used to calculate the gradient value T f Calculate the iteration coefficient γ of the p+1th generation p+1 ; Iteration threshold module (25), used to calculate the iteration threshold D according to the quantity A a ; The optimization coefficient module (26) is used to optimize the coefficient according to the iteration threshold D a Determine whether the iteration is finished; If p+1<D a , then the optimization coefficient γ is repeatedly optimized yh Calculation of If p+1≥D a , then the iteration ends and the final optimization coefficient γ is obtained yh ; The optimization alignment module (27) is used to optimize the coefficient γ yh Calculate the alignment tensor D i .

3. The inverter remote management system based on Internet of Things according to claim 1, characterized in that: The hybrid feature module (3) comprises: The scalar value tensor module (31) is used to align the tensor D i The scalar value tensor B is obtained by local normalization of the sliding window h ; The time domain feature module (32) is used to generate the time domain feature according to the scale value tensor B h The time domain feature S is obtained by sliding differential method y ; Band energy module (33), used to calculate the energy of the frequency band according to the scale value tensor B h Calculate the frequency band energy characteristic P used to evaluate the frequency band energy fluctuation of the inverter signal d ; A cross-metric module (34) is used to select a sensor pair (a) having a physical association from among sensors of the frequency converter that collects raw data Z(y). m ,b n ), the sensor pair (a m ,b n ) of the scalar-valued tensor B m and B n And calculate the cross metric J c ; The splicing module (35) is used to generate the splicing image according to the time domain feature S y , frequency band energy characteristics P d and cross metric J c Calculate the mixed feature H t .

4. The inverter remote management system based on Internet of Things according to claim 1, characterized in that: The victory unit module (4) includes: Preparation module (41), used to combine multiple mixed features H t As the input of the input layer, and define the initial learning rate Cs x , weight w O and neighborhood radius R l ; Euclidean distance module (42), used to calculate multiple mixed features H t The Euclidean distance Jl o ; Competition module (43), for the Euclidean distance Jl o Calculate the victory element S of the competition layer L ; Update module (44) is used to update the x and neighborhood radius R l Calculate the updated learning rate Cs x+1 and update the neighborhood radius R l+1 ; Control module (45), for updating learning rate Cs according to x+1 Control Victory Element S L iterations; If Cs x+1 ≤0.001, the iteration ends and the current victory element S is output L ; If Cs x+1 >0.001, then recalculate the victory yuan S L .

5. The inverter remote management system based on Internet of Things according to claim 1, characterized in that: The state monitoring module (5) comprises: Implicit module (51) is used to convert multiple victory elements S L As input, the implicit value Y is calculated through the implicit function of the hidden layer h ; Output module (52), used to output the implicit value Y h Calculate the output value Sc of the output layer i ; The mean square error module (53) is used to define the expected value Q according to the model w , calculate the output value Sc i and the expected value Q w The mean square error value W f ; The optimization model module (54) is used to optimize the model according to multiple basic weights w C , intercept value Y J and the intercept threshold y z Calculate the optimized weight Q y , iterative intercept YC u and iterative cutoff threshold J d , based on the optimized weight Q y , iterative intercept YC u and iterative cutoff threshold J d Construct optimization model YH; Dynamic screening module (55), used to select the f Calculate dynamic filter value Do s ; Model building module (56), used to dynamically filter the value Do s Complete the establishment of the inverter status monitoring model; If W f <Do s , then the end and the inverter status monitoring model is obtained; If W f ≥Do s , then re-train the model.

6. The inverter remote management system based on Internet of Things according to claim 5, characterized in that: The expression of the optimization model YH is: Among them, Yh q represents the qth optimization variable value, and p represents the victory element S L The number of 7. The inverter remote management system based on Internet of Things according to claim 5, characterized in that: The optimization model module (54) includes: Initialization module (541) is used to initialize the basic weight w C , intercept value Y J and the intercept threshold y z Converted into multiple particles Li through vector encoding z , based on particle Li z The encoded value defines the initial position C of each particle w , speed v c , individual optimal position G z and the group optimal position Q z ; Fitness module (542), used to construct the fitness Sy for guiding particles to migrate to a better area d (Z); Update speed module (543), used to update the speed v c Calculate the updated velocity value v of the cth particle of the Dth generation c (D+1); Update position module (544), used to update the speed value v c (D+1) Calculate the updated position value Gw x (D+1); The optimal position module (545) is used to update the position value Gw according to x (D+1) calculate the optimal position value Zy a ; A judgment module (546) is used to obtain the current number of iterations U and determine whether the iteration should be stopped according to the current number of iterations; If U < 1000, continue to iterate; If U≥1000, stop the iteration and set the optimal position value Zy at this time a Perform reverse encoding to obtain the optimized weight Q y , iterative intercept YC u and iterative cutoff threshold J d .

8. The inverter remote management system based on Internet of Things according to claim 7, characterized in that: The optimal position value Zy a The calculation formula is: Among them, Zy a Indicates the ath optimal position value, Sy d [Gw x (D+1)] represents the fitness Sy d The independent variable of (Z) is the updated position value Gw x (D+1), Sy d (Q z ) represents the fitness Sy d The independent variable of (Z) is the group optimal position Q z .

Images