Transformer life prediction method based on digital-analog linkage and gradient descent
Through the method based on digital-to-analog linkage and gradient descent, the life prediction of transformers is optimized, and the problem of model complexity and data-driven methods in the prior art is solved, and more accurate life prediction and transformer life cycle extension are achieved.
Patent Information
- Application Number
- CN202411861973.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-05-16
AI Technical Summary
The prior art has problems such as increasing model complexity and lack of transparency in the transformation life prediction, resulting in inaccurate prediction and ineffective extension of the transformer life cycle.
Using a method based on digital-to-analog linkage and gradient descent, accurate prediction of transformer life and remaining life is achieved by building data sets, establishing composite health indicators and stochastic degradation models, and optimizing model parameters and weighting coefficients.
It improves the accuracy of the remaining life prediction of the distribution network transformer, forms a closed-loop cross-linking mechanism between the data and the model, effectively extending the life cycle of the transformer.
Smart Images
Figure CN120012542A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of transformers, and in particular to a transformer life prediction method based on digital-analog linkage and gradient descent. Background Art
[0002] The distribution network transformer is the core equipment connecting the distribution network and the user's electricity consumption side. The stable operation of the transformer plays a vital role in the safe and stable operation of the entire power network. At present, the two methods of regular maintenance and post-maintenance are generally used for the distribution network transformers in operation. However, these two methods will cause the problems of "over-maintenance" and "disrepair". On the one hand, it will waste a lot of manpower and material resources. On the other hand, it may cause the transformer to be out of operation unplanned, affecting the stable operation of the power system. Therefore, the prediction of the remaining life of the transformer is a necessary link for the safe and stable operation of the distribution network transformer. At the same time, accurate prediction can effectively extend the life cycle of the distribution network transformer and improve its performance and economic benefits.
[0003] At present, the transformer life prediction technology is mainly divided into two categories: failure physical model method and data-driven method. The failure physical model method is highly dependent on the random degradation model. When the number of devices is large, individual differences will increase the complexity of the model. The data-driven method uses artificial intelligence technology to learn the degradation pattern of the system from existing observation data, but it lacks transparency, the results are difficult to interpret, and it is impossible to characterize the uncertainty of the remaining life distribution during operation. The above two methods have their own advantages and disadvantages. In order to solve the shortcomings of model-driven and data-driven, a distribution network transformer remaining life prediction method that optimizes model parameters through data-driven optimization plays a vital role in ensuring the safe and stable operation of distribution network transformers. Summary of the invention
[0004] In order to solve the above technical problems, the present invention provides a transformer life prediction method based on digital-analog linkage and gradient descent, comprising the following steps:
[0005] Step 1: Construct a data set based on the historical operation data of the distribution network transformer and perform normalization preprocessing on the data set;
[0006] Step 2: Weighted fusion of various indicators of the preprocessed data set to construct a composite health indicator;
[0007] Step 3: Establish a transformer random degradation process model of nonlinear Wiener process;
[0008] Step 4: The life span and remaining life span prediction is realized by the time when the composite health indicator reaches the failure threshold for the first time at the current moment;
[0009] Step 5: Establish a mean square error target optimization function based on the predicted remaining life and the actual remaining life;
[0010] Step 6: Use the gradient descent algorithm to optimize the target optimization function and perform reverse optimization adjustment on the data set weight coefficients and model parameters;
[0011] Step 7: Predict the remaining life of the transformer based on the optimized model parameters and weighting coefficients.
[0012] In a preferred embodiment, the transformer historical operation data set includes the following primary data sets, gas data set, insulating oil data set, and electrical data set; the gas data set includes the following secondary data sets: hydrogen content, methane content, acetylene content, ethylene content, ethane content, and carbon oxide content; the insulating oil data set includes the following secondary data sets: insulating oil dielectric loss, insulation resistance absorption ratio, oil breakdown voltage, water content in oil, gas content in oil, and furfural content; the electrical data set includes the following secondary data sets: insulation resistance value, partial discharge amount, dielectric loss, leakage resistance value, insulation resistance absorption ratio, and volume resistivity.
[0013] In a preferred embodiment, the data set normalization preprocessing method comprises:
[0014] Let x i,j (t) is the j{1≤j≤S}th secondary historical data set of the i{1≤i≤N}th transformer in the distribution network transformer data set at t(t=0,1,2,…,K i ) time, the performance degradation history data of the transformer after normalization x' i.j (t) is as follows:
[0015]
[0016] Among them, min(x i,j )、max(x i,j ) are the minimum and maximum values of the performance degradation historical data in the j-th secondary historical data set of the ith distribution network transformer, respectively.
[0017] In the preferred embodiment, in step 2, the composite health index Z after integrating multiple secondary data sets is obtained based on the historical data set of the distribution network transformer. i (t) is as follows:
[0018]
[0019] where a j is the weight of the j-th indicator.
[0020] In a preferred embodiment, a random degradation function of the degradation model is established, and the nonlinear Wiener process modeling of the standard Brownian motion and the transformer composite health index Z are considered. iThe degradation process of (t) changing with time, the random degradation function of the composite health index is constructed including:
[0021]
[0022] Among them, B(t) is the standard Brownian motion, θ i , β i , σ i is the model parameter, and S i (t) represents the logarithmic transformation of the i-th random degradation system at time t, and its formula is:
[0023]
[0024] Let θ' i =lnθ i , The formula is:
[0025] S i (t) = θ′ i +β′ i t+σ i B(t).
[0026] In a preferred embodiment, in step 4, the life value T of the distribution network transformer random degradation is defined as i :
[0027] T i = inf{t i :S i (t i )≥b|s i,0 <b}
[0028] s i,0 =lnz i,0
[0029] where t i The historical data set time recorded for the i-th transformer, s i,0 is the natural logarithm of the initial composite health index of the i-th transformer, and b is the critical threshold corresponding to the composite health index;
[0030] Define the remaining life value L of the distribution network transformer random degradation i，k :
[0031] L i,k = inf{l i,k :S i (t i,k +l i,k )≥b|s i,k <b}
[0032] s i,k =lnzi,k
[0033] where s i,k is the natural logarithm of the composite health index of the i-th transformer at time k, t i,k is the life value of the i-th transformer at time k, l i,k is the remaining life value of the i-th transformer at time k, and b is the critical threshold corresponding to the natural logarithm of the composite health index;
[0034] Calculating the expected life expectancy includes,
[0035]
[0036] Among them, t i,0 The initial time of the i-th transformer, t i,Ki The final moment of the i-th transformer, s i,Ki is the natural logarithm of the transformer composite health index of the i-th transformer at the final moment;
[0037] The expected prediction for calculating the remaining life of distribution network transformers includes:
[0038]
[0039] In a preferred embodiment, the mean square error objective optimization function constructed based on the predicted remaining life and the actual remaining life includes:
[0040]
[0041] Where A is the indicator weight a j A collection of; and l i,k is the predicted remaining life value and actual remaining life value of the i-th transformer at time k.
[0042] In a preferred embodiment, the target optimization function is optimized using a gradient descent algorithm, and reverse optimization adjustment of the data set weighting coefficients and model parameters includes:
[0043]
[0044] Where A' is the optimal indicator weight set, b' is the optimal critical threshold; argmin represents the parameter value at which the function achieves the minimum value in its domain.
[0045] In a preferred embodiment, a gradient descent algorithm is used to optimize the target optimization function, and its parameter update process is iteratively expressed as:
[0046]
[0047] Where Υ represents the parameters (A, b) to be updated, η represents the learning rate, and r represents the iteration step. The iteration process will stop when the parameters converge or the number of iterations reaches a predetermined value.
[0048] In a preferred embodiment, in step 7, the logarithmic transformation of the i-th stochastic degradation system at time t is updated to S′ i (t),
[0049]
[0050] The predicted expected update of the remaining life of the distribution network transformer is:
[0051]
[0052] Compared with the prior art, the present invention has the following beneficial technical effects:
[0053] According to the transformer life prediction method based on digital-analog linkage and gradient descent in an embodiment of the present invention, the method combines the transformer historical data with a quasi-exponential random degradation model to form a closed-loop cross-linkage mechanism between the data and the model, thereby improving the accuracy of the remaining life prediction of the distribution network transformer. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] Figure 1 It is a flow chart of a transformer life prediction method based on digital-analog linkage and gradient descent provided by the present invention. DETAILED DESCRIPTION
[0055] The following is combined with Figure 1 The specific implementation of the present invention is further described in detail. The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. ② The same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.
[0056] In the description of the present invention, it is necessary to understand that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.
[0057] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0058] Embodiment 1:
[0059] like Figure 1 As shown, the specific steps of the transformer life prediction method based on digital-analog linkage and gradient descent of the present invention are shown, and the transformer life prediction method specifically includes the following steps:
[0060] Step 1: Construct a data set based on the historical operation data of distribution network transformers and perform normalization preprocessing on the data set.
[0061] The transformer historical operation data set includes three primary data sets: gas data set, insulating oil data set, and electrical data set;
[0062] The three primary datasets include six secondary datasets.
[0063] Secondary datasets include,
[0064] Gas data set: hydrogen content, methane content, acetylene content, ethylene content, ethane content, carbon oxide content;
[0065] Insulating oil data set: insulating oil dielectric loss, insulation resistance absorption ratio, oil breakdown voltage, water content in oil, gas content in oil, furfural content;
[0066] Electrical data set: insulation resistance value, partial discharge, dielectric loss, leakage resistance value, insulation resistance absorption ratio, volume resistivity;
[0067] The data set normalization preprocessing methods include:
[0068] Let x i,j (t) is the j{1≤j≤S}th secondary historical data set of the i{1≤i≤N}th transformer in the distribution network transformer data set at t(t=0,1,2,…,K i ) time, the performance degradation history data of the transformer after normalization x' i.j (t) is as follows:
[0069]
[0070] Among them, min(x i,j )、max(x i,j ) are the minimum and maximum values of the performance degradation historical data in the j-th secondary historical data set of the ith distribution network transformer, respectively.
[0071] Step 2: Weighted fusion of various indicators of the preprocessed data set to construct a composite health indicator;
[0072] According to the historical data set of the i-th distribution network transformer, the composite health index Z after integrating multiple secondary data sets i (t) is as follows:
[0073]
[0074] where a j is the weight of the j-th indicator;
[0075] Step 3: Establish a transformer random degradation process model of nonlinear Wiener process;
[0076] The random degradation function of the degradation model is established, and the nonlinear Wiener process modeling of the standard Brownian motion and the composite health index Z of the i-th transformer are considered. i The degradation process of (t) changing with time, the random degradation function of the composite health index is constructed including:
[0077]
[0078] Among them, B(t) is the standard Brownian motion, which reflects the randomness of time changes in the degradation process; θ i , β i , σ i is a model parameter, reflecting the time-varying uncertainty of the degradation process. For the established exponential random degradation model, it is linearized by logarithmic transformation. i (t) represents the logarithmic transformation of the i-th random degradation system at time t, and its formula is:
[0079]
[0080] Let θ' i =lnθ i , The formula is:
[0081] S i (t) = θ′ i +β′ i t+σ i B(t)
[0082] Step 4: The life span and remaining life span prediction is realized by the time when the composite health indicator reaches the failure threshold for the first time at the current moment.
[0083] Define the life value T of the random degradation of the distribution network transformer i , as follows:
[0084] T i = inf{ti :S i (t i )≥b|s i,0 <b}
[0085] s i,0 =lnz i,0
[0086] Among them, Z i,0 represents the initial composite health index of the i-th transformer, t i The historical data set time recorded for the i-th transformer, s i,0 is the natural logarithm of the initial composite health index of the i-th transformer, b is the critical threshold corresponding to the composite health index, and inf represents the definition;
[0087] Define the remaining life value L of the distribution network transformer random degradation i，k , as follows:
[0088] L i,k = inf{l i,k :S i (t i,k +l i,k )≥b|s i,k <b}
[0089] s i,k =lnz i,k
[0090] Among them, z i,k represents the composite health index of the i-th transformer at time k, s i,k is the natural logarithm of the composite health index of the i-th transformer at time k, t i,k is the life value of the i-th transformer at time k, l i,k is the remaining life value of the i-th transformer at time k, and b is the critical threshold corresponding to the natural logarithm of the composite health index;
[0091] Calculate the predicted expectation of derived life expectancy value include,
[0092]
[0093] Among them, E[] represents the expected value, t i,0 The initial time of the i-th transformer, t i,Ki The final moment of the i-th transformer, s i,Ki is the natural logarithm of the transformer composite health index of the i-th transformer at the final moment.
[0094] Calculate and derive the expected remaining life of distribution network transformers include,
[0095]
[0096] Step 5: Establish the mean square error objective optimization function based on the predicted remaining life and the actual remaining life:
[0097]
[0098] Where A is the indicator weight a j A collection of; and l i,k is the predicted remaining life value and actual remaining life value of the i-th transformer at time k. There are i transformers here, and K of each transformer i Different, K i Represents all sampling points of each transformer.
[0099] Step 6: Use the gradient descent algorithm to optimize the target optimization function and reversely optimize the data set weight coefficients and model parameters:
[0100]
[0101] Where A' is the optimal indicator weight set, b' is the optimal critical threshold; argmin represents the parameter value at which the function achieves the minimum value in its domain.
[0102] In order to solve the complexity of the target optimization function, the gradient descent algorithm is used to optimize the target function, and its parameter update process is expressed in an iterative way as follows:
[0103]
[0104] Where γ represents the parameters (A, b) to be updated, η represents the learning rate, and r represents the iteration step. The iteration process stops when the parameters converge or the number of iterations reaches a predetermined value.
[0105] Step 7: Predict the remaining life of the transformer based on the optimized model parameters and weighting coefficients.
[0106] Based on the optimized parameters and weights, the logarithmic transformation of the i-th stochastic degradation system at time t is updated to S′ i (t),
[0107] The predicted expected update of the remaining life of the distribution network transformer is:
[0108]
[0109] In summary, the implementation method of the present invention is based on the traditional remaining life prediction, optimizes the weights and parameters of the composite health index and the random degradation model according to the transformer data set, so that the weight coefficient distribution is reasonable, and realizes the change of the composite weight and the critical threshold according to the actual value of the historical data, thereby improving the accuracy of the prediction method.
[0110] According to the transformer life prediction method based on digital-analog linkage and gradient descent in an embodiment of the present invention, the method combines the transformer historical data with a quasi-exponential random degradation model to form a closed-loop cross-linkage mechanism between the data and the model, thereby improving the accuracy of the remaining life prediction of the distribution network transformer.
[0111] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to an embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any verified manner in one or more embodiments or examples.
[0112] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.
Claims
1. A transformer life prediction method based on digital-analog linkage and gradient descent, characterized in that: The following steps are involved: Step 1: Construct a data set based on the historical operation data of the distribution network transformer and perform normalization preprocessing on the data set; Step 2: Weighted fusion of various indicators of the preprocessed data set to construct a composite health indicator; Step 3: Establish a transformer random degradation process model of nonlinear Wiener process; Step 4: The life span and remaining life span prediction is realized by the time when the composite health indicator reaches the failure threshold for the first time at the current moment; Step 5: Establish a mean square error target optimization function based on the predicted remaining life and the actual remaining life; Step 6: Use the gradient descent algorithm to optimize the target optimization function and reversely optimize the data set weight coefficients and model parameters; Step 7: Predict the remaining life of the transformer based on the optimized model parameters and weighting coefficients.
2. The transformer life prediction method based on digital-analog linkage and gradient descent according to claim 1 is characterized in that: The transformer historical operation data set includes the following primary data sets, gas data set, insulating oil data set, and electrical data set; the gas data set includes the following secondary data sets: hydrogen content, methane content, acetylene content, ethylene content, ethane content, and carbon oxide content; the insulating oil data set includes the following secondary data sets: insulating oil dielectric loss, insulation resistance absorption ratio, oil breakdown voltage, water content in oil, gas content in oil, and furfural content; the electrical data set includes the following secondary data sets: insulation resistance value, partial discharge amount, dielectric loss, leakage resistance value, insulation resistance absorption ratio, and volume resistivity.
3. The transformer life prediction method based on digital-analog linkage and gradient descent according to claim 2 is characterized in that: The data set normalization preprocessing method comprises: Let x i,j (t) is the j{1≤j≤S}th secondary historical data set of the i{1≤i≤N}th transformer in the distribution network transformer data set at t(t=0,1,2,…,K i ) time, the performance degradation history data of the transformer after normalization x' i.j (t) is as follows: Among them, min(x i,j )、max(x i,j ) are the minimum and maximum values of the performance degradation historical data in the j-th secondary historical data set of the ith distribution network transformer, respectively.
4. The transformer life prediction method based on digital-analog linkage and gradient descent according to claim 3 is characterized in that: In step 2, the composite health index Z after integrating multiple secondary data sets is obtained based on the historical data set of the distribution network transformer. i (t) is as follows: where a j is the weight of the j-th indicator.
5. The transformer life prediction method based on digital-analog linkage and gradient descent according to claim 4 is characterized in that: The random degradation function of the degradation model is established, and the nonlinear Wiener process modeling and transformer composite health index Z are considered based on the standard Brownian motion. i The degradation process of (t) changing with time, the random degradation function of the composite health index is constructed including: Among them, B(t) is the standard Brownian motion, θ i , β i , σ i is the model parameter, and S i (t) represents the logarithmic transformation of the i-th random degradation system at time t, and its formula is: Let θ' i =lnθ i , The formula is: S i (t)=θ′ i +b′ i t+s i B(t).
6. The transformer life prediction method based on digital-analog linkage and gradient descent according to claim 5 is characterized in that: In step 4, the life value T of the distribution network transformer random degradation is defined i : T i =inf{t i :S i (t i )≥b|s i,0 <b} s i,0 =lnz i,0 where t i The historical data set time recorded for the i-th transformer, s i,0 is the natural logarithm of the initial composite health index of the i-th transformer, and b is the critical threshold corresponding to the composite health index; Define the remaining life value L of the distribution network transformer with random degradation i，k : L i,k inf{l i,k :S i (t i,k +l i,k )≥b|s i,k <b} s i,k =lnz i,k where s i,k is the natural logarithm of the composite health index of the i-th transformer at time k, t i,k is the life value of the i-th transformer at time k, l i,k is the remaining life value of the i-th transformer at time k, and b is the critical threshold corresponding to the natural logarithm of the composite health index; Calculating the expected life expectancy includes, Among them, t i,0 The initial time of the i-th transformer, t i,Ki The final moment of the i-th transformer, s i,Ki is the natural logarithm of the transformer composite health index of the i-th transformer at the final moment; The expected prediction for calculating the remaining life of distribution network transformers includes:
7. The transformer life prediction method based on digital-analog linkage and gradient descent according to claim 6 is characterized in that: The mean square error objective optimization function based on the predicted remaining life and the actual remaining life includes: Where A is the indicator weight a j A collection of; and l i,k is the predicted remaining life value and actual remaining life value of the i-th transformer at time k.
8. The transformer life prediction method based on digital-analog linkage and gradient descent according to claim 1 is characterized in that: The gradient descent algorithm is used to optimize the target optimization function, and the reverse optimization adjustment of the data set weight coefficient and model parameters includes: Where A' is the optimal indicator weight set, b' is the optimal critical threshold; argmin represents the parameter value at which the function achieves the minimum value in its domain.
9. The transformer life prediction method based on digital-analog linkage and gradient descent according to claim 8 is characterized in that: The gradient descent algorithm is used to optimize the target optimization function, and its parameter update process is expressed in an iterative way as follows: Where Υ represents the parameters (A, b) to be updated, η represents the learning rate, and r represents the iteration step. The iteration process will stop when the parameters converge or the number of iterations reaches a predetermined value.
10. The transformer life prediction method based on digital-analog linkage and gradient descent according to claim 1 is characterized in that: In step 7, the logarithmic transformation of the i-th random degradation system at time t is updated to S′ i (t), The predicted expected update of the remaining life of the distribution network transformer is: