Data-driven and multi-factor-based insulation life prediction method for power generator-phase modifier

Abstract

The application provides a power generation-phase modifier insulation life prediction method based on data driving and multiple factors. A large motor stator insulation parameter diagnosis set containing a thermal conductivity coefficient is established, and each parameter threshold is calibrated; insulation historical operation data, online monitoring data and offline experimental data of the large power generation-phase modifier under different working conditions are obtained to form an experimental database; an insulation health degree evaluation system is constructed based on the experimental data, and the insulation health degree grade is calculated based on a matter element method and an adaptive variable weight method; a finite element coupling model of a real machine stator electric-thermal-flow multi-field is established, insulation voltage and temperature data under different insulation aging states are obtained through simulation, and the data are combined with the aforementioned experimental data to form a complete database; a relationship among insulation residual breakdown voltage, insulation temperature and insulation residual life is established, data are trained and fitted through a fruit fly-BP neural network, an electric-thermal multi-factor life model is established, the model is corrected by using a test set; and insulation residual life is output.

Description

Technical Field

[0001] This invention relates to the field of motor technology, and in particular to a method for predicting the remaining life of stator insulation in large generator-condenser based on data-driven and electro-thermal multi-factor Drosophila-BP neural network. Background Technology

[0002] Modern power systems are characterized by a high proportion of renewable energy sources (such as wind and solar power), which, while driving energy transformation, also bring new challenges. These intermittent and fluctuating power sources lack the rotational inertia and short-circuit capacity of traditional synchronous generators, leading to decreased system stability, manifested in problems such as increased voltage fluctuations, weakened frequency support, and reduced disturbance immunity. To compensate for this lack of critical support capabilities and enhance the resilience and stability of the power grid, synchronous condensers are widely used as an important solution. A synchronous condenser is essentially a synchronous motor specifically designed to operate under no-load or light-load conditions. Its core value lies in its ability to actively and rapidly inject or absorb reactive power into the grid, providing strong dynamic voltage support. Furthermore, when the system suffers disturbances (such as short-circuit faults), it provides crucial instantaneous reactive current and rotational inertia support, effectively suppressing voltage collapse and maintaining system frequency stability. It is the "stabilizing force" ensuring the safe and stable operation of the new power system.

[0003] However, as a large rotating equipment, the long-term safe and reliable operation of a synchronous condenser hinges on the health of its insulation system. Its stator windings, rotor windings, main insulation, and inter-turn insulation are subjected to the combined effects of high voltage, high current, high temperature, strong electromagnetic fields, and mechanical stress over extended periods. Insulation aging, deterioration, and even failure are among the main causes of synchronous condenser downtime. Once a critical insulation component breaks down or short-circuits, it not only causes severe damage to the synchronous condenser itself, high repair costs, and prolonged downtime, but also directly weakens the critical support capacity of the power grid, threatening the safe and stable operation of the entire system with potentially disastrous consequences. Therefore, condition monitoring and lifespan prediction of the synchronous condenser's insulation system are crucial and absolutely necessary.

[0004] Current insulation lifetime prediction methods all use residual breakdown voltage as the criterion for judging insulation failure, and parameters related to electrical factors such as dielectric loss, maximum discharge capacity, and discharge capacitance as input samples. This only considers the influence of electrical factors on insulation thermal damage. However, thermal factors are the root cause of insulation thermal degradation, and the combined effect of electrical and thermal factors can accelerate insulation thermal damage. Considering only a single factor and single variable leads to low prediction accuracy. Currently, there is a lack of insulation lifetime prediction methods that incorporate multiple electrical and thermal factors.

[0005] Therefore, it is necessary to propose a data-driven and electro-thermal multi-factor-based fruit fly-BP neural network method for predicting the remaining life of stator insulation in large-scale generator-condenser phases. Firstly, this method overcomes the limitations of traditional single / empirical models, solving the problem of limited accuracy due to empirical parameter settings. Secondly, it overcomes the shortcomings of traditional BP neural networks, such as being prone to getting trapped in local optima and slow convergence. The fruit fly-BP neural network has strong global search capabilities, enabling it to find a better or near-global optimal starting point for network parameters, significantly improving the convergence speed and prediction accuracy of the BP network, and enhancing the model's stability. Summary of the Invention

[0006] The embodiments of the present invention provide a data-driven and multi-factor-based method for predicting the insulation life of a generator-modulator, which is used to solve the technical problems existing in the prior art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution.

[0008] A data-driven and multi-factor-based method for predicting the insulation life of a generator-modulator phase-time transformer is characterized by comprising: S1 calibrates the diagnostic threshold for insulation parameters by incorporating the thermal conductivity of insulation into the existing set of insulation diagnostic parameters, combined with the DL / T series guidelines and IEEE series standards. S2 constructs a test database for large generator-condenser by acquiring historical insulation operation data, online monitoring data, and offline test data under different operating conditions. S3, based on the constructed experimental database of large-scale generator-synchronous condenser, calculates and obtains the evaluation index of insulation health through the matter-element method and adaptive variable weight algorithm. S4 constructs a simulation database for large-scale generator-condenser based on the actual size parameters of the generator-condenser. S5 uses the experimental and simulation databases of large generator-condenser to calculate the predicted results of the remaining life of the stator insulation of large generator-condenser through the fruit fly-BP neural network algorithm. The predicted results of the remaining life of the stator insulation of large generator-condenser are used for condition monitoring of the condenser insulation system.

[0009] Preferably, step S1 includes: S11 incorporates thermal conductivity into the existing set of insulation diagnostic parameters and calibrates its diagnostic threshold based on accelerated aging tests; thermal conductivity is expressed using the Fourier heat flow equation. (1) (2) Calculated; in equation (1): R sThe thermal resistance of the sample; T U The surface temperature of the upper plate; T L The temperature of the base plate; q The heat flow rate through the sample; R int The contact thermal resistance between the sample and the upper and lower plates; in equation (2): d For sample thickness, the measured thermal conductivity of multi-adhesive molded epoxy mica insulation is not less than 0.22; for multi-adhesive molded epoxy mica insulation, the measured thermal conductivity is not less than 0.29; the thermal conductivity attenuation Δ of epoxy mica insulation is... λ Within 3%; S12 combines the DL / T series guidelines and IEEE series standards to calibrate the diagnostic thresholds for common insulation characteristic parameters. The diagnostic thresholds for common insulation characteristic parameters include insulation resistance, polarization index, dielectric loss, capacitance increment, partial discharge, AC / DC withstand voltage, visual inspection, and end dynamic characteristics.

[0010] Preferably, step S2 includes: S21 acquires historical insulation operation data of large generator-synchronous condenser under different operating conditions, including cumulative operating time, insulation temperature, operating environment, annual average load rate and historical fault and defect information; S22 acquires online monitoring data through a host computer, including insulation temperature, GCM readings from the operating condition monitor, vibration displacement and frequency of the end winding, cooling water pressure, and hydrogen detection data in the water. S23 obtains offline test data of insulation through offline testing, including insulation resistance, polarization index, dielectric loss, capacitance increment, partial discharge, leakage current, and withstand voltage time; S24 constructs a test database for large generator-condenser using historical insulation operation data, online monitoring data, and offline test data under different operating conditions.

[0011] Preferably, step S3 includes: S31 divides insulation health into five levels: healthy, sub-healthy, slightly diseased, diseased, and failed; establishes a qualitative description of insulation health under each level, and constructs a comprehensive evaluation framework for insulation health that includes target layer, test layer, and index layer based on the experimental database of large generator-condenser. S32 quantifies the various parameter indices of the experimental database of a large-scale generator-synchronous condenser through calculations, using the formula... (3) The relative degradation degree of the stator insulation is calculated; where, x ij This is the actual value of the indicator.i Indicates the experimental sequence number. j Indicates the serial number of the parameter index; x 0 These are the test values ​​for the stator insulation at the time of manufacture or handover; x 1 The threshold value for insulation parameters; α This is the relative degradation rate index; if the stator insulation is in a state of natural aging, its value is taken as 1. S33 Based on the matter-element method, an organic relationship between the degree and quantity of correlation between various parameters in the experimental database is established through correlation functions. The multi-index assessment of stator insulation health is transformed into a single-index assessment. A comprehensive state assessment model combining qualitative and quantitative methods is established, and the classical domain of each test layer is determined. S34 constructs a judgment matrix based on the nine-scale method, performs normalization to obtain the single-layer weight values ​​of each indicator, and performs consistency checks. S35 introduces an equilibrium function The formula (4) Transform into formula (5) The operation modifies the weights of the indicators; where B is an aggregation function that combines the relative degradation of multiple indicators into a single degradation index, thereby reflecting the overall health status of the system or equipment; m is the number of indicators. It is the relative degradation raised to the power of T, where T is the equilibrium coefficient; S36 Based on the original static weights of the experimental layer, an adaptive weight adjustment is performed based on the distribution characteristics of the relative degradation of the indicators; S37 Through-type (6) Calculate the correlation scale K(x) between the quantified indicators and the stator insulation at each level; and use a weighted summation formula. (7) The correlation between each material element and the insulation condition level is calculated; where: K ( x ) is the first n The correlation function of state C corresponding to each test item; ρ ( X , X 0) represents the quantity X to be measured and the finite interval. X A distance of 0; ρ ( X , X i () represents the element quantity to be measured. XWith finite interval X i The distance; K j ( P ) The overall correlation value of the object to be evaluated.

[0012] Preferably, sub-step S36 includes: S361 design For the first i The static weight values ​​for each experiment, thus the static weight vector. The adjusted weight vector is Through the formula (8) (9) This maximizes the dispersion of the indicators; where: z i This indicates the experimental layer after a single weight adjustment. U i The weights of each indicator Linear combinations under the following conditions; S362 Through type (10) The results of the second-order weight adjustment of the experimental layer were calculated. W (1) In the formula, , , , , It is the first j The arithmetic mean of the indicators over all samples; It is the centralized data, that is, the original value minus the mean of the corresponding indicator.

[0013] Preferably, step S4 includes: Based on the actual machine size, S41 establishes a coupled finite element calculation model of the electric-thermal-fluid field under different insulation damage states of the generator-condenser, and sets different insulation damage schemes.

[0014] S42 determines the boundary conditions of the solution domain, including heat dissipation surface, zero potential surface, velocity inlet, pressure outlet, adiabatic surface, and zero voltage surface.

[0015] S43 uses the finite volume method to solve for the insulation electric field strength, thermal conductivity, and maximum temperature under different insulation damage states, thus forming a simulation database for a large-scale generator-condenser.

[0016] Preferably, the finite element calculation model of the electric-thermal-fluid field coupling under different insulation damage states of the generator-condenser in sub-step S41 satisfies the control equations of the electric-thermal-fluid coupling field, including the mass conservation, energy conservation and momentum conservation equations. Electric field calculation satisfies the formula , (11) In the formula: J The surface current density; J e The surface current density introduced from the outside; σ The conductivity; E Electric field strength; U Voltage; The equations for calculating the temperature field satisfy the following expression: (12) (13) In the formula: ρ e Resistivity; κ K is the Seebeck coefficient; It is the heat flux vector; Q The density of the thermal fluid.

[0017] Preferably, step S5 includes: S51 combines experimental databases and simulation data to form a complete database; S52 establishes the relationship between insulation residual breakdown voltage, insulation temperature and insulation residual life based on the annual average aging rate method, negative power theorem, Arrhenius law and lifetime loss accumulation theory. S53 uses the Pearson product-moment method to screen and perform correlation analysis on the data, selecting insulation characteristic parameters with strong correlation to insulation life as input samples. Data is then digitized and compressed to achieve normalization. S54 trains and fits the input samples using the fruit fly-backpropagation neural network algorithm, verifies the accuracy of the model using test data, and calculates the predicted insulation remaining life based on the electrical-thermal dual factor through a surrogate model.

[0018] Preferably, sub-step S54 includes: S541 creates a BP neural network structure, determines the number of nodes in each layer, and initializes the network weights and thresholds; S542 treats all network weights as fruit flies and initializes the fruit fly population positions. X i = X a ;Y i = Y a ; S543 gives a fruit fly a random optimization direction and distance, i.e. X i = X a + R ; Y i = Y a + R In the formula, X a , Y a , R For random direction and distance; S544 Based on the condition that the location of the food cannot be determined, first estimate the distance from the origin. D Then calculate the taste concentration judgment value. S , ; S =1 / D; S545 Substitute S into the flavor concentration determination function to calculate the current flavor concentration and save the value; S546 Find the optimal value of the flavor concentration determination function. At this point, the minimum value should be selected, and the current network weights should be saved. S547 Updates the optimal value of the fitness function to determine the fruit fly's position. At this point, the weight coordinates are: X a = X b ; Y a = Y b ; S548 The iteration begins. Determine whether the fitness function value is better than the previous generation. If so, proceed to step 7; otherwise, proceed to sub-step S543. S549 uses the optimal weights for training the BP neural network; after training, the performance of the BP network is tested using test data.

[0019] As can be seen from the technical solutions provided by the embodiments of the present invention above, the present invention solves the problem of low prediction accuracy caused by considering only a single factor and single variable in the prediction of the stator insulation life of a generator-condenser under the background of new power systems. The proposed method for predicting the remaining life of stator insulation of large generator-condenser based on data-driven and electro-thermal multi-factor fruit fly-BP neural network breaks through the limitations of traditional single / empirical models and solves the problem of limited accuracy due to model parameter dependence on empirical settings. The fruit fly-BP neural network overcomes the defects of traditional BP neural networks that are prone to getting trapped in local optima and slow convergence. It has a strong global search capability and can find a better or closer to the global optimum network parameter starting point, significantly improving the convergence speed and prediction accuracy.

[0020] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of the invention. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 The flowchart illustrates a data-driven and multi-factor-based method for predicting the insulation life of a generator-modulator phase-shifting camera, as provided in an embodiment of the present invention. Figure 2 This is a partial schematic diagram of a diagnostic set of motor stator insulation parameters including thermal conductivity, which is provided in an embodiment of the present invention for a data-driven and multi-factor-based method for predicting the insulation life of a generator-condenser. Figure 3 This is a partial schematic diagram of a simulation database containing thermal conductivity, which is provided in an embodiment of the present invention for a data-driven and multi-factor-based method for predicting the insulation life of a generator-modulator. Figure 4 The present invention provides a data-driven and multi-factor-based method for predicting the insulation life of a generator-condenser, which includes an insulation health evaluation system framework that considers offline testing and online monitoring. Figure 5 The flowchart of the insulation health assessment method based on the matter-element method and weight adaptive adjustment is provided by the present invention for a data-driven and multi-factor-based method for predicting the insulation life of a generator-condenser. Figure 6The flowchart of the insulation lifetime prediction method based on electro-thermal multi-factor and fruit fly-BP neural network provided by the present invention is as follows: Figure 7 The flowchart of the training process of the fruit fly-BP neural network for a data-driven and multi-factor-based method for predicting the insulation life of a generator-modulator phase-shifting camera is provided in this invention. Detailed Implementation

[0023] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0024] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0025] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0026] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0027] See Figures 1 to 3This invention provides a method for predicting the remaining life of stator insulation in large-scale generator-condenser based on data-driven and electro-thermal multi-factor Drosophila-BP neural network. The method includes insulation parameter diagnostic threshold calibration, experimental database construction, insulation health assessment, simulation database construction, and insulation life prediction based on the Drosophila-BP neural network algorithm and electro-thermal multi-factors. Specifically, it can be implemented through the following steps: S1 calibrates the diagnostic threshold for insulation parameters by incorporating the thermal conductivity of insulation into the existing set of insulation diagnostic parameters, combined with the DL / T series guidelines and IEEE series standards. S2 constructs a test database for large generator-condenser by acquiring historical insulation operation data, online monitoring data, and offline test data under different operating conditions. S3, based on the constructed experimental database of large-scale generator-synchronous condenser, calculates and obtains the evaluation index of insulation health through the matter-element method and adaptive variable weight algorithm. S4 constructs a simulation database for large-scale generator-condenser based on the actual size parameters of the generator-condenser. Based on the experimental and simulation databases of large generator-condenser phase converters, S5 uses the fruit fly-BP neural network algorithm to calculate the predicted results of the remaining life of the stator insulation of large generator-condenser phase converters.

[0028] The predicted results of the remaining life of the stator insulation of large generator-condenser are used for condition monitoring of the insulation system of large generator-condenser, providing strong support for the maintenance optimization of large generator-condenser.

[0029] In the preferred embodiment provided by the present invention, the specific execution process of each step is as follows.

[0030] I. Insulation Parameter Diagnostic Threshold Calibration

[0031] S11. Incorporate the thermal conductivity of insulation into the existing set of insulation diagnostic parameters and calibrate its diagnostic threshold based on accelerated aging tests. The thermal conductivity is measured using the heat flow meter method. The test system consists of main components such as an upper heating plate, a reference heat flow meter, a heating guard plate, and a lower heating plate. The method is based on the Fourier heat flow equation:

[0032] In the formula: R s The thermal resistance of the sample; T U The surface temperature of the upper plate; T L The temperature of the base plate; q The heat flow rate through the sample; R int Let be the contact thermal resistance between the sample and the upper and lower plates. From this, the thermal conductivity of the insulation can be obtained as:

[0033] In the formula, d The thickness is the sample thickness. For multi-adhesive molded epoxy mica insulation, the measured thermal conductivity (50℃) shall not be less than 0.22; for multi-adhesive molded epoxy mica insulation, the measured thermal conductivity (50℃) shall not be less than 0.29. The thermal conductivity attenuation Δ of epoxy mica insulation is also considered. λ It should be within 3%.

[0034] S12. In conjunction with the DL / T series guidelines and IEEE series standards, calibrate the diagnostic thresholds for conventional insulation characteristic parameters (insulation resistance, polarization index, dielectric loss, capacitance increment, partial discharge, AC / DC withstand voltage, as well as visual inspection, end dynamic characteristics, etc.).

[0035] II. Construction of Experimental Database

[0036] like Figure 4 As shown, it includes the following sub-steps: S21. Obtain historical operating data of insulation under different operating conditions of large generator-synchronous condenser (cumulative operating time, insulation temperature, operating environment, annual average load rate, and historical fault and defect information).

[0037] S22. Obtain online monitoring data through the host computer, such as insulation temperature, GCM reading of the operating condition monitor, vibration displacement and frequency of the end winding, cooling water pressure and hydrogen detection in the water, etc.

[0038] S23. Insulation resistance, polarization index, dielectric loss, capacitance increment, partial discharge, leakage current, and withstand voltage time are obtained through offline tests. The above historical operating data, online monitoring data, and offline test data constitute the experimental database of the large-scale generator-condenser.

[0039] S24. Construct a test database for large generator-condenser using historical insulation operation data, online monitoring data, and offline test data under different operating conditions.

[0040] III. Insulation Health Assessment Based on Matter-Element Method and Adaptive Variable Weight

[0041] like Figure 5 As shown, it includes the following sub-steps: S31. Insulation health is divided into five levels: healthy, sub-healthy, slightly diseased, diseased, and failed. A qualitative description of insulation health is established for each level. Based on the experimental database of large-scale generator-condenser, a comprehensive evaluation framework for insulation health, including target, test, and index layers, is constructed. S32. Quantify the parameters of the experimental database of large-scale generator-synchronous condenser through calculation, and then use the formula...

[0042] The relative degradation degree is calculated; where, x ij This is the actual value of the indicator. j Indicates the serial number of the parameter index; x 0 These are the test values ​​for insulation at the time of manufacture or handover. x 1 The threshold value for insulation parameters; α This is the relative degradation rate index; if the insulation is in the process of natural aging, its value is taken as 1. S33. Based on the matter-element method, an organic connection between the degree and quantity of correlation between indicators is established through correlation functions. The multi-indicator assessment of stator insulation health is transformed into a single-indicator assessment. A comprehensive state assessment model combining qualitative and quantitative methods is established, and the classical domain of each test layer is determined. S34. Construct a judgment matrix based on the nine-scale method, perform normalization to obtain the single-layer weight values ​​of each indicator, and perform consistency verification. S35. To avoid the problem that a static weight calculation mode might mask certain information due to an underweight indicator, resulting in a seemingly good overall evaluation that deviates from the true state, an adaptive secondary weight adjustment scheme is proposed. First, the indicator layer is reweighted using the following formula: (4) Introducing an equilibrium function The above formula can be rewritten as: (5).

[0043] In the formula: B is an aggregation function, which integrates the relative deterioration of multiple indicators into an overall deterioration index, thereby reflecting the overall health status of the system or equipment; m is the number of indicators; It is the relative degradation raised to the power of T, where T is the equilibrium coefficient.

[0044] S36. Based on the original static weights of the experimental layer, perform an adaptive weight adjustment guided by the distribution characteristics of the relative degradation of the indicators. Specifically, this includes: S361, Let For the first i The static weight values ​​for each experiment, thus the static weight vector. The adjusted weight vector is The variance measure is used to assess the dispersion of relative degradation, and the formula is used to... (8) (9) This maximizes the dispersion of the indicators; where: z i This indicates the experimental layer after a single weight adjustment. U i The weights of each indicator Linear combinations under the following conditions; S362, Through-type (10) The results of the second-order weight adjustment of the experimental layer were calculated. W (1) In the formula, , , , , It is the first j The arithmetic mean of the indicators over all samples; It is the centralized data, that is, the original value minus the mean of the corresponding indicator.

[0045] S37, Through-type (6) Calculate the correlation scale K(x) between the quantified indicators and the stator insulation at each level; and use a weighted summation formula. (7) The correlation between each material element and the insulation condition level is calculated; where: K ( x ) is the first n The correlation function of state C corresponding to each test item; ρ ( X , X 0) represents the quantity X to be measured and the finite interval. X A distance of 0; ρ ( X , X i () represents the element quantity to be measured. X With finite interval X i Distance (of the classical domain); K j ( P ) The overall correlation value of the object to be evaluated.

[0046] IV. Simulation Database Construction

[0047] Based on the actual machine size, S41 establishes a coupled finite element calculation model of the electric-thermal-fluid field under different insulation damage states of the generator-condenser, and sets different insulation damage schemes.

[0048] The finite element calculation model of the electric-thermal-fluid field coupling under different insulation damage states of the generator-condenser satisfies the control equations of the electric-thermal-fluid coupled field, including the mass conservation, energy conservation and momentum conservation equations. Electric field calculation satisfies the formula , (11) In the formula: J The surface current density; J e The surface current density introduced from the outside; σ The conductivity; E Electric field strength; U Voltage; The equations for calculating the temperature field satisfy the following expression: (12) (13) In the formula: ρ e Resistivity; κ K is the Seebeck coefficient; It is the heat flux vector; Q The density of the thermal fluid.

[0049] S42 determines the boundary conditions of the solution domain, including heat dissipation surface, zero potential surface, velocity inlet, pressure outlet, adiabatic surface, and zero voltage surface.

[0050] S43 uses the finite volume method to solve for the insulation electric field strength, thermal conductivity, and maximum temperature under different insulation damage states, thus forming a simulation database for a large-scale generator-condenser.

[0051] V. Insulation Life Prediction Based on Fruit Fly-BP Neural Network Algorithm and Electrothermal Multifactors

[0052] like Figure 6 As shown, it can be the following sub-steps: S51. Combine the experimental database and simulation data to form a complete database; S52. Based on the annual average aging rate method, negative power theorem, Arrhenius law and lifetime loss accumulation theory, establish the relationship between insulation residual breakdown voltage, insulation temperature and insulation residual lifetime. S53. Based on the Pearson product-moment method, data is screened and correlation analysis is performed to select insulation characteristic parameters with strong correlation to insulation life as input samples. Data is then digitized and compressed to complete normalization. S54. The input samples are trained and fitted using the fruit fly-BP neural network algorithm. The accuracy of the model is verified using test data. The prediction results of the remaining insulation lifetime are calculated using a surrogate model based on the electrical-thermal dual factor.

[0053] Sub-step S54 specifically includes: S541. Create a BP neural network structure, determine the number of nodes in each layer, and initialize the weights and thresholds of the network. S542. Treat all network weights as fruit flies and initialize the fruit fly population positions. X i = X a ; Y i = Y a ; S543. Given a fruit fly a random optimization direction and distance, i.e. X i = X a + R ; Y i = Y a + R In the formula, X a , Y a , R For random direction and distance; S544. Since the location of the food cannot be determined, first estimate the distance from the origin. D Then calculate the taste concentration judgment value. S , ; S =1 / D; S545. Substitute S into the flavor concentration determination function, calculate the current flavor concentration, and save the value. S546. Find the optimal value of the flavor concentration determination function. At this point, the minimum value should be selected, and the current network weights should be saved. S547. When updating the optimal value of the fitness function, the fruit fly position is as follows: X a = X b ; Y a = Yb ; S548. The iteration begins. Determine whether the fitness function value is better than the previous generation. If so, proceed to step 7; otherwise, proceed to sub-step S543. S549. Utilize the optimal weights for BP neural network training; after training, test the performance of the BP network using test data.

[0054] In some feasible embodiments, the fruit fly-backpropagation (BP) neural network has a 3-layer architecture with 6 input parameters and the network output is the remaining insulation lifetime. Therefore, the number of nodes in the input and output layers is 7 and 1, respectively, and the number of nodes in the intermediate layers is 21. The transfer function for the intermediate layers is logsig, and the transfer function for the output layer is purelin. trainlm is used as the training function. The network's training objective error is 0.001, the learning efficiency is 0.2, and the number of training iterations is 5000. The number of fruit flies in the population will be determined based on the designed BP neural network structure, set to 20, with 100 consecutive iterations. The fitness function is selected as the mean square error between the expected and actual results.

[0055] In summary, this invention provides a data-driven, multi-factor (electro-thermal) fruit fly-BP neural network-based method for predicting the remaining life of stator insulation in large-scale generator-condenser phases. This method overcomes the limitations of traditional single-factor, single-model parameter-dependent methods with limited accuracy. By considering breakdown voltage and thermal conductivity attenuation factors, it improves the accuracy of insulation life prediction. The fruit fly-BP neural network overcomes the shortcomings of traditional BP neural networks, such as being prone to getting trapped in local optima and slow convergence, thus improving convergence speed and prediction accuracy.

[0056] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0057] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0058] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0059] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A data-driven and multi-factor-based method for predicting the insulation life of a generator-modulator phase-time transformer, characterized in that, include: S1 calibrates the diagnostic threshold for insulation parameters by incorporating the thermal conductivity of insulation into the existing set of insulation diagnostic parameters, and combining the DL / T series guidelines and IEEE series standards. S2 constructs a test database for large generator-condenser by acquiring historical insulation operation data, online monitoring data, and offline test data under different operating conditions. S3, based on the constructed experimental database of large-scale generator-synchronous condenser, calculates and obtains the evaluation index of insulation health through the matter-element method and adaptive variable weight algorithm. S4 constructs a simulation database for large-scale generator-condenser based on the actual size parameters of the generator-condenser. S5 uses the experimental and simulation databases of large generator-condenser to calculate the predicted results of the remaining life of the stator insulation of large generator-condenser through the fruit fly-BP neural network algorithm. The predicted results of the remaining life of the stator insulation of large generator-condenser are used for condition monitoring of the condenser insulation system.

2. The method according to claim 1, characterized in that, Step S1 includes: S11 incorporates thermal conductivity into the existing set of insulation diagnostic parameters and calibrates its diagnostic threshold based on accelerated aging tests; thermal conductivity is expressed using the Fourier heat flow equation. （1） （2） Calculated; in equation (1): R s The thermal resistance of the sample; T U The surface temperature of the upper plate; T L The temperature of the base plate; q The heat flow rate through the sample; R int The contact thermal resistance between the sample and the upper and lower plates; in equation (2): d For sample thickness, the measured thermal conductivity of multi-adhesive molded epoxy mica insulation is not less than 0.22; for multi-adhesive molded epoxy mica insulation, the measured thermal conductivity is not less than 0.29; the thermal conductivity attenuation Δ of epoxy mica insulation is... λ Within 3%; S12 combines the DL / T series guidelines and IEEE series standards to calibrate the diagnostic thresholds for conventional insulation characteristic parameters; the diagnostic thresholds for conventional insulation characteristic parameters include insulation resistance, polarization index, dielectric loss, capacitance increment, partial discharge, AC / DC withstand voltage, visual inspection, and end dynamic characteristics.

3. The method according to claim 2, characterized in that, Step S2 includes: S21 acquires historical insulation operation data of large generator-synchronous condenser under different operating conditions, including cumulative operating time, insulation temperature, operating environment, annual average load rate and historical fault and defect information; S22 acquires online monitoring data through a host computer, including insulation temperature, GCM readings from the operating condition monitor, vibration displacement and frequency of the end winding, cooling water pressure, and hydrogen detection data in the water. S23 obtains offline test data of insulation through offline testing, including insulation resistance, polarization index, dielectric loss, capacitance increment, partial discharge, leakage current, and withstand voltage time; S24 constructs a test database for large generator-condenser using historical insulation operation data, online monitoring data, and offline test data under different operating conditions.

4. The method according to claim 3, characterized in that, Step S3 includes: S31 divides insulation health into five levels: healthy, sub-healthy, slightly diseased, diseased, and failed; establishes a qualitative description of insulation health under each level, and constructs a comprehensive evaluation framework for insulation health that includes target layer, test layer, and index layer based on the experimental database of large generator-condenser. S32 quantifies the various parameter indices of the experimental database of a large-scale generator-synchronous condenser through calculations, using the formula... （3） The relative degradation degree of the stator insulation is calculated; where, x ij This is the actual value of the indicator. i Indicates the experimental sequence number. j Indicates the serial number of the parameter index; x 0 These are the test values ​​for the stator insulation at the time of manufacture or handover; x 1 The threshold value for insulation parameters; α This is the relative degradation rate index; if the stator insulation is in a state of natural aging, its value is taken as 1. S33 Based on the matter-element method, an organic relationship between the degree and quantity of correlation between various parameters in the experimental database is established through correlation functions. The multi-index assessment of stator insulation health is transformed into a single-index assessment. A comprehensive state assessment model combining qualitative and quantitative methods is established, and the classical domain of each test layer is determined. S34 constructs a judgment matrix based on the nine-scale method, performs normalization to obtain the single-layer weight values ​​of each indicator, and performs consistency checks. S35 introduces an equilibrium function The formula （4） Transform into formula （5） The operation modifies the weights of the indicators; where B is an aggregation function that combines the relative degradation of multiple indicators into a single degradation index, thereby reflecting the overall health status of the system or equipment; m is the number of indicators. It is the relative degradation raised to the power of T, where T is the equilibrium coefficient; S36 Based on the original static weights of the experimental layer, an adaptive weight adjustment is performed based on the distribution characteristics of the relative degradation of the indicators; S37 Through-type （6） Calculate the correlation scale K(x) between the quantified indicators and the stator insulation at each level; and use a weighted summation formula. （7） The correlation between each material element and the insulation condition level is calculated; where: K ( x ) is the first n The correlation function of state C corresponding to each test item; ρ ( X , X 0) represents the quantity X to be measured and the finite interval. X A distance of 0; ρ ( X , X i () represents the element quantity to be measured. X With finite interval X i The distance; K j ( P ) The overall correlation value of the object to be evaluated.

5. The method according to claim 4, characterized in that, Sub-step S36 includes: S361 design For the first i The static weight values ​​of each experiment, thus the static weight vector. The adjusted weight vector is Through the formula （8） （9） This maximizes the dispersion of the indicators; where: z i This indicates the experimental layer after a single weight adjustment. U i The weight of each indicator in the middle Linear combinations under the following conditions; S362 Through Type （10） The results of the second-order weight adjustment of the experimental layer were calculated. W (1) In the formula, , , , , It is the first j The arithmetic mean of the indicators over all samples; It is the centralized data, that is, the original value minus the mean of the corresponding indicator.

6. The method according to claim 5, characterized in that, Step S4 includes: Based on the actual machine size, S41 establishes a coupled finite element calculation model of the electric-thermal-fluid field under different insulation damage states of the generator-condenser, and sets different insulation damage schemes. S42 determines the boundary conditions of the solution domain, including heat dissipation surface, zero potential surface, velocity inlet, pressure outlet, adiabatic surface, and zero voltage surface. S43 uses the finite volume method to solve for the insulation electric field strength, thermal conductivity, and maximum temperature under different insulation damage states, thus forming a simulation database for a large-scale generator-condenser.

7. The method according to claim 6, characterized in that, The finite element calculation model of the electric-thermal-fluid field coupling under different insulation damage states of the generator-condenser in sub-step S41 satisfies the control equations of the electric-thermal-fluid coupling field, including the mass conservation, energy conservation and momentum conservation equations. Electric field calculation satisfies the formula ， （11） In the formula: J The surface current density; J e The surface current density introduced from the outside; σ The conductivity; E Electric field strength; U Voltage; The equations for calculating the temperature field satisfy the following formula: （12） （13） In the formula: ρ e Resistivity; κ K is the Seebeck coefficient; It is the heat flux vector; Q The density of the thermal fluid.

8. The method according to claim 6, characterized in that, Step S5 includes: S51 combines experimental databases and simulation data to form a complete database; S52 establishes the relationship between insulation residual breakdown voltage, insulation temperature and insulation residual life based on the annual average aging rate method, negative power theorem, Arrhenius law and lifetime loss accumulation theory. S53 uses the Pearson product-moment method to screen and perform correlation analysis on the data, selecting insulation characteristic parameters with strong correlation to insulation life as input samples. Data is then digitized and compressed to achieve normalization. S54 trains and fits the input samples using the fruit fly-backpropagation neural network algorithm, verifies the accuracy of the model using test data, and calculates the predicted insulation remaining life based on the electrical-thermal dual factor through a surrogate model.

9. The method according to claim 7, characterized in that, Sub-step S54 includes: S541 creates a BP neural network structure, determines the number of nodes in each layer, and initializes the network weights and thresholds; S542 treats all network weights as fruit flies and initializes the fruit fly population positions. X i = X a ; Y i = Y a ; S543 gives a fruit fly a random optimization direction and distance, i.e. X i = X a + R ; Y i = Y a + R In the formula, X a , Y a , R For random direction and distance; S544 Based on the condition that the location of the food cannot be determined, first estimate the distance from the origin. D Then calculate the taste concentration judgment value. S , ; S =1 / D; S545 Substitute S into the flavor concentration determination function to calculate the current flavor concentration and save the value; S546 Find the optimal value of the flavor concentration determination function. At this point, the minimum value should be selected, and the current network weights should be saved. S547 Updates the optimal value of the fitness function to determine the fruit fly's position. At this point, the weight coordinates are: X a = X b ; Y a = Y b ; S548 The iteration begins. Determine whether the fitness function value is better than the previous generation. If so, proceed to step 7; otherwise, proceed to sub-step S543. S549 uses the optimal weights for training the BP neural network; after training, the performance of the BP network is tested using test data.

Images