Transformer temperature rise model non-parameter estimation method and device based on intelligent fusion terminal

Abstract

The invention relates to the field of state estimation of transformers, and particularly discloses a transformer temperature rise model non-parameter estimation method and device based on an intelligent fusion terminal, and the method comprises the steps: collecting the historical operation condition data of the intelligent fusion terminal; combining historical operation condition data to form a time sequence transformer operation state data set; performing non-parametric independence screening on characteristic attributes of the operation state data of the transformer, and excluding operation data which is not strongly associated with the temperature rise of the transformer; a Group Lasso non-parametric penalty regression algorithm is adopted to realize online estimation rule extraction of the temperature of the transformer; and the model prediction accuracy is checked, and the feature importance is regularly sorted again so as to adapt to the complex operation environment and variable operation conditions of the transformer. The transformer hot-spot temperature online estimation method has the advantages that intelligent fusion terminal equipment and a non-parameter estimation method are combined, and compared with a conventional estimation method, the transformer hot-spot temperature online estimation method has better interpretability, so that the transformer hot-spot temperature can be quickly estimated online.

Description

Technical Field

[0001] The present invention relates to the field of transformer state estimation, and in particular to a transformer temperature rise model non-parametric estimation method and device based on an intelligent fusion terminal. Background Art

[0002] With the development of my country's economic strength, the scale of the power grid continues to increase. In the energy consumption of terminals, the proportion of electric energy is increasing, and the growth rate of power load has generally remained relatively rapid. my country's peak load level continues to rise, and the imbalance between supply and demand during peak loads continues to worsen. The gradual growth of electric vehicles and distributed photovoltaics are likely to cause heavy overload and reverse overload of distribution transformers, thereby affecting their penetration rate. At the same time, if a heavy overload occurs in the distribution transformer area, it will cause problems such as accelerated aging and burning of the transformer, which will have an impact on the safe operation of the distribution network.

[0003] Distribution transformers are one of the most expensive assets in the grid infrastructure. If the temperature of the transformer exceeds the nameplate limit, it increases the leakage flux in the core and the outside, which heats the metal parts of the transformer. This may further affect the thermodynamics inside the transformer because, in this case, the insulating oil composition changes, resulting in an increase in gas content. As the transformer losses (i.e. coil windings, core and impurity losses) increase, the temperature rises. If the hottest areas of the transformer and the oil temperature exceed the dynamic thermal limit of the transformer itself, it will increase the aging rate of the insulation layer and shorten the transformer life, thus endangering the reliability of the entire network. Therefore, new methods are needed to sense the temperature of the transformer in time and thus limit its aging to reduce the need for expensive network upgrades.

[0004] At present, there are mainly the following prediction methods for the current-carrying temperature rise model of distribution transformers. Some scholars have developed a dynamic rating algorithm to determine the real-time dynamic thermal limit of the transformer by evaluating the hot spot temperature of the transformer and measuring the top oil temperature, tap position and cooling operation when the insulation life is shortened. On this basis, some scholars have considered the dependence of oil viscosity and winding loss on temperature and improved the model, thereby improving the accuracy of short-term dynamic load temperature estimation. Because the above model is too complicated, some scholars have proposed a simplified hot spot temperature calculation regression model based on experimental measurement of optical fiber temperature sensor. This method can reduce the error rate without increasing the input data. For the harmonic temperature rise caused by nonlinear loads, some scholars have proposed an unbalanced load model considering the harmonic distortion of different phases and integrated it into the dynamic transformer rating algorithm. In terms of prediction of dynamic applications, some scholars have developed a transformer hot spot temperature prediction model based on artificial neural network, some scholars have proposed a method for predicting transformer hot spot temperature under limited information, and some scholars have established a transformer winding hot spot temperature fluctuation prediction model based on fuzzy information granulation and chaotic particle swarm optimization wavelet neural network.

[0005] In the prior art, transformer temperature rise models are mostly based on physical laws, such as the IEEE standard model, but these models have many assumptions and are difficult to accurately describe the complex and dynamic actual operating environment. In addition, these models often rely on fixed parameters and are difficult to adapt to the dynamic changes of transformer load and environmental conditions. Therefore, there is an urgent need for a data-driven non-parametric estimation method to improve the accuracy and adaptability of the transformer current-carrying temperature rise model. In addition, most models focus on academic research and rarely consider engineering practice and corresponding hardware devices. Summary of the invention

[0006] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a non-parametric estimation method and device for a transformer temperature rise model based on an intelligent fusion terminal.

[0007] The object of the present invention is achieved by the following technical solution: a non-parametric estimation method of transformer temperature rise model based on an intelligent fusion terminal, the method comprising the following steps:

[0008] Step S1, collecting historical operating condition data of the intelligent fusion terminal;

[0009] Step S2, combining historical operating condition data to form a sequential transformer operating status data set;

[0010] Step S3, non-parametric independence screening of characteristic attributes of transformer operation status data, and excluding operation data that is not strongly correlated with transformer temperature rise;

[0011] Step S4, using the Group Lasso non-parametric penalized regression algorithm to build a transformer current-carrying temperature rise prediction model to achieve online transformer temperature estimation rule extraction;

[0012] Step S5: test the prediction accuracy of the transformer current-carrying temperature rise prediction model, and regularly re-rank the feature importance to adapt to the complex operating environment and variable operating conditions of the transformer.

[0013] Specifically, in step S1, the transformer voltage, current, oil temperature operating condition data of the oil-immersed transformer, and distribution box temperature and distribution box humidity environment data are collected through the intelligent fusion terminal installed on the secondary side of the transformer.

[0014] Specifically, the specific steps of step S2 are:

[0015] Step S21, constructing a time series data set;

[0016] Step S22: Data cleaning: for missing values ​​x in the collected data i, for data with time series properties, linear interpolation is used to process the data. The specific formula is as follows:

[0017]

[0018] In the formula, x j is the value of other valid samples under the same feature, and n is the number of valid samples;

[0019] For data that do not have the nature of time series, the mean filling method is used to process the data. The specific formula is as follows:

[0020]

[0021] In the formula, x i-1 and x i+1 are the known data points on both sides of the missing value;

[0022] Step S23, abnormal data detection and processing, using the interquartile range Q1 and Q3 to define the upper and lower limits of the abnormal value, the specific formula is as follows:

[0023] lb=Q1-1.5×(Q3-Q1)

[0024] ub=Q3+1.5×(Q3-Q1)

[0025] In the formula, lb is the lower limit of the outlier value, ub is the upper limit of the outlier value, and data points outside this range are outliers.

[0026] Specifically, the specific steps of step S3 are:

[0027] Step S31: Transformer operating condition characteristic attribute and environmental characteristic attribute X j Perform BS expansion and take X j The i-th sample x ij For example, its B-spline expansion form is:

[0028]

[0029] Where: a ij1 ,a ij2 ,…,a ijn Represents x ij The regression coefficients of each item after BS expansion are obtained by the least squares method; b j4 ,b j5 ,…,b jn Indicates the interpolation node, in X j The value range of f is determined by a fixed ratio; n represents the degree of freedom, which determines f Bij (x ij ) contains the number of items, n is at least 3;

[0030] Step S32: Use cross validation to obtain the optimal degree of freedom and extract the fitting error ε corresponding to each degree of freedom i ;

[0031] Step S33: fitting error ε i Sort, corresponding to ε i The smallest one is the optimal degree of freedom of the regression function of this attribute;

[0032] Step S34: BS expand each feature attribute according to its optimal degree of freedom. The simplified form of the expansion is:

[0033] f Bij (x ij )=φ Bij1 (x ij1 )+φ Bij2 (x ij2 )+φ Bij3 (x ij3 )+φ Bijn (x ij )

[0034] Step S35: Use all feature attributes to perform regression fitting on each target attribute and calculate its corresponding ε i The values ​​are sorted from high to low, with ε i The inflection point where the value drops significantly is the dividing line. The feature attributes before the dividing line have a weak correlation with the target attribute under investigation and need to be eliminated. The feature attributes after the dividing line are retained as input samples for subsequent steps.

[0035] Step S36: Perform unified centralization processing on the filtered feature attributes. The specific formula is:

[0036]

[0037] Where: φ Bijgmean Represents φ Bijg (x ijg ) mean; sd(φ Bijg ) represents φ Bijg (x ijg ) standard deviation; the obtained φ ijg (x ijg ) is the expanded item of characteristic attributes after centralization;

[0038] Step S37: Obtain the sample after centralization. The specific formula is:

[0039]

[0040] Where, d f represents the optimal degrees of freedom.

[0041] Specifically, the specific steps of step S32 are:

[0042] Step S321, set the degree of freedom value interval [3, n], and initialize the degree of freedom d = 3;

[0043] Step S322, divide all samples into m groups in order, and initialize i=1;

[0044] Step S323, taking the degree of freedom as d, and using m-1 groups of samples other than the i-th group as training sets to perform nonparametric regression;

[0045] Step S324: Taking the i-th group of samples as the test set, calculate the fitting error ε of the regression function i ;

[0046] Step S325: if i=m, obtain m fitting errors ε when the degree of freedom is d i The mean of the fitting error ε with degrees of freedom d i ; If i≠m, i is incremented by 1 and the process returns to step S323;

[0047] Step S326: If d=n, the calculation ends and the fitting error ε corresponding to each degree of freedom is extracted. i ; If d≠n, d is incremented by 1 and returns to step S322.

[0048] Specifically, the specific steps of step S4 are:

[0049] Step S41, randomly dividing the data set into two groups according to 70% training set and 30% test set, the training set data packet is used for parameter optimization and establishing the transformer current-carrying temperature rise prediction model, and the test set data is used for testing the transformer current-carrying temperature rise prediction model;

[0050] Step S42: The idea of ​​solving the association expression by Group Lasso is expressed as an optimization problem:

[0051] minL(μ,β)

[0052] st

[0053]

[0054]

[0055] In the formula, μ is the constant term to be determined; λ is the penalty factor; N is the number of samples; p is the number of cubic B-spline functions; T n is the target attribute of the nth sample; ω i is the weight of the ith feature. In the absence of prior knowledge, all feature weights are set to 1; β is the coefficient set to be determined; d iis the degree of freedom of the spline function; β i is the coefficient set of the i-th added term, and there is

[0056] β=[β1,β2,…,β p ],β i =[β i1 ,β i2 ,…,β i,di ] ↓

[0057] Step S43: gradually increase the penalty factor λ to further eliminate weakly correlated features and set the coefficient set β i The solution is 0;

[0058] Step S44: Successively test the optimal penalty factor, and determine the optimal penalty factor according to the principle of minimizing the prediction error, and its value range is {λ|λ=0.5z,1≤z≤20,z∈Z}

[0059] Step S44, obtaining a nonlinear mathematical relationship with the hot spot temperature of the transformer as the dependent variable and the filtered characteristic attribute as the independent variable:

[0060]

[0061] The relational expression is used to quickly estimate the hot spot temperature of the transformer in the value space composed of all characteristic attributes.

[0062] Specifically, the specific steps of step S5 are:

[0063] Step S51, using mean square error MSE and mean absolute percentage error MAPE as evaluation indicators of transformer current-carrying temperature rise prediction model, MSE is an indicator used to measure the difference between the actual observation value and the model estimated value, MAPE is used to measure whether the prediction model is biased, which can accurately reflect the size of the actual estimation error, and the calculation formulas of the two are as follows:

[0064]

[0065]

[0066] In the formula, represents the model estimate, which is the predicted value of the transformer temperature, y i represents the actual measured value, which is the measured value of the transformer temperature, and n is the number of test set samples.

[0067] Step S52: Model testing: The model is tested by calculating two statistics: the overall relative error Rs and the absolute average relative error E:

[0068]

[0069]

[0070] Step S53: Feature importance sorting: Group Lasso outputs the importance of variables during training.

[0071] Specifically, the formula for the feature importance in step S53 is as follows:

[0072]

[0073] In the formula, y* is the sample prediction value before replacement, y is the actual value, is the predicted value after replacement, oob is the number of samples, and m is the maximum absolute value of the label value in all training samples.

[0074] A non-parametric estimation device for transformer temperature rise model based on an intelligent fusion terminal includes:

[0075] A data acquisition and processing module, which is connected to the intelligent fusion terminal and is used to monitor the load current, ambient temperature, transformer load rate and operating time in real time, and to collect data for preprocessing;

[0076] Feature grouping module, the processed data is sent to the feature grouping module, and the load-related features, environment-related features and operating status feature attributes are screened according to the non-parametric independence screening method, leaving the feature attributes with a greater correlation with the transformer current-carrying temperature rise;

[0077] Model training module, the filtered feature attributes are sent to the model training module for sparse regression training;

[0078] Online estimation module,The trained model is deployed in the online estimation module to generate real-time predictions of hot spot temperatures of transformers and send alerts to the monitoring center;

[0079] The detection and update module tests the accuracy of the model with MSE and MAPE, and updates the model by reordering the importance of feature attributes:

[0080]

[0081]

[0082] In the formula, represents the model estimate, which is the predicted value of the transformer temperature, y i represents the actual measured value, which is the measured value of the transformer temperature, and n is the number of test set samples.

[0083] An electronic device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a non-parametric estimation method of a transformer temperature rise model based on an intelligent fusion terminal are implemented.

[0084] A non-transitory computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a non-parametric estimation method for a transformer temperature rise model based on an intelligent fusion terminal.

[0085] The present invention has the following advantages:

[0086] The present invention automatically selects the most important feature group for transformer temperature rise prediction through the Group Lasso method, thereby improving the explanatory power and accuracy of the model. In addition, the model adopted by the present invention does not need to assume a fixed physical model or parameters. The model can be dynamically adjusted according to real-time data to adapt to the complex operating environment of the transformer and has strong adaptability. In addition, the present invention combines the sparse characteristics of the intelligent fusion terminal and the Group Lasso, and can quickly process large-scale data and perform real-time online temperature rise prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 It is a flow chart of a non-parametric estimation method of a current-carrying temperature rise model of a distribution transformer based on an intelligent fusion terminal of the present invention;

[0088] Figure 2 The Group-Lasso model online estimation and update flow chart of the present invention;

[0089] Figure 3 It is a schematic diagram of a non-parametric estimation device of a transformer current-carrying temperature rise model based on an intelligent fusion terminal of the present invention. DETAILED DESCRIPTION

[0090] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and 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, that is, the embodiments described are only part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0091] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0092] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0093] The present invention is further described below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited to the following description. Figures 1 to 3 As shown, a non-parametric estimation method of transformer temperature rise model based on intelligent fusion terminal is characterized in that the method comprises the following steps:

[0094] Step S1, collecting historical operating condition data of the intelligent fusion terminal;

[0095] Step S2, combining historical operating condition data to form a sequential transformer operating status data set;

[0096] Step S3, non-parametric independence screening of characteristic attributes of transformer operation status data, and excluding operation data that is not strongly correlated with transformer temperature rise;

[0097] Step S4, using the Group Lasso non-parametric penalized regression algorithm to construct a transformer non-parametric estimation temperature rise model prediction, and realize the extraction of transformer temperature online estimation rules;

[0098] Step S5: test the prediction accuracy of the transformer non-parametric estimation temperature rise model, and regularly re-rank the feature importance to adapt to the complex operating environment and changeable operating conditions of the transformer.

[0099] Furthermore, in step S1, the transformer voltage, current, oil temperature of the oil-immersed transformer or winding temperature of the dry-type transformer operating condition data and distribution box temperature, distribution box humidity environmental data are collected through the intelligent fusion terminal installed on the secondary side of the transformer. In order to improve the accuracy of diagnosis, the data should be collected regularly at certain intervals and saved in the database for subsequent analysis.

[0100] Further, the transformer historical operating condition data and historical environmental data collected by the intelligent fusion terminal are combined to form a time series transformer operation data set, and the collected multidimensional data is preprocessed. The specific steps of step S2 are:

[0101] Step S21, constructing a time series data set; combining the historical operating condition data of the transformer collected by the intelligent fusion terminal, such as load current, operating time, transformer load rate, etc., with historical environmental data, such as ambient temperature, humidity, etc., sorting these data in time series to construct a time series transformer operation data set;

[0102] Step S22: Data cleaning: for missing values ​​x in the collected data i , for data with time series properties, linear interpolation is used to process the data. The specific formula is as follows:

[0103]

[0104] In the formula, x j is the value of other valid samples under the same feature, and n is the number of valid samples;

[0105] For data that do not have the nature of time series, the mean filling method is used to process the data. The specific formula is as follows:

[0106]

[0107] In the formula, x i-1 and x i+1 are the known data points on both sides of the missing value;

[0108] Step S23, abnormal data detection and processing, using the interquartile range Q1 and Q3 to define the upper and lower limits of the abnormal value, the specific formula is as follows:

[0109] lb=Q1-1.5×(Q3-Q1)

[0110] ub=Q3+1.5×(Q3-Q1)

[0111] In the formula, lb is the lower limit of the outlier value, ub is the upper limit of the outlier value, and data points outside this range are outliers.

[0112] Furthermore, since the data dimension involved in the transformer current-carrying temperature rise model is relatively large, a non-parametric independence screening method is used to first exclude characteristic attributes that are not strongly correlated with the transformer temperature rise. The specific steps of step S3 are:

[0113] Step S31: Transformer operating condition characteristic attribute and environmental characteristic attribute X j Perform BS expansion and take X j The i-th sample xij For example, its B-spline expansion form is:

[0114]

[0115] Where: a ij1 ,a ij2 ,…,a ijn Represents x ij The regression coefficients of each item after BS expansion are obtained by the least squares method; b j4 ,b j5 ,…,b jn Indicates the interpolation node, in X j The value range of f is determined by a fixed ratio; n represents the degree of freedom, which determines f Bij (x ij ) contains the number of items, n is at least 3;

[0116] Step S32: Use cross validation to obtain the optimal degree of freedom and extract the fitting error ε corresponding to each degree of freedom i ;

[0117] Step S33: fitting error ε i Sort, corresponding to ε i The smallest one is the optimal degree of freedom of the regression function of this attribute;

[0118] Step S34: BS expand each feature attribute according to its optimal degree of freedom. The simplified form of the expansion is:

[0119] f Bij (x ij )=φ Bij1 (x ij1 )+φ Bij2 (x ij2 )+φ Bij3 (x ij3 )+φ Bijn (x ij )

[0120] Step S35: Use all feature attributes to perform regression fitting on each target attribute and calculate its corresponding ε i The values ​​are sorted from high to low, with ε i The inflection point where the value drops significantly is the dividing line. The feature attributes before the dividing line have a weak correlation with the target attribute under investigation and need to be eliminated. The feature attributes after the dividing line are retained as input samples for subsequent steps.

[0121] Step S36: The selected characteristic attributes have different influences on the transformer temperature rise during actual operation of the transformer. In order to avoid distorting the actual influence of the characteristic attributes, the selected characteristic attributes are centrally processed uniformly. The specific formula is:

[0122]

[0123] Where: φ Bijgmean Represents φ Bijg (x ijg ) mean; sd(φ Bijg ) represents φ Bijg (x ijg ) standard deviation; the obtained φ ijg (x ijg ) is the expanded item of characteristic attributes after centralization;

[0124] Step S37: Obtain the sample after centralization. The specific formula is:

[0125]

[0126] Where, d f represents the optimal degrees of freedom.

[0127] Furthermore, the specific steps of step S32 are:

[0128] Step S321, set the degree of freedom value interval [3, n], and initialize the degree of freedom d = 3;

[0129] Step S322, divide all samples into m groups in order, and initialize i=1;

[0130] Step S323, taking the degree of freedom as d, and using m-1 groups of samples other than the i-th group as training sets to perform nonparametric regression;

[0131] Step S324: Taking the i-th group of samples as the test set, calculate the fitting error ε of the regression function i ;

[0132] Step S325: if i=m, obtain m fitting errors ε when the degree of freedom is d i The mean of the fitting error ε with degrees of freedom d i ; If i≠m, i is incremented by 1 and the process returns to step S323;

[0133] Step S326: If d=n, the calculation ends and the fitting error ε corresponding to each degree of freedom is extracted. i ; If d≠n, d is incremented by 1 and returns to step S322.

[0134] Furthermore, the specific steps of step S4 are:

[0135] Step S41, randomly dividing the data set into two groups according to 70% training set and 30% test set, the training set data packet is used for parameter optimization and establishing the transformer current-carrying temperature rise prediction model, and the test set data is used for testing the transformer current-carrying temperature rise prediction model;

[0136] Step S42: The idea of ​​solving the association expression by Group Lasso is expressed as an optimization problem:

[0137] minL(μ,β)

[0138] st

[0139]

[0140]

[0141] In the formula, μ is the constant term to be determined; λ is the penalty factor; N is the number of samples; p is the number of cubic B-spline functions; T n is the target attribute of the nth sample; ω i is the weight of the ith feature. In the absence of prior knowledge, all feature weights are set to 1; β is the coefficient set to be determined; d i is the degree of freedom of the spline function; β i is the coefficient set of the i-th added term, and there is

[0142]

[0143] Step S43: gradually increase the penalty factor λ to further eliminate weakly correlated features and set the coefficient set β i The solution is 0;

[0144] Step S44: Successively test the optimal penalty factor, and determine the optimal penalty factor according to the principle of minimizing the prediction error, and its value range is {λ|λ=0.5z,1≤z≤20,z∈Z}

[0145] Step S44, obtaining a nonlinear mathematical relationship with the hot spot temperature of the transformer as the dependent variable and the filtered characteristic attribute as the independent variable, that is, a transformer current-carrying temperature rise prediction model:

[0146]

[0147] The relational expression is used to quickly estimate the hot spot temperature of the transformer in the value space composed of all characteristic attributes.

[0148] Furthermore, the transformer current-carrying temperature rise prediction model has a high dimension of input variables, and the model prediction accuracy needs to be tested, and the feature importance needs to be re-ranked regularly to adapt to the complex operating environment and variable operating conditions of the transformer. The specific steps of step S5 are:

[0149] Step S51, using mean square error MSE and mean absolute percentage error MAPE as evaluation indicators of transformer current-carrying temperature rise prediction model, MSE is an indicator used to measure the difference between the actual observation value and the model estimated value, MAPE is used to measure whether the prediction model is biased, which can accurately reflect the size of the actual estimation error, and the calculation formulas of the two are as follows:

[0150]

[0151]

[0152] In the formula, represents the model estimate, which is the predicted value of the transformer temperature, y i Represents the actual measured value, which is the measured value of the transformer temperature. n is the number of test set samples. The smaller the MSE, the better the model effect and the higher the prediction accuracy. A MAPE value less than 10% indicates that the model prediction accuracy is high; if it is greater than 50%, the prediction is wrong.

[0153] Step S52: Model testing: The model is tested by calculating two statistics: the overall relative error Rs and the absolute average relative error E:

[0154]

[0155]

[0156] Step S53, feature importance ranking. Group Lasso outputs the importance of variables during training, that is, which feature is more useful. There are two methods to implement it: Gini method and permutation method. The permutation method is to change the impact of feature observation on the result. If the feature is important, then the result should be sensitive to it. Feature importance plays an important role in predictive modeling. It can help understand the data set and model, perform feature selection, and even improve the predictive model, thereby improving the efficiency and effectiveness of the predictive model. The feature importance formula for regression problems is as follows:

[0157]

[0158] In the formula, y* is the sample prediction value before replacement, y is the actual value, is the predicted value after replacement, oob is the number of samples, and m is the maximum absolute value of the label value in all training samples.

[0159] A non-parametric estimation device for transformer temperature rise model based on an intelligent fusion terminal includes:

[0160] A data acquisition and processing module is installed in the cabinet of the intelligent fusion terminal to facilitate quick access to the data stream of the terminal. The data acquisition and processing module is connected to the intelligent fusion terminal and is used to monitor the load current, ambient temperature, transformer load rate and operating time in real time, and collect data to pre-process the data;

[0161] Feature grouping module, the processed data is sent to the feature grouping module, and the load-related features, environment-related features and operating status feature attributes are screened according to the non-parametric independence screening method, leaving the feature attributes with a greater correlation with the transformer current-carrying temperature rise;

[0162] Model training module, the filtered feature attributes are sent to the model training module for sparse regression training;

[0163] Online estimation module,The trained model is deployed in the online estimation module to generate real-time predictions of hot spot temperatures of transformers and send alerts to the monitoring center;

[0164] The detection and update module tests the accuracy of the model with MSE and MAPE, and updates the model by reordering the importance of feature attributes:

[0165]

[0166]

[0167] In the formula, represents the model estimate, which is the predicted value of the transformer temperature, y i represents the actual measured value, which is the measured value of the transformer temperature, and n is the number of test set samples.

[0168] An electronic device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a non-parametric estimation method of a transformer temperature rise model based on an intelligent fusion terminal are implemented.

[0169] A non-transitory computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a non-parametric estimation method for a transformer temperature rise model based on an intelligent fusion terminal.

[0170] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any technician familiar with the art can make many possible changes and modifications to the technical solution of the present invention by using the above-mentioned technical content without departing from the scope of the technical solution of the present invention, or modify it into an equivalent embodiment of equivalent changes. Therefore, any changes, modifications, equivalent changes and modifications made to the above embodiments based on the technology of the present invention without departing from the content of the technical solution of the present invention belong to the protection scope of the present technical solution.

Claims

1. A non-parametric estimation method for transformer temperature rise model based on intelligent fusion terminal, characterized by: The method comprises the following steps: Step S1, collecting historical operating condition data of the intelligent fusion terminal; Step S2, combining historical operating condition data to form a sequential transformer operating status data set; Step S3: non-parametric independence screening of characteristic attributes of transformer operation status data to exclude operation data that is not strongly correlated with transformer temperature rise; Step S4, using the Group Lasso non-parametric penalized regression algorithm to build a transformer current-carrying temperature rise prediction model to achieve online transformer temperature estimation rule extraction; Step S5: test the prediction accuracy of the transformer current-carrying temperature rise prediction model, and regularly re-rank the feature importance to adapt to the complex operating environment and variable operating conditions of the transformer.

2. The non-parametric estimation method of transformer temperature rise model based on intelligent fusion terminal according to claim 1 is characterized in that: In step S1, the transformer voltage, current, oil temperature operating condition data of the oil-immersed transformer, and distribution box temperature and distribution box humidity environment data are collected through the intelligent fusion terminal installed on the secondary side of the transformer.

3. The non-parametric estimation method of transformer temperature rise model based on intelligent fusion terminal according to claim 1 is characterized in that: The specific steps of step S2 are: Step S21, constructing a time series data set; Step S22: Data cleaning: for missing values ​​x in the collected data i , for data with time series properties, linear interpolation is used to process the data. The specific formula is as follows: In the formula, x j is the value of other valid samples under the same feature, and n is the number of valid samples; For data that do not have the nature of time series, the mean filling method is used to process the data. The specific formula is as follows: In the formula, x i-1 and x i+1 are the known data points on both sides of the missing value; Step S23, abnormal data detection and processing, using the interquartile range Q1 and Q3 to define the upper and lower limits of the abnormal value, the specific formula is as follows: lb=Q1-1.5×(Q3-Q1) ub=Q3+1.5×(Q3-Q1) In the formula, lb is the lower limit of the outlier value, ub is the upper limit of the outlier value, and data points outside this range are outliers.

4. The non-parametric estimation method of transformer temperature rise model based on intelligent fusion terminal according to claim 1 is characterized in that: The specific steps of step S3 are: Step S31: Transformer operating condition characteristic attribute and environmental characteristic attribute X j Perform BS expansion and take X j The i-th sample x ij For example, its B-spline expansion form is: Where: a ij1 ,a ij2 ,…,a ijn Represents x ij The regression coefficients of each item before BS expansion are obtained by the least squares method; bj4,b j5 ,…,b jn Indicates the interpolation node, in X j The value range of f is determined by a fixed ratio; n represents the degree of freedom, which determines f Bij (x ij ) contains the number of items, n is at least 3; Step S32: Use cross validation to obtain the optimal degree of freedom and extract the fitting error ε corresponding to each degree of freedom i ; Step S33: fitting error ε i Sort, corresponding to ε i The smallest one is the optimal degree of freedom of the regression function of this attribute; Step S34: BS expand each feature attribute according to its optimal degree of freedom. The simplified form of the expansion is: f Bij (x ij )=φ Bij1 (x ij1 )+φ Bij2 (x ij2 )+φ Bij3 (x ij3 )+φ Bijn (x ij ) Step S35: Use all feature attributes to perform regression fitting on each target attribute and calculate its corresponding ε i The values ​​are sorted from high to low, with ε i The inflection point where the value drops significantly is the dividing line. The feature attributes before the dividing line have a weak correlation with the target attribute under investigation and need to be eliminated. The feature attributes after the dividing line are retained as input samples for subsequent steps. Step S36: Perform unified centralization processing on the filtered feature attributes. The specific formula is: Where: φ Bijgmean Represents φ Bijg (x ijg ) mean; sd(φ Bijg ) represents φ Bijg (x ijg ) standard deviation; the obtained φ ijg (x ijg ) is the expanded item of characteristic attributes after centralization; Step S37: Obtain the sample after centralization. The specific formula is: Where, d f represents the optimal degrees of freedom.

5. The non-parametric estimation method of transformer temperature rise model based on intelligent fusion terminal according to claim 4 is characterized in that: The specific steps of step S32 are: Step S321, set the degree of freedom value interval [3, n], and initialize the degree of freedom d = 3; Step S322, divide all samples into m groups in order, and initialize i=1; Step S323, taking the degree of freedom as d, and using m-1 groups of samples other than the i-th group as training sets to perform nonparametric regression; Step S324: Taking the i-th group of samples as the test set, calculate the fitting error ε of the regression function i ; Step S325: if i=m, obtain m fitting errors ε when the degree of freedom is d i The mean of the fitting error ε with degrees of freedom d i ; If i≠m, i is incremented by 1 and the process returns to step S323; Step S326: If d=n, the calculation ends and the fitting error ε corresponding to each degree of freedom is extracted. i ; If d≠n, d is incremented by 1 and returns to step S322.

6. The non-parametric estimation method of transformer temperature rise model based on intelligent fusion terminal according to claim 4 is characterized in that: The specific steps of step S4 are: Step S41, randomly divide the data set into two groups according to 70% training set and 30% test set, the training set data packet is used for Group Lasso parameter optimization and establishing a transformer current-carrying temperature rise prediction model through Group Lasso, and the test set data is used to test the transformer current-carrying temperature rise prediction model obtained by non-parametric estimation; Step S42: Express the idea of ​​solving the association expression by Group Lasso as an optimization problem: minL(μ,β) st In the formula, μ is the constant term to be determined; λ is the penalty factor; N is the number of samples; p is the number of cubic B-spline functions; T n is the target attribute of the nth sample; ω i is the weight of the ith feature. In the absence of prior knowledge, all feature weights are set to 1; β is the coefficient set to be determined; d i is the degree of freedom of the spline function; β i is the coefficient set of the i-th added term, and there is Step S43: gradually increase the penalty factor λ to further eliminate weakly associated features and set the coefficient set β i The solution is 0; Step S44, test the optimal penalty factor one by one, and determine the optimal penalty factor according to the principle of minimizing the prediction error, and the value range of the optimal penalty factor is {λ|λ=0.5z,1≤z≤20,z∈Z} Step S44, obtaining a nonlinear mathematical relationship with the hot spot temperature of the transformer as the dependent variable and the filtered characteristic attribute as the independent variable: The relational expression is used to quickly estimate the hot spot temperature of the transformer in the value space composed of all characteristic attributes.

7. The non-parametric estimation method of transformer temperature rise model based on intelligent fusion terminal according to claim 4 is characterized in that: The specific steps of step S5 are: Step S51, using mean square error MSE and mean absolute percentage error MAPE as evaluation indicators of transformer current-carrying temperature rise prediction model, MSE is an indicator used to measure the difference between actual observation value and model estimated value, MAPE is used to measure whether the prediction model is biased, and the calculation formulas of the two are as follows: In the formula, represents the model estimate, which is the predicted value of the transformer temperature, y i represents the actual measured value, which is the measured value of the transformer temperature, and n is the number of test set samples. Step S52: Model testing: The model is tested by calculating two statistics: the overall relative error Rs and the absolute average relative error E: Step S53: Feature importance sorting: Group Lasso outputs the importance of variables during training.

8. The non-parametric estimation method of transformer temperature rise model based on intelligent fusion terminal according to claim 7 is characterized in that: The formula of feature importance in step S53 is as follows: In the formula, y* is the sample prediction value before replacement, y is the actual value, is the predicted value after replacement, oob is the number of samples, and m is the maximum absolute value of the label value in all training samples.

9. A non-parametric estimation device for transformer temperature rise model based on an intelligent fusion terminal, characterized in that: include: A data acquisition and processing module, which is connected to the intelligent fusion terminal and is used to monitor the load current, ambient temperature, transformer load rate and operating time in real time, and to collect data for preprocessing; Feature grouping module, the processed data is sent to the feature grouping module, and the load-related features, environment-related features and operating status feature attributes are screened according to the non-parametric independence screening method, leaving the feature attributes with a greater correlation with the transformer current-carrying temperature rise; Model training module, the filtered feature attributes are sent to the model training module for sparse regression training; Online estimation module,The trained model is deployed in the online estimation module to generate real-time predictions of hot spot temperatures of transformers and send alerts to the monitoring center; The detection and update module tests the accuracy of the model with MSE and MAPE, and updates the model by reordering the importance of feature attributes: In the formula, represents the model estimate, which is the predicted value of the transformer temperature, y i represents the actual measured value, which is the measured value of the transformer temperature, and n is the number of test set samples.

Images