Substation equipment full life cycle management method and system based on correlation analysis

Abstract

The present invention relates to a method and system for managing the whole life cycle of substation equipment based on correlation analysis, the method comprising: obtaining real-time state sampling time series data of substation equipment, screening out influencing factors as cleaning measurements; filling the cleaning measurements into a pre-set variable-length moving window, and outputting a variable-length moving window data matrix; using differential privacy technology to add noise, inputting into a recurrent neural network state prediction model based on a time series pattern attention mechanism, and outputting the real-time state warning prediction of substation equipment; judging whether the set warning value is reached, if so, calculating the correlation between the real-time state warning prediction and the corresponding influencing factors, screening out the fault factors, otherwise not doing anything; performing fault repair based on the fault factors, and repeating the above steps to achieve the whole life cycle management of substation equipment. Compared with the prior art, the present invention has the advantages of improving the intelligent management level of substation equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment life cycle management, and in particular to a method and system for substation equipment life cycle management based on correlation analysis. Background Art

[0002] Substation is an important part of the power system, and the state of its equipment directly affects the safety and stable operation of the power system. However, due to the wide variety of substation equipment and the complex working environment, the operating status of the equipment is difficult to grasp in real time. Therefore, the full life cycle management of equipment has become a severe challenge. Full life cycle management refers to the effective management and control of the entire process from equipment design, manufacturing, transportation, installation, operation, maintenance to scrapping to ensure the safe, reliable and economical operation of the equipment. This management mode can comprehensively and systematically understand and grasp the status of the equipment, prevent and diagnose equipment failures, extend equipment life, reduce equipment operating costs, and improve the operating efficiency of the power system. However, traditional equipment management methods often only focus on a certain stage of the equipment, ignoring the correlation between the various stages of the equipment, resulting in blind spots and inefficiency in equipment management. Therefore, the research and application of the full life cycle management of substation equipment has important practical significance and broad application prospects. Through full life cycle management, the healthy, intelligent and efficient operation of substation equipment can be achieved, providing guarantee for the safety and stability of the power system, and it is of great value to promote the sustainable development of the power industry.

[0003] Patent application CN113344504A discloses a full life cycle management method for substation equipment based on physical ID. This method has the disadvantages of managing equipment only by ID and being unable to deeply analyze the status of substation equipment. Existing intelligent substation equipment monitoring mainly monitors key quantity measurements through the threshold method, which lacks the ability to predict the risk of potential faulty equipment and the risk of failure during peak power consumption. Moreover, the threshold method does not consider the impact of other variables on key measurements, is easily affected by the external environment and the failure of the sampling equipment itself, and has poor robustness. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for managing the entire life cycle of substation equipment based on correlation analysis, which realizes the risk prediction and early warning capability of substation equipment to improve the intelligent management level of substation equipment.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A method for managing the entire life cycle of substation equipment based on correlation analysis includes the following steps:

[0007] S1. Acquire real-time state sampling time series data of substation equipment, and select influencing factors as cleaning measurements based on prior knowledge, wherein the real-time state sampling time series data includes electrical quantity measurement, non-electrical quantity measurement and external environmental variables of substation equipment;

[0008] S2, filling the cleaning measurement into a preset variable-length moving window, and outputting a variable-length moving window data matrix;

[0009] S3, based on the variable-length moving window data matrix, adding noise by using differential privacy technology, inputting the pre-built recurrent neural network state prediction model based on the time series pattern attention mechanism, and outputting the real-time state warning prediction of the substation equipment;

[0010] S4, judging whether the real-time status warning prediction amount reaches the set warning value, if so, calculating the correlation between the real-time status warning prediction amount and the corresponding influencing factor, and screening out the fault factor, if not, doing nothing;

[0011] S5. Perform fault repair based on the fault factor, and return to step S1 to implement full life cycle management of substation equipment.

[0012] Furthermore, the electrical quantity measurements include the load current, core grounding current, and high-frequency partial discharge of the equipment; the non-electrical quantity measurements include the main transformer infrared temperature, pipe mother joint temperature, oil temperature, body temperature, body humidity, and gas content in the oil of the equipment; and the external environmental variables include ambient temperature and ambient humidity.

[0013] Furthermore, the differential privacy technology is a differential privacy protection technology based on the Laplace mechanism, and the expression for adding noise using the differential privacy technology is:

[0014]

[0015] In the formula, Δf is the global sensitivity of f, f is the original data, and D 1 , D 2 are adjacent data points, f' is the data after differential privacy is performed on f, ε is the privacy budget, ranging from 0 to 1, and η is the noise, which obeys the Laplace distribution with parameter f / ε.

[0016] Furthermore, the recurrent neural network state prediction model based on the temporal pattern attention mechanism includes an LSTM network and a TPA attention mechanism network, and the steps of constructing the recurrent neural network state prediction model based on the temporal pattern attention mechanism include:

[0017] 1) Obtaining historical state sampling time series data of substation equipment within a set search time period as historical cleaning measurement, and setting corresponding online warning measurement, wherein the set search time period includes multiple lines of historical state sampling time series data, and the time span between each line is the same;

[0018] 2) Based on the historical cleaning measurement, a variable-length moving window data cleaning method is used to obtain a variable-length moving window data matrix, and the variable-length moving window data matrix is ​​used together with the online warning measurement as a training input of a recurrent neural network state prediction model based on a temporal pattern attention mechanism, wherein the online warning measurement is used as a label, and the format of the variable-length moving window data matrix is: cleaning measurement×moving window length×number of features;

[0019] 3) The LSTM network extracts features from the training input to obtain the hidden state h of each time step i ', and use the TPA attention mechanism network to calculate the attention weight of each time step to output the prediction result;

[0020] 4) Using the Adam optimizer to perform weight optimization to adjust the hyperparameters of the recurrent neural network state prediction model based on the temporal pattern attention mechanism, and record the current training round and loss value;

[0021] 5) Return to step 3) to perform iterative training until the preset conditions are met and the training is terminated, thereby obtaining a recurrent neural network state prediction model based on the temporal pattern attention mechanism.

[0022] Furthermore, the step of obtaining the historical cleaning measurement includes:

[0023] Get the timestamp of each row of historical state sampling time series data within the set time interval from top to bottom;

[0024] According to the timestamp, determine whether the time interval between two rows is less than the set time interval,

[0025] If so, it is determined that there is no sampling missing or interruption between the two rows, and the multiple rows of historical state sampling time series data within the set search time period are used as historical cleaning measurements.

[0026] If not, determine that there is no sampling missing or interruption between the two rows, select Skip, and continue to select downward until the time span between the two rows is the same, and use multiple rows of historical state sampling time series data within the set search time period as historical cleaning measurements.

[0027] Furthermore, the step of obtaining a variable-length moving window matrix also includes:

[0028] Determine whether the number of available sampling values ​​of the cleaning measurement in the variable-length moving window matrix is ​​less than a preset value. If so, determine that the cleaning measurement is missing, use interpolation to fill the gaps, and obtain a complete variable-length moving window matrix. If not, do nothing.

[0029] Furthermore, the interpolation method includes one of linear interpolation, polynomial interpolation, Lagrange interpolation, cubic spline interpolation and spline interpolation.

[0030] Further, in step 3), the step of outputting the prediction result includes the following:

[0031] The TPA attention mechanism network obtains the hidden state h of the training input at each time step i ', use the attention mechanism to calculate the attention weights of the training input at each time step:

[0032] e i =h i ' T tanh(W i h i '+b i )

[0033]

[0034] The weighted sum of the attention weights of each time step is used to obtain the hidden layer weight v at the tth moment t :

[0035]

[0036] Output the substation equipment status prediction results of the TPA attention mechanism network:

[0037] h t =W h (W h' h t '+W v v t )

[0038] In the formula, h i ' T h i ' is the transpose; tanh is the hyperbolic tangent function; W i and b i is the weight matrix; α i is the attention weight for each time step; e i is the correlation between the input element and the context at time step i, e m is the correlation degree when the time step is m, and the value range of m is [1, i]; h tis the substation equipment status prediction result at the tth moment; W h , W h' , W v is the weight matrix.

[0039] Furthermore, the Pearson correlation coefficient is used to calculate the correlation:

[0040]

[0041] In the formula, ρ X,Y is the Pearson correlation coefficient, cov represents the covariance, X, Y are the state warning prediction and the corresponding related state operation data, σ X , σ Y are the standard deviations of X and Y respectively, E represents the expectation, μ X , μ Y are the average values ​​of X and Y respectively.

[0042] This embodiment also provides a substation equipment full life cycle management system based on correlation analysis, including:

[0043] Data acquisition module: acquires real-time state sampling time series data of substation equipment, and selects influencing factors as cleaning measurements based on prior knowledge, wherein the real-time state sampling time series data includes electrical quantity measurement, non-electrical quantity measurement and external environmental variables of substation equipment;

[0044] Cleaning module: fills the cleaning measurement into a preset variable length moving window and outputs a variable length moving window data matrix;

[0045] Noise addition and prediction module: Based on the variable-length moving window data matrix, noise is added using differential privacy technology, input into a pre-built recurrent neural network state prediction model based on the time series pattern attention mechanism, and output the real-time state warning prediction of the substation equipment;

[0046] Judgment module: judge whether the real-time status warning prediction quantity reaches the set warning value, if so, calculate the correlation between the real-time status warning prediction quantity and the corresponding influencing factor, and screen out the fault factor, if not, do nothing;

[0047] Management module: perform fault inspection and repair based on the fault factors, and repeat the processing of the above modules to achieve full life cycle management of substation equipment.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) The present invention takes into account the potential failure risks of substation equipment, uses a variable-length moving window to perform data cleaning, and uses a model input formed by a differential privacy protection mechanism to perform state prediction and early warning through a recurrent neural network model based on a temporal pattern attention mechanism, thereby realizing the risk prediction and early warning capabilities of substation equipment, and through correlation analysis, analyzes the potential failure factors of substation equipment, and then performs maintenance, which is repeated in turn, thereby improving the intelligent management level of substation equipment.

[0050] (2) The present invention uses differential privacy technology to add noise to synthesize new data sets to replace previous data sets. These synthesized new data sets are similar to the original data sets in statistical characteristics, but do not contain individual information of the original data, and can protect data privacy.

[0051] (3) The present invention proposes a set of efficient, robust, and versatile variable-length moving window data cleaning methods to process substation equipment status sampling, which solves the irregularity, discontinuity, non-alignment, and granularity unevenness of substation equipment status sampling as spatiotemporal big data.

[0052] (4) The present invention uses a recurrent neural network model based on the time series pattern attention mechanism for prediction, and uses the good time series prediction performance of the recurrent neural network to adaptively adjust the weight of each time step through the attention layer, so that the model pays more attention to important time steps and predicts and warns input measurements. Compared with fault warnings based on low-dimensional data, it has higher accuracy and can perform risk management before the equipment fails, reducing the risk of equipment failure during peak power consumption periods. In addition, the trained neural network model is combined with online data sampling, and the real-time status warning quantity prediction value can be obtained by standardizing the online data window input, realizing the online status warning of substation equipment.

[0053] (5) The present invention predicts the relationship between the state quantity and the influencing factors through correlation coefficient analysis. The Pearson correlation coefficient can use the absolute value or change amount to screen out potential fault factors, provide a quantitative reference for maintenance work, facilitate targeted maintenance, and improve the availability of equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0055] Figure 2 It is a schematic diagram of the TPA-LSTM model structure of the present invention;

[0056] Figure 3 It is a structural diagram of the LSTM Unit in the present invention. DETAILED DESCRIPTION

[0057] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0058] Example 1

[0059] This embodiment provides a method for managing the entire life cycle of substation equipment based on correlation analysis. Figure 1 As shown, the method comprises the following steps:

[0060] Step 1: According to the substation equipment management requirements, obtain the substation equipment status sampling time series data.

[0061] The status sampling time series data of the substation equipment is arranged in rows, for example, the sampling data at 0:00 is in the first row, and the sampling data at 1:00 is in the second row. The sampled data includes three types: electrical quantity measurement, non-electrical quantity measurement and external environmental variables. The electrical quantity measurement includes the load current, core grounding current, high-frequency partial discharge, etc. of the equipment. The non-electrical quantity measurement includes the main transformer infrared temperature, pipe mother joint temperature, oil temperature, body temperature, body humidity, gas content in oil, etc. of the equipment. The external environmental variables include ambient temperature, ambient humidity, etc.

[0062] Step 2: Clean the unevenly sampled data and add noise.

[0063] Step 2-1, first obtain the timestamps of the target measurement fixed intervals from top to bottom, such as each row and its next three rows, and then determine whether the interval between the two is less than the set threshold. For example, the threshold is set to one day, that is, when the time difference between the upper and lower three rows exceeds one day, it is determined that the measurement has sampling missing or interruption, then skip the row and continue to select downward.

[0064] Step 2-2: When the time span of two rows of target measurement meets the requirements, this time span is set as the search time period. In the state sampling data, search for cleaning measurements with timestamps within the search time period and fill in the set variable length moving window data. For example, if five cleaning measurements are selected and the time window length is set to 5, then the variable length moving window data matrix M T The shape is 5×5, which means 25 data.

[0065] Step 2-3, in general, due to the unstable quality of the actual sampling data, the original data cannot fill the entire moving time window, and the interpolation method is used to fill the remaining gaps in the time window matrix. The interpolation method of this embodiment includes one of linear interpolation, polynomial interpolation, Lagrange interpolation, cubic spline interpolation and spline interpolation. At the same time, in order to ensure the information density of the cleaned data, the interpolation number threshold can be set. For example, when the available sampling values ​​of the cleaned measurement are less than 3 within the search time period, it is determined that there is data missing in a certain measurement within the time period, and the time period is abandoned, and the target measurement timestamp is selected downward.

[0066] Step 2-4: If the variable-length moving window data is successfully formed, the matrix M T Output. At the same time, the start and end timestamps of the corresponding target measurement and the corresponding measurement value under the timestamp can be output according to the needs for subsequent data analysis or label addition.

[0067] The above steps propose a set of efficient, robust, and versatile variable-length moving window data cleaning methods to preprocess the substation equipment status sampling, which solves the irregularity, discontinuity, non-alignment and granularity unevenness of substation equipment status sampling as spatiotemporal big data.

[0068] This step also uses differential privacy protection technology based on the Laplace mechanism to add noise. Specifically:

[0069] Step 2-5: Let the original data be f, and assume that the global sensitivity of f is Where D 1 , D 2 are adjacent data points; the privacy budget is ε, ranging from 0 to 1; the noise η follows a Laplace distribution with parameter f / ε; then the differentially private version of f Serves as data input for subsequent online early warnings.

[0070] This step uses differential privacy technology to add noise to synthesize new data sets to replace the previous data sets. These synthesized new data sets are similar to the original data sets in statistical characteristics, but do not contain individual information of the original data, and can protect data privacy.

[0071] Step 3: Construct a recurrent neural network state prediction model based on the temporal pattern attention mechanism to realize online early warning of substation equipment status.

[0072] Step 3-1, set up online warning measurement, and pre-screen influencing factors (load current, core grounding current, high-frequency partial discharge, main transformer infrared temperature, oil temperature, body temperature, gas content in oil, etc.) as cleaning measurements based on prior knowledge.

[0073] Step 3-2: Use the variable-length moving window data cleaning method to obtain the influencing factor output matrix in the format of: number of samples × moving window length × number of features. At the same time, use the custom feature output of the designed data cleaning method to output the warning measurement as a label according to the required fixed prediction period for subsequent recurrent neural network training.

[0074] Step 3-3: Use historical influencing factors as input features to build a recurrent neural network state prediction model (TPA-LSTM) based on the temporal pattern attention mechanism. Figure 2 , given a time series of w-1 data, it is represented as X = {x t-w+1 ,...,x i ,...,x t-1},enter Figure 3 In the LSTM Unit shown, x i Represents the observation value at time i. The LSTM network extracts features from the time series and outputs the hidden state h i '. The forward propagation step of the LSTM includes the following calculations:

[0075] Calculate the input gate to determine the amount by which the memory cell should be updated:

[0076] I i =σ(W I ·[h′ i-1 ,x i ]+b I ) (1)

[0077] Calculate the forget gate to determine how much information should be removed from the memory cell:

[0078] F i =σ(W F ·[h′ i-1 ,x i ]+b F ) (2)

[0079] Calculate the output gate, which determines the amount of information to be output to the next layer or as the final output:

[0080] O i =σ(W O ·[h′ i-1 ,x i ]+b O ) (3)

[0081] Update the state of the memory cell and the value of the memory cell C i :

[0082]

[0083] Calculate the final output of LSTM:

[0084] h i '=O i tanh(C i ) (6)

[0085] Among them, x i is the current input in the input sequence; h′ i-1 is the hidden state of the previous time step; W I ,W F ,W O ,W C is the weight matrix; b I ,b F ,b O ,b C is the bias vector; σ is the sigmoid function; tanh is the hyperbolic tangent function.

[0086] The TPA attention mechanism selects the important temporal patterns related to the current time step by calculating the attention weights. t When predicting, add all time series pairs h to the original formula t The specific steps are as follows:

[0087] 1) After passing through the LSTM layer, the hidden state sequence h of each time step is obtained i '.

[0088] 2) Use the attention mechanism to calculate the attention weights for each time step.

[0089] e i =h i ' T tanh(W i h′ i +b i ) (7)

[0090]

[0091] Among them, W i and b i is the weight matrix.

[0092] 3) Perform weighted summation on each time step to get v t , represents the time step h t The combined effect of the invention is the temporal attention proposed.

[0093]

[0094] Output the final result h t :

[0095] h t =W h (W h' h′ t +W v v t ) (10)

[0096] Where W h , W h' , W v is the weight matrix.

[0097] Step 3-4: Use the Adam optimizer to optimize the model hyperparameters and optimize the training process. The model parameters and optimizer status are saved regularly through the breakpoint resume mechanism, and the number of current training rounds and loss value and other information are recorded. The network weights with the best training effect can be saved, and the network can be adaptively optimized after adding historical data training sets in the future.

[0098] Step 3-5: Combine the trained neural network model with online data sampling, and obtain the real-time status warning quantity prediction value by standardizing the online data window input, so as to realize the online status warning of substation equipment.

[0099] This model uses the good time series prediction performance of the recurrent neural network, and adaptively adjusts the weight of each time step through the attention layer, so that the model pays more attention to important time steps and predicts and warns input measurements. Compared with fault warnings based on low-dimensional data, it has higher accuracy and can perform risk management before the equipment fails, reducing the risk of equipment failure during peak power consumption periods. In addition, the trained neural network model is combined with online data sampling. By standardizing the windowed input of online data, the real-time status warning quantity prediction value can be obtained, realizing online status warning of substation equipment.

[0100] Step 4: Implement full life cycle management of substation equipment based on correlation analysis.

[0101] When an alarm occurs, the relationship between the predicted value of the real-time state warning quantity and the influencing factors is analyzed by the Pearson correlation coefficient. The calculation formula of the Pearson coefficient is shown in formula (11).

[0102]

[0103] In the formula, ρ X,Y is the Pearson correlation coefficient, cov represents the covariance, X, Y are the state warning prediction and the corresponding influencing factors, σ X , σ Y are the standard deviations of X and Y respectively, E represents the expectation, μ X , μ Y are the average values ​​of X and Y respectively.

[0104] When the absolute value of the Pearson coefficient exceeds 0.8, it can be considered to be highly correlated, and potential fault factors can be screened out to provide a quantitative reference for maintenance work, so that the corresponding maintenance work can be carried out to troubleshoot the faults. Then, real-time status detection can be performed according to the above steps to achieve full life cycle status monitoring of the equipment and full life cycle equipment management.

[0105] Example 2

[0106] This embodiment provides a substation equipment full life cycle management system based on correlation analysis, which is characterized by including:

[0107] Data acquisition module: acquires real-time state sampling time series data of substation equipment, and selects influencing factors as cleaning measurements based on prior knowledge, wherein the real-time state sampling time series data includes electrical quantity measurement, non-electrical quantity measurement and external environmental variables of substation equipment;

[0108] Cleaning module: fills the cleaning measurement into a preset variable length moving window and outputs a variable length moving window data matrix;

[0109] Noise addition and prediction module: Based on the variable-length moving window data matrix, noise is added using differential privacy technology, input into a pre-built recurrent neural network state prediction model based on the time series pattern attention mechanism, and output the real-time state warning prediction of the substation equipment;

[0110] Judgment module: judge whether the real-time status warning prediction quantity reaches the set warning value, if so, calculate the correlation between the real-time status warning prediction quantity and the corresponding influencing factor, and screen out the fault factor, if not, do nothing;

[0111] Management module: perform fault inspection and repair based on the fault factors, and repeat the processing of the above modules to achieve full life cycle management of substation equipment.

[0112] The rest is the same as in Example 1.

[0113] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0114] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0115] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0116] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0117] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0118] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0119] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for managing the entire life cycle of substation equipment based on correlation analysis, characterized in that: The following steps are involved: S1. Acquire real-time state sampling time series data of substation equipment, and select influencing factors as cleaning measurements based on prior knowledge, wherein the real-time state sampling time series data includes electrical quantity measurement, non-electrical quantity measurement and external environmental variables of substation equipment; S2, filling the cleaning measurement into a preset variable-length moving window, and outputting a variable-length moving window data matrix; S3, based on the variable-length moving window data matrix, adding noise by using differential privacy technology, inputting the pre-built recurrent neural network state prediction model based on the time series pattern attention mechanism, and outputting the real-time state warning prediction of the substation equipment; S4, judging whether the real-time status warning prediction amount reaches the set warning value, if so, calculating the correlation between the real-time status warning prediction amount and the corresponding influencing factor, and screening out the fault factor, if not, doing nothing; S5, perform fault repair based on the fault factor, and return to step S1 to implement full life cycle management of substation equipment; The differential privacy technology is a differential privacy protection technology based on the Laplace mechanism. The expression for adding noise using the differential privacy technology is: Where Δf is the global sensitivity of f, f is the original data, D1 and D2 are adjacent data points, f' is the data after differential privacy of f, ε is the privacy budget, ranging from 0 to 1, η is the noise, and obeys the Laplace distribution with parameter Δf / ε; The recurrent neural network state prediction model based on the temporal pattern attention mechanism includes an LSTM network and a TPA attention mechanism network. The steps of constructing the recurrent neural network state prediction model based on the temporal pattern attention mechanism include: 1) Obtaining historical state sampling time series data of substation equipment within a set search time period as historical cleaning measurement, and setting corresponding online warning measurement, wherein the set search time period includes multiple lines of historical state sampling time series data, and the time span between each line is the same; 2) Based on the historical cleaning measurement, a variable-length moving window data cleaning method is used to obtain a variable-length moving window data matrix, and the variable-length moving window data matrix is ​​used together with the online warning measurement as a training input of a recurrent neural network state prediction model based on a temporal pattern attention mechanism, wherein the online warning measurement is used as a label, and the format of the variable-length moving window data matrix is: cleaning measurement×moving window length×number of features; 3) The LSTM network extracts features from the training input to obtain the hidden state h of the training input at each time step i ', and use the TPA attention mechanism network to calculate the attention weight of the training input at each time step to output the substation equipment status prediction result; 4) Using the Adam optimizer to perform weight optimization to adjust the hyperparameters of the recurrent neural network state prediction model based on the temporal pattern attention mechanism, and record the current training round and loss value; 5) Return to step 3) to perform iterative training until the preset conditions are met and the training is terminated, thereby obtaining a recurrent neural network state prediction model based on the temporal pattern attention mechanism; The step of obtaining a variable length moving window data matrix also includes: Determine whether the number of available sampling values ​​of the cleaning measurement in the variable-length moving window matrix is ​​less than a preset value. If so, determine that the cleaning measurement is missing, and use interpolation to fill the gap to obtain a complete variable-length moving window matrix. If not, do nothing. In step 3), the step of outputting the substation equipment status prediction result includes the following: The TPA attention mechanism network obtains the hidden state h of the training input at each time step i ', use the attention mechanism to calculate the attention weights of the training input at each time step: The attention weights of the training input at each time step are weighted summed to obtain the hidden layer weight v at the tth moment t : Output the substation equipment status prediction results of the TPA attention mechanism network: h t =W h (W h' h t '+W v v t ) In the formula, h i ' is the transpose; tanh is the hyperbolic tangent function; W i and b i is the weight matrix; α i is the attention weight for each time step; e i is the correlation between the input element and the context at time step i, e m is the correlation degree when the time step is m, and the value range of m is [1, i]; h t is the substation equipment status prediction result at the tth moment; W h , W h' , W v is the weight matrix.

2. A method for managing the entire life cycle of substation equipment based on correlation analysis according to claim 1, characterized in that: The electrical quantity measurements include the equipment's load current, core grounding current, and high-frequency partial discharge; the non-electrical quantity measurements include the equipment's main transformer infrared temperature, pipe mother joint temperature, oil temperature, body temperature, body humidity, and gas content in oil; the external environmental variables include ambient temperature and ambient humidity.

3. A method for managing the entire life cycle of substation equipment based on correlation analysis according to claim 1, characterized in that: The steps of obtaining the historical cleaning measurement include: Get the timestamp of each row of historical state sampling time series data within the set time interval from top to bottom; According to the timestamp, determine whether the time interval between two rows is less than the set time interval, If so, it is determined that there is no sampling missing or interruption between the two rows, and the multiple rows of historical state sampling time series data within the set search time period are used as historical cleaning measurements. If not, determine that there is no sampling missing or interruption between the two rows, select Skip, and continue to select downward until the time span between the two rows is the same, and use multiple rows of historical state sampling time series data within the set search time period as historical cleaning measurements.

4. A method for managing the entire life cycle of substation equipment based on correlation analysis according to claim 1, characterized in that: The interpolation method includes one of a linear interpolation method, a polynomial interpolation method, a Lagrange interpolation method, a cubic spline interpolation method and a spline interpolation method.

5. A method for managing the entire life cycle of substation equipment based on correlation analysis according to claim 1, characterized in that: The correlation was calculated using the Pearson correlation coefficient: In the formula, ρ X,Y is the Pearson correlation coefficient, cov represents the covariance, X, Y are the state warning prediction and the corresponding related state operation data, σ X , σ Y are the standard deviations of X and Y respectively, E represents the expectation, μ X , μ Y are the average values ​​of X and Y respectively.

6. A management system for a full life cycle management method of substation equipment based on correlation analysis according to any one of claims 1 to 5, characterized in that: include: Data acquisition module: acquires real-time state sampling time series data of substation equipment, and selects influencing factors as cleaning measurements based on prior knowledge, wherein the real-time state sampling time series data includes electrical quantity measurement, non-electrical quantity measurement and external environmental variables of substation equipment; Cleaning module: fills the cleaning measurement into a preset variable length moving window and outputs a variable length moving window data matrix; Noise addition and prediction module: Based on the variable-length moving window data matrix, noise is added using differential privacy technology, input into a pre-built recurrent neural network state prediction model based on the time series pattern attention mechanism, and output the real-time state warning prediction of the substation equipment; Judgment module: judge whether the real-time status warning prediction quantity reaches the set warning value, if so, calculate the correlation between the real-time status warning prediction quantity and the corresponding influencing factor, and screen out the fault factor, if not, do nothing; Management module: perform fault inspection and repair based on the fault factors, and repeat the processing of the above modules to achieve full life cycle management of substation equipment.

Images