Method and device for predicting content of dissolved gas in transformer oil

Abstract

The embodiment of the invention provides a method and device for predicting the content of dissolved gas in transformer oil, and the method comprises the steps: inputting to-be-measured data of the dissolved gas in the transformer oil into a pre-trained combined model of VMD (variational mode decomposition) and LSTM (long short-term memory) network, and obtaining an intrinsic mode function prediction result of the dissolved gas in each type of oil; combining the intrinsic mode function prediction results of the dissolved gas in each type of oil to obtain a corresponding prediction result; according to the method, the prediction accuracy of nonlinear and non-stationary time sequences can be improved based on a combined model of the VMD and the LSTM, real-time monitoring and fault early warning of dissolved gas data are realized, and the stability of the model in a dynamic scene is enhanced; the device can continuously provide reliable prediction results under different operation conditions, and the fault early warning capability and the system operation safety are improved.

Description

Technical Field

[0001] This application relates to the field of power grid equipment monitoring, and particularly to a method and device for predicting the content of dissolved gases in transformer oil. Background Art

[0002] As a core device in the power system, the operating state of a transformer is directly related to the stability and reliability of the power grid. However, during actual operation, due to the effects of various factors such as electricity, heat, and machinery, problems such as partial discharge, overheating, and insulation aging may be triggered, resulting in the generation of specific decomposed gases in the insulating oil of the transformer. The concentration changes of these gases are usually early signals of potential faults. By predicting the change trend of gas concentration, potential hidden dangers inside the transformer can be detected in a timely manner, the fault warning ability of the equipment can be improved, and the impact on the power grid caused by sudden faults can be reduced. Therefore, accurately predicting the concentration of dissolved gases in the oil is crucial for evaluating the transformer state and taking preventive measures.

[0003] Dissolved gas analysis has become an important means for evaluating the transformer state. However, traditional off-line analysis methods have disadvantages such as long monitoring intervals and poor real-time performance, making it difficult to meet the requirements for early fault identification in actual operation and maintenance. With the continuous development of transformer on-line monitoring technology, it has become possible to collect and analyze dissolved gas data in real time, which has also promoted gas concentration prediction technology to become a research hotspot at home and abroad.

[0004] Currently, domestic scholars have proposed various prediction methods based on machine learning. For example, a model is constructed by combining transformer operation data and the extreme learning machine algorithm to improve the real-time performance and accuracy of prediction. In addition, combined models of wavelet neural network, grey neural network, and support vector machine, as well as deep recursive belief network, have further improved the prediction performance. In international research, scholars have tried to use time convolutional network and graph convolutional network for modeling to capture complex dynamic features. At the same time, the time series method based on seasonal autoregressive model effectively focuses on the periodic change characteristics of gas concentration. There are also studies combining advanced technologies such as honey badger algorithm and variational mode decomposition to conduct in-depth research on the non-linear and non-stationary characteristics of gases.

[0005] However, the existing technologies have insufficient ability to capture the complex dynamic characteristics of gas concentration in engineering applications, and the prediction accuracy for non-linear and non-stationary sequences needs to be improved. In addition, problems such as dealing with noise and data mutations still require further optimization of the model structure to meet the requirements of actual operation and maintenance. Summary of the Invention

[0006] In view of the problems in the prior art, the present application provides a method and device for predicting the content of dissolved gases in transformer oil, which can improve the prediction accuracy of non-linear and non-stationary time series based on a combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), realize real-time monitoring and fault warning of dissolved gas data, enhance the stability of the model in dynamic scenarios, ensure that the device continuously provides reliable prediction results under different operating conditions, and improve the fault warning ability and the safety of system operation.

[0007] To solve at least one of the above problems, the present application provides the following technical solutions:

[0008] According to the first aspect of the embodiments of the present application, the present application provides a method for predicting the content of dissolved gases in transformer oil, including:

[0009] Input the data to be measured of the dissolved gases in the transformer oil into a pre-trained combined model of variational mode decomposition VMD and long short-term memory network LSTM to obtain the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil, where the combined model is obtained by decomposing the oil chromatogram sequence of the historical data of the dissolved gases in the transformer oil into multiple mode functions with different frequencies through VMD, and capturing the time series characteristics and non-linear relationships of multi-dimensional data through LSTM for optimization;

[0010] Merge the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil to obtain the corresponding prediction results.

[0011] According to any implementation manner of the present application, the training process of the pre-trained combined model of variational mode decomposition VMD and long short-term memory network LSTM includes:

[0012] Collect the historical data of the dissolved gases in the transformer oil within a preset time range, and perform missing value replacement processing on the historical data based on the forward method;

[0013] Perform time-frequency synchronous decomposition on the sequences of different gases in the historical data through VDM to obtain the corresponding multiple frequency-aligned multivariate intrinsic mode functions IMFs and residuals, where the frequency alignment means that each gas is aligned inside the multiple IMFs decomposed from the sequences of other influencing factors;

[0014] Perform Z-score normalization processing on the multiple IMFs, and construct corresponding multiple multi-step LSTM network models according to the normalized multiple IMFs.

[0015] According to any implementation manner of the present application, performing time-frequency synchronous decomposition on the sequences of different gases in the historical data through VDM to obtain the corresponding multiple frequency-aligned multivariate intrinsic mode functions IMFs and residuals includes:

[0016] Perform Hilbert transform on each dissolved gas sequence to extract unilateral spectrum information;

[0017] Shift the spectrum of each dissolved gas component to the baseband region by modulation and align the center frequencies;

[0018] Perform regularization on the gradient of the modulated signal and apply Gaussian smoothing method to obtain the frequency bandwidth of each component;

[0019] Input each processed dissolved gas sequence into the VMD model and decompose it into multiple IMFs by iterative optimization method. Each IMF represents an independent signal component in a frequency range, while retaining the residual information.

[0020] According to any embodiment of the present application, the Z-score normalization processing of the multiple IMFs includes:

[0021] Calculate the mean and standard deviation of each IMF to obtain its data distribution characteristics;

[0022] Subtract the mean from each data point and divide the de-meaned data by the standard deviation to adjust it to the standard normal distribution state.

[0023] According to any embodiment of the present application, the construction of the corresponding multiple multi-step LSTM network models based on the normalized multiple IMFs includes:

[0024] Use the normalized multiple IMFs sequences as input data, and the corresponding IMFs sequences for the next week as the target output. Independently construct an LSTM network model for each IMF component, and initialize the weight parameters and biases of the input gate, forget gate, and output gate, and prepare for model training;

[0025] Input each IMFs sequence into the corresponding LSTM model. The model learns the relationship between the current sequence and the future trend, and uses the iterative optimization method to gradually adjust the parameters to generate the multiple multi-step LSTM network models that can accurately capture the characteristics of the time series.

[0026] According to any embodiment of the present application, the merging of the prediction results of the intrinsic mode functions of each type of dissolved gas in oil to obtain the corresponding prediction results includes:

[0027] Use the reverse normalization technique to restore the normalized IMFs results predicted by the LSTM model to the original numerical range through weighted calculation;

[0028] The predicted values of each IMF after normalization are gradually added according to the gas types they belong to, and the corresponding prediction results are obtained.

[0029] According to the second aspect of the embodiments of the present application, the present application provides a prediction device for the content of dissolved gases in transformer oil, including:

[0030] A data input module, configured to: input the data to be measured of the dissolved gases in the transformer oil into a pre-trained combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), and obtain the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil, where the combined model is obtained by decomposing the oil chromatogram sequence of the historical data of the dissolved gases in the transformer oil into multiple mode functions with different frequencies through VMD, and optimizing by capturing the time series characteristics and non-linear relationships of multi-dimensional data through LSTM;

[0031] A result determination module, configured to: merge the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil to obtain the corresponding prediction results.

[0032] According to any implementation manner of the present application, the training process of the pre-trained combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM) includes:

[0033] A data acquisition module, configured to: acquire the historical data of the dissolved gases in the transformer oil within a preset time range, and perform missing value replacement processing on the historical data based on the forward method;

[0034] A data decomposition module, configured to: perform time-frequency synchronous decomposition on the sequences of different gases in the historical data through VDM to obtain the corresponding multiple frequency-aligned multivariate intrinsic mode functions (IMFs) and residuals, where the frequency alignment means that each gas is aligned inside the multiple IMFs decomposed from the sequences of other influencing factors;

[0035] A model construction module, configured to: perform Z-score normalization processing on the multiple IMFs, and construct corresponding multiple multi-step LSTM network models according to the normalized multiple IMFs.

[0036] According to any implementation manner of the present application, the data decomposition module includes:

[0037] A transformation unit, configured to: perform Hilbert transformation on each dissolved gas sequence to extract unilateral spectrum information;

[0038] An alignment unit, configured to: move the spectrum of each dissolved gas component to the baseband region by modulation and align the center frequencies;

[0039] A bandwidth unit, configured to: perform regularization processing on the modulated signal gradient and apply a Gaussian smoothing method to obtain the frequency bandwidth of each component;

[0040] An iterative unit, configured to: input each processed dissolved gas sequence into a VMD model, and decompose it into multiple IMFs through an iterative optimization method. Each IMF represents an independent signal component within a frequency range, while retaining residual information.

[0041] According to any embodiment of the present application, the model construction module performs Z-score normalization processing on the multiple IMFs, including:

[0042] A feature acquisition unit, configured to: calculate the mean and standard deviation of each IMF to obtain its data distribution characteristics;

[0043] A distribution adjustment unit, configured to: subtract the mean from each data point and divide the mean-subtracted data by the standard deviation to adjust it to a standard normal distribution state.

[0044] According to any embodiment of the present application, the model construction module constructs corresponding multiple multi-step LSTM network models based on the normalized multiple IMFs, including:

[0045] A model construction unit, configured to: use the normalized multiple IMF sequences as input data, and use the corresponding IMF sequences for the next week as the target output. Independently construct an LSTM network model for each IMF component, and initialize the weight parameters and biases of the input gate, forget gate, and output gate, and perform model training;

[0046] A parameter adjustment unit, configured to: input each IMF sequence into the corresponding LSTM model. The model learns the relationship between the current sequence and the future trend, and uses an iterative optimization method to gradually adjust the parameters to generate the multiple multi-step LSTM network models that can accurately capture the characteristics of the time series.

[0047] According to any embodiment of the present application, the step of merging the prediction results of the intrinsic mode functions of the dissolved gases in each type of oil to obtain the corresponding prediction results includes:

[0048] A data restoration unit, configured to: use an inverse normalization technique to restore the normalized IMF results predicted by the LSTM model to the original numerical range through weighted calculation;

[0049] A data merging unit, configured to: gradually add the predicted values of each normalized IMF according to the gas types to which they belong to obtain the corresponding prediction results.

[0050] According to the third aspect of the embodiments of the present application, the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the method for predicting the content of dissolved gases in transformer oil are implemented.

[0051] According to the fourth aspect of the embodiments of the present application, the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for predicting the content of dissolved gases in transformer oil are implemented.

[0052] According to the fifth aspect of the embodiments of the present application, the present application provides a computer program product, including a computer program / instructions. When the computer program / instructions are executed by a processor, the steps of the method for predicting the content of dissolved gases in transformer oil are implemented.

[0053] As can be seen from the above technical solutions, the present application provides a method and device for predicting the content of dissolved gases in transformer oil. By inputting the data to be measured of the dissolved gases in transformer oil into a pre-trained combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil are obtained, and the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil are combined to obtain the corresponding prediction results. It is possible to improve the prediction accuracy of non-linear and non-stationary time series based on the combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), realize the real-time monitoring and fault warning of dissolved gas data, enhance the stability of the model in dynamic scenarios, ensure that the device continuously provides reliable prediction results under different operating conditions, and improve the fault warning ability and the safety of system operation. Description of the Drawings

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 It is one of the flow charts of the method for predicting the content of dissolved gases in transformer oil in the embodiments of the present application;

[0056] Figure 2 It is another flow chart of the method for predicting the content of dissolved gases in transformer oil in the embodiments of the present application;

[0057] Figure 3It is the third schematic diagram of the process for the prediction method of the dissolved gas content in transformer oil in the embodiments of the present application;

[0058] Figure 4 It is the fourth schematic diagram of the process for the prediction method of the dissolved gas content in transformer oil in the embodiments of the present application;

[0059] Figure 5 It is the fifth schematic diagram of the process for the prediction method of the dissolved gas content in transformer oil in the embodiments of the present application;

[0060] Figure 6 It is the overall framework schematic diagram of VMD-LSTM for the dissolved gas content in transformer oil in the embodiments of the present application;

[0061] Figure 7 It is the schematic diagram of the convergence curve of the LSTM model for the dissolved gas content in transformer oil in the embodiments of the present application;

[0062] Figure 8 It is the schematic diagram of the fitting effect of the training set of the first intrinsic mode function after hydrogen decomposition of the dissolved gas content in transformer oil in the embodiments of the present application;

[0063] Figure 9 It is the schematic diagram of the fitting effect of the training set of the second intrinsic mode function after hydrogen decomposition of the dissolved gas content in transformer oil in the embodiments of the present application;

[0064] Figure 10 It is the schematic diagram of the fitting effect of the training set of the third intrinsic mode function after hydrogen decomposition of the dissolved gas content in transformer oil in the embodiments of the present application;

[0065] Figure 11 It is the schematic diagram of the prediction result of hydrogen in the dissolved gas content in transformer oil in the embodiments of the present application;

[0066] Figure 12 It is the first structural diagram of the prediction device for the dissolved gas content in transformer oil in the embodiments of the present application;

[0067] Figure 13 It is the second structural diagram of the prediction device for the dissolved gas content in transformer oil in the embodiments of the present application;

[0068] Figure 14 It is the third structural diagram of the prediction device for the dissolved gas content in transformer oil in the embodiments of the present application;

[0069] Figure 15 It is the fourth structural diagram of the prediction device for the dissolved gas content in transformer oil in the embodiments of the present application;

[0070] Figure 16This is the fifth structural diagram of the prediction device for the content of dissolved gases in transformer oil in the embodiments of the present application;

[0071] Figure 17 This is the schematic structural diagram of the electronic device in the embodiments of the present application. Detailed implementation manners

[0072] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some but not all of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0073] In the technical solutions of the present application, the acquisition, storage, use, processing, etc. of data all comply with the relevant regulations of national laws and regulations.

[0074] Considering the problems that the existing technology has insufficient ability to capture the complex dynamic characteristics of gas concentration in engineering applications, the prediction accuracy of non-linear and non-stationary sequences needs to be improved, and the problems such as processing noise and data mutation still require further optimization of the model structure to meet the actual operation and maintenance requirements, the present application provides a prediction method and device for the content of dissolved gases in transformer oil. Based on a combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), the prediction accuracy of non-linear and non-stationary time series is improved, real-time monitoring and fault warning of dissolved gas data are realized, the stability of the model in dynamic scenarios is enhanced, reliable prediction results are continuously provided for the device under different operating conditions, and the fault warning ability and the safety of system operation are improved.

[0075] In order to improve the prediction accuracy of non-linear and non-stationary time series based on a combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), realize real-time monitoring and fault warning of dissolved gas data, enhance the stability of the model in dynamic scenarios, ensure that the device continuously provides reliable prediction results under different operating conditions, and improve the fault warning ability and the safety of system operation, the present application provides an embodiment of a prediction method for the content of dissolved gases in transformer oil. Refer to Figure 1 , the prediction method for the content of dissolved gases in transformer oil specifically includes the following content:

[0076] Step S101: Input the data to be measured of the dissolved gases in the transformer oil into a pre-trained combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM) to obtain the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil. The combined model decomposes the oil chromatogram sequence of the historical data of the dissolved gases in the transformer oil into multiple mode functions with different frequencies through VMD, and captures the temporal characteristics and non-linear relationships of multi-dimensional data through LSTM for optimization.

[0077] Among them, the data to be measured of the dissolved gases in the transformer oil is first input into the pre-trained VMD-LSTM combined model for processing. By combining the technical advantages of variational mode decomposition (VMD) and long short-term memory network (LSTM), the model completes the in-depth analysis of the data to be measured. Specifically, the role of VMD is to decompose the input data of the dissolved gases in the oil into multiple mode functions with relatively stable frequencies, and each mode function represents a certain frequency component in the original data, so as to effectively separate the noise and trend components in the data.

[0078] Subsequently, the LSTM model further processes the decomposed mode functions, and captures the temporal characteristics and non-linear change laws of each mode function through its powerful temporal modeling ability. LSTM can not only identify the short-term dynamic characteristics in the data, but also remember the long-term dependence characteristics in the data, thus optimizing the prediction accuracy and robustness of the overall model. Through the synergistic effect of VMD and LSTM, the model can accurately capture the complex dynamic characteristics of the dissolved gas data and generate the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil.

[0079] Step S102: Combine the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil to obtain the corresponding prediction results.

[0080] Among them, the combined prediction results reflect the concentration change trend of various dissolved gases in the transformer oil in a future period of time. This result can be directly used for the evaluation of the equipment operation status and the early warning of potential faults, providing reliable technical support for the safe and stable operation of the transformer. By integrating the multi-modal prediction results into an overall trend, the practicality and interpretability of the prediction are further improved.

[0081] As can be seen from the above description, the prediction method for the content of dissolved gases in transformer oil provided by the embodiments of the present application can improve the prediction accuracy of non-linear and non-stationary time series based on the combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), realize the real-time monitoring and fault early warning of dissolved gas data, enhance the stability of the model in dynamic scenarios, ensure that the equipment continuously provides reliable prediction results under different operating conditions, improve the fault early warning ability and the safety of system operation.

[0082] In an embodiment of the method for predicting the content of dissolved gases in transformer oil of the present application, refer to Figure 2 , the training process of the pre-trained combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM) includes:

[0083] Step S001: Collect historical data of dissolved gases in transformer oil within a preset time range, and perform missing value replacement processing on the historical data based on the forward filling method.

[0084] Among them, after collecting the historical data of dissolved gases in transformer oil within a preset time range, first perform statistical analysis on the collected samples to identify the missing values in the data. For these missing values, the present application uses the forward filling method for processing, that is, replacing the missing values with the most recent valid observation value before them, that is, ensuring the integrity and consistency of the data by utilizing the continuity characteristics of time series data.

[0085] Step S002: Perform time-frequency synchronous decomposition on the sequences of different gases in the historical data through VDM to obtain corresponding multiple frequency-aligned multivariate intrinsic mode functions (IMFs) and residuals, where the frequency alignment means that each gas is aligned inside the multiple IMFs decomposed from other influencing factor sequences.

[0086] Exemplarily, the present application uses VMD to perform time-frequency synchronous decomposition on one-dimensional sequences of hydrogen, acetylene, total hydrocarbons, carbon monoxide, and carbon dioxide {X1, X2, …, X d}, and decomposes them into frequency-aligned multivariate intrinsic mode functions and residuals. Each original sequence is decomposed into k intrinsic mode functions (IMFs), and the d-dimensional sequence is decomposed into d×k-dimensional frequency-aligned IMFs, that is, IMF 1,1 , IMF 1,2 , …, IMF 1,k , …, IMF d,k , where the hydrogen sequence is decomposed into IMF 1,1 , IMF 1,2 , …, IMF 1,k . Frequency alignment means that the k IMFs decomposed from hydrogen and other influencing factor sequences are aligned inside, such as the center frequencies of IMF 1,1 , IMF 2,1 , …, IMF d,1 that make up IMF1 are the same, but the center frequencies between IMF1, IMF2, …, IMF k are different.

[0087] Step S003: Perform Z-score normalization on the multiple IMFs, and construct corresponding multiple multi-step LSTM network models based on the normalized multiple IMFs.

[0088] In an alternative embodiment, referring to Figure 3 , the time-frequency synchronous decomposition of the sequences of different gases in the historical data by VDM to obtain corresponding multiple frequency-aligned multivariate intrinsic mode function IMFs and residuals includes:

[0089] Step S002A: Perform Hilbert transform on each dissolved gas sequence to extract unilateral spectrum information;

[0090] Step S002B: Move the spectrum of each dissolved gas component to the baseband region by modulation and align the center frequencies;

[0091] Step S002C: Perform regularization processing on the gradient of the modulated signal, and apply the Gaussian smoothing method to obtain the frequency bandwidth of each component;

[0092] Step S002D: Input each processed dissolved gas sequence into the VMD model, and decompose it into multiple IMFs by the iterative optimization method. Each IMF represents an independent signal component in a frequency range, and at the same time, the residual information is retained.

[0093] Exemplarily, the specific steps are as follows:

[0094] Perform Hilbert transform on each photovoltaic component to obtain the unilateral spectrum;

[0095] Transfer the spectra of the dissolved gases in oil components to the baseband region by mixing an exponential tuned to their respective estimated center frequencies.

[0096] Estimate the bandwidth of each photovoltaic component by Gaussian smoothing of the L2 regularization of the demodulated signal gradient.

[0097] Among them, the VMD of the dissolved gas output sequence in oil can be represented by formula (1):

[0098]

[0099] Among them, represents taking the partial derivative; δ(t) is the Dirac distribution function; represents the convolution operation; K is the total number of components; f(t) is the original photovoltaic power output signal.

[0100] Convert the above constrained mechanism problem into an unconstrained problem for solution through the Lagrange multiplier λ and the quadratic penalty term α, as shown in formula (2):

[0101]

[0102] Each component u k and the corresponding center frequency w k can be optimally solved by the alternating direction method of multipliers, and its update method is shown in formula (3):

[0103]

[0104] In the formula: and are the Fourier transforms of f(w), u k (w) and λ(w) respectively; w is the frequency; n is the number of iterations.

[0105] In summary, in this application, the unilateral spectrum information is first extracted by the Hilbert transform, and the frequency characteristics of the dissolved gas signal can be clearly analyzed. Then, the gas component spectrum is shifted to the baseband region by modulation and the center frequencies are aligned, which helps to unify the frequency ranges of the components and avoid the problem of spectrum overlap. Next, the frequency bandwidth is extracted through regularization processing and Gaussian smoothing methods, significantly reducing noise interference and improving the stability and accuracy of the data. Finally, the VMD model is used to decompose the gas signal into multiple IMFs, which can not only separate independent frequency components but also retain the residual information, ensuring signal integrity and dynamic feature expression ability.

[0106] The above decomposition method can more accurately capture the time-frequency characteristics of complex gas data, provide a high-quality input basis for the accurate prediction of subsequent models, and at the same time enhance the processing ability for nonlinear and non-stationary data.

[0107] In an optional embodiment, referring to Figure 4 , the Z-score normalization processing of the multiple IMFs includes:[[]]

[0108] Step S003A: Calculate the mean and standard deviation of each IMF to obtain its data distribution characteristics;

[0109] Step S003B: Subtract the mean from each data point and divide the de-meaned data by the standard deviation to adjust it to a standard normal distribution state.

[0110] In order to improve the stability and convergence speed of the model values, all the decomposed IMFs in this application need to be Z–score normalized before being input into the model. The specific normalization formula is shown in formula (4):

[0111]

[0112] Among them, Z imfDenote the standardized IMFs; μ represents the average value of the IMFs; σ represents the standard deviation of the IMFs.

[0113] When this application performs standardization processing on the IMFs, it first calculates the average value and standard deviation of each IMF to obtain the overall characteristics of its data distribution. Then, each data point in the IMFs is subtracted by its average value, thereby completing the mean removal operation and moving the center of the data to zero. Next, the mean-removed data is divided by the standard deviation to adjust it to the standard normal distribution state, enabling the data to have the standardized characteristics of a mean of zero and a variance of one, thereby improving the convergence speed and stability of the model.

[0114] In an optional embodiment, constructing a corresponding plurality of multi-step LSTM network models according to the standardized plurality of IMFs includes:

[0115] Step S003C: Use the standardized plurality of IMFs sequences as input data, and use the IMFs sequences corresponding to the next week as the target output. Independently construct an LSTM network model for each IMF component, and initialize the weight parameters and biases of the input gate, forget gate, and output gate to prepare for model training.

[0116] Among them, in order to construct a plurality of multi-step LSTM network models, first use the standardized plurality of IMFs sequences as input data, and use the IMFs sequences of the next week as the target output. An LSTM network model is independently constructed for each IMF component, thereby ensuring that each model focuses on the modal function it represents. The LSTM network consists of an input gate, a forget gate, and an output gate, and their functions are to control the selection of input information, forget useless information, and output valid information respectively. When constructing the model, it is necessary to initialize the weight parameters and biases of these gates. The initialization process ensures that the model has a reasonable initial state, lays a foundation for the subsequent training process, and at the same time ensures that each model can independently and efficiently process the data corresponding to its IMF component.

[0117] Step S003D: Input each IMF sequence into the corresponding LSTM model. The model learns the relationship between the current sequence and the future trend, and gradually adjusts the parameters using the iterative optimization method to generate the plurality of multi-step LSTM network models that can accurately capture the characteristics of the time series.

[0118] Next, each standardized IMFs sequence is input into the corresponding LSTM network for training. The LSTM captures the short-term dynamic changes and long-term dependence characteristics in time series data through its unique recurrent structure and memory units. The model learns the relationship between the current IMFs sequence and the future trend by gradually adjusting the parameters of the input gate, forget gate, and output gate. During the training process, an iterative optimization method is adopted to continuously reduce the prediction error of the model, thereby improving the accuracy. Finally, multiple multi-step LSTM network models are generated, and each model can accurately predict the future trend of the corresponding IMFs. This method can not only retain the temporal characteristics of the data but also significantly improve the accuracy and robustness of the prediction, providing strong support for subsequently combining the results of multiple IMFs into an overall predicted value.

[0119] Exemplarily, the LSTM forward propagation process is as follows:

[0120] c t = f t c t-1 + i t z t Cell state (2 - 1)

[0121]

[0122] Wherein, σ(·) represents the Sigmoid function; represents the tangent function; w z , w i , w f , w o respectively represent the weights of the input x t and the cell input, input gate, forget gate, and output gate; r z , r i , r f , r o respectively represent the weights of the hidden layer h t-1 and the cell input, input gate, forget gate, and output gate; b z , b i , b f , b o respectively represent the biases of the input x t and the cell input, input gate, forget gate, and output gate.

[0123] The above content shows the specific process of LSTM forward propagation, whose core consists of a cell state, a hidden state, an input gate, a forget gate, and an output gate. First, the LSTM updates the cell state, determines which information to retain through the forget gate, and injects useful new information by combining new cell inputs through the input gate. The hidden state is the output controlled by the output gate after the current cell state is processed by the tangent function. In the formula, the input data and the previous hidden state jointly participate in the calculation through the weight matrix and bias. The Sigmoid function is used to control the opening and closing degree of the gate, while the tangent function extracts non-linear features. This structure ensures that the model can effectively capture short-term and long-term dependencies in the time series and meet the prediction requirements of complex dynamic data.

[0124] In an alternative embodiment, referring to Figure 5 , the step of combining the intrinsic mode function prediction results of the dissolved gases in each type of oil to obtain the corresponding prediction results includes:

[0125] Step S102A: Using the inverse normalization technique, the normalized IMFs results predicted by the LSTM model are restored to the original numerical range through weighted calculation.

[0126] To restore the IMF before normalization, this application adopts the inverse normalization technique and uses the mean and standard deviation before normalization to restore the prediction data, as shown in formula (5):

[0127]

[0128] Where, represents the predicted result of the j-th normalized intrinsic function in the i-th dissolved gas in oil predicted by the LSTM; represents the predicted result of the j-th intrinsic function in the i-th dissolved gas in oil.

[0129] Step S102B: Gradually add the predicted values of each normalized IMF according to the gas types they belong to to obtain the corresponding prediction results.

[0130] The intrinsic mode function prediction results of the dissolved gases in each type of oil are combined to form the prediction results of the original data. The combination formula is as shown in formula (6):

[0131]

[0132] Where, represents the predicted value of the i-th dissolved gas.

[0133] As can be seen from the above embodiments, after the prediction is completed, each IMF represents the independent prediction result of a certain frequency range of the dissolved gas data. These IMFs are refined features obtained by decomposing the gas data, reflecting different frequency components of the gas concentration. In order to synthesize these independent components into a complete prediction result, it is necessary to gradually superimpose all the IMFs of the same gas.

[0134] First, restore the predicted values of each IMF to ensure that these values maintain the same physical meaning as the original data. Then add up the predicted values of all the IMFs belonging to the same dissolved gas, that is, use the characteristics of each mode function to restore the overall gas concentration value.

[0135] Finally, the merged result not only completely retains the contributions of each frequency component to the gas concentration change, but also accurately reflects the future change trend of the overall gas concentration, providing reliable support for further fault prediction and equipment status evaluation.

[0136] In summary, the beneficial effects of this application include the following aspects:

[0137] (1) Accuracy. This application adopts a variational mode decomposition (VMD) technology, which successfully overcomes the accuracy problem of the traditional long short-term memory (LSTM) network in non-linear and non-stationary time series prediction. Through this method, the accuracy of the prediction result is significantly improved, providing more reliable decision-making support for practical applications.

[0138] (2) Real-time performance. This application uses online oil chromatographic data to achieve timely detection and early warning of equipment failures. By real-time monitoring and analyzing the oil chromatographic data, this technology can identify potential failures in advance, thus significantly improving the equipment's fault early warning ability and ensuring the safe and stable operation of the equipment.

[0139] (3) Stability. This application significantly enhances the stability of the model in dynamic prediction scenarios by introducing a VMD-LSTM combined model. This method can effectively process non-stationary and non-linear oil chromatographic data, thus ensuring continuous provision of reliable prediction results under different operating conditions and reducing the sensitivity of the system to external disturbances.

[0140] To further illustrate this solution, this application also provides an overall framework of VMD-LSTM, see Figure 6 , which specifically includes the following content:

[0141] First, historical time series data of dissolved gases in transformer oil, such as hydrogen, acetylene, carbon dioxide, etc., are collected as the initial input of the model. Since the original gas data contains complex dynamic characteristics and multiple frequency components, to improve the processing ability of the model, these data are subjected to time-frequency synchronous decomposition by the VMD method. VMD can decompose the gas data into multiple intrinsic mode functions (IMFs) with different frequencies, and each IMF corresponds to a relatively independent frequency range, thus effectively separating the noise and trend components.

[0142] Next, each decomposed IMF is separately input into an independent LSTM network model for prediction. Through the interaction of the input gate, forget gate, and output gate, LSTM can capture the time dependence and non-linear relationship of the IMF sequence, thereby learning the dynamic characteristics of each mode function. The LSTM network predicts the IMF values in the future time period by gradually adjusting the weights and biases.

[0143] Finally, the predicted results of the IMFs output by all LSTM models are combined, and the predicted values of each mode are gradually superimposed to restore the overall prediction result of the dissolved gas concentration. The combined result can comprehensively reflect the change trend of various dissolved gases in the future time period, providing an accurate assessment of the transformer state. This framework organically combines the advantages of VMD and LSTM, not only being able to effectively process non-linear and non-stationary data, but also significantly improving the prediction accuracy, real-time performance, and stability, providing a reliable technical support for the fault warning and condition monitoring of transformer equipment.

[0144] Taking hydrogen as an example, the model training results are as Figures 7 to 11 shown.

[0145] Figure 7 This is the convergence curve of the LSTM model, showing the change trend of the loss function during the training process of the LSTM model. The horizontal axis represents the number of training iterations (Epoch), and the vertical axis represents the loss value (Loss). It can be seen from the figure that the loss value of the model drops rapidly in the initial stage, indicating that the model is continuously learning data features and optimizing parameters. After 50 iterations, the loss value tends to be stable, indicating that the model has achieved a good convergence effect, the training is stable, and the prediction results are reliable.

[0146] Figure 8 This is the fitting effect of the training set of intrinsic mode function 1 after hydrogen decomposition, showing the fitting effect of the LSTM model on the first IMF (intrinsic mode function) of hydrogen. The blue line is the actual value (Actual), and the red line is the predicted value (Predicted). It can be seen from the figure that the model can fit the actual value well, especially in the region with large fluctuations, and the deviation between the predicted value and the actual value is very small, indicating that the LSTM model captures the dynamic changes of the mode characteristics more accurately.

[0147] Figure 9 For the fitting effect of the training set of the second intrinsic mode function after hydrogen decomposition, it shows the fitting result of the LSTM model for the second IMF. Compared with the first IMF, the second mode has a higher frequency and a larger fluctuation amplitude. As can be seen from the figure, the predicted values (red line) are highly consistent with the actual values (blue line), indicating that the model can handle high-frequency characteristics and better restore the dynamic changes in the signal.

[0148] Figure 10 For the fitting effect of the training set of the third intrinsic mode function after hydrogen decomposition, it demonstrates the model's learning ability for higher-frequency modes. Although the fluctuations of this mode are relatively complex, the predicted values still highly coincide with the actual values, indicating that the LSTM model has good prediction ability for data with multi-frequency and multi-dynamic characteristics.

[0149] Figure 11 For the hydrogen prediction result, it shows the overall prediction result of the hydrogen concentration. The blue line is the actual concentration value, and the red line is the predicted concentration value. As can be seen from the figure, the predicted values can better fit the actual values. Especially at some mutation points (such as near the 25th point), the model's response to sudden changes is relatively accurate. It indicates that after decomposing the signal, the VMD-LSTM model can accurately predict the overall change trend of the hydrogen concentration through multi-modal fitting and merging, providing a reliable basis for transformer fault warning.

[0150] In order to improve the prediction accuracy of non-linear and non-stationary time series based on the combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), realize the real-time monitoring and fault warning of dissolved gas data, enhance the stability of the model in dynamic scenarios, ensure that the device continuously provides reliable prediction results under different operating conditions, improve the fault warning ability and the safety of system operation, this application provides an embodiment of a prediction device for dissolved gas content in transformer oil for implementing all or part of the content of the prediction method for the dissolved gas content in transformer oil. See Figure 12 The prediction device for dissolved gas content in transformer oil specifically includes the following:

[0151] The data input module 1101 is used to: input the data to be measured of the dissolved gas in the transformer oil into the pre-trained combined model of variational mode decomposition VMD and long short-term memory network LSTM to obtain the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil, where the combined model decomposes the oil chromatogram sequence of the historical data of the dissolved gas in the transformer oil into multiple mode functions with different frequencies through VMD, and captures the temporal characteristics and non-linear relationships of multi-dimensional data through LSTM for optimization;

[0152] A result determination module 1102, configured to: combine the intrinsic mode function prediction results of the dissolved gases in each type of oil to obtain corresponding prediction results.

[0153] According to any implementation manner of the present application, refer to Figure 13 , the training process of the pre-trained combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM) includes:

[0154] A data acquisition module 0001, configured to: acquire historical data of dissolved gases in transformer oil within a preset time range, and perform missing value replacement processing on the historical data based on the forward method;

[0155] A data decomposition module 0002, configured to: perform time-frequency synchronous decomposition on the sequences of different gases in the historical data through VDM to obtain corresponding multiple frequency-aligned multivariate intrinsic mode functions (IMFs) and residuals, where the frequency alignment indicates that each gas is aligned inside the multiple IMFs decomposed from other influencing factor sequences;

[0156] A model construction module 0003, configured to: perform Z-score normalization processing on the multiple IMFs, and construct corresponding multiple multi-step LSTM network models according to the normalized multiple IMFs.

[0157] According to any implementation manner of the present application, refer to Figure 14 , the data decomposition module includes:

[0158] A transformation unit 0002A, configured to: perform Hilbert transformation on each dissolved gas sequence to extract unilateral spectrum information;

[0159] An alignment unit 0002B, configured to: move the spectrum of each dissolved gas component to the baseband region through modulation and align the center frequencies;

[0160] A bandwidth unit 0002C, configured to: perform regularization processing on the gradient of the modulated signal, and apply the Gaussian smoothing method to obtain the frequency bandwidth of each component;

[0161] An iteration unit 0002D, configured to: input each processed dissolved gas sequence into the VMD model, and decompose it into multiple IMFs through an iterative optimization method, where each IMF represents an independent signal component in a frequency range, and at the same time, residual information is retained.

[0162] According to any implementation manner of the present application, refer to Figure 15 , the model construction module performs Z-score normalization processing on the multiple IMFs, including:

[0163] Feature acquisition unit 0003A, configured to: calculate the mean and standard deviation of each IMF to obtain its data distribution characteristics;

[0164] Distribution adjustment unit 0003B, configured to: subtract the mean from each data point and divide the mean - removed data by the standard deviation to adjust it to a standard normal distribution state.

[0165] According to any implementation manner of the present application, the model construction module constructs corresponding multiple multi - step LSTM network models based on the standardized multiple IMFs, including:

[0166] Model construction unit 0003C, configured to: use the standardized multiple IMF sequences as input data, use the corresponding IMF sequences in the next week as the target output, independently construct an LSTM network model for each IMF component, and initialize the weight parameters and biases of the input gate, forget gate, and output gate, and perform model training;

[0167] Parameter adjustment unit 0003D, configured to: input each IMF sequence into the corresponding LSTM model. The model learns the relationship between the current sequence and the future trend, and gradually adjusts the parameters using an iterative optimization method to generate the multiple multi - step LSTM network models that can accurately capture the characteristics of the time series.

[0168] According to any implementation manner of the present application, see Figure 16 , the merging of the intrinsic mode function prediction results of the dissolved gases in each type of oil to obtain the corresponding prediction results includes:

[0169] Data reduction unit 1103A, configured to: use the reverse normalization technique to restore the standardized IMF results predicted by the LSTM model to the original numerical range through weighted calculation;

[0170] Data merging unit 1103B, configured to: gradually add the predicted values of each normalized IMF according to the gas types to which they belong to obtain the corresponding prediction results.

[0171] As can be seen from the above description, the prediction device for the content of dissolved gases in transformer oil provided by the embodiments of the present application can improve the prediction accuracy of non - linear and non - stationary time series based on the combined model of variational mode decomposition (VMD) and long short - term memory network (LSTM), realize real - time monitoring of dissolved gas data and fault warning, enhance the stability of the model in dynamic scenarios, ensure that the device continuously provides reliable prediction results under different operating conditions, and improve the fault warning ability and the safety of system operation.

[0172] At the hardware level, in order to improve the prediction accuracy of non-linear and non-stationary time series based on a combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), realize real-time monitoring and fault warning of dissolved gas data, enhance the stability of the model in dynamic scenarios, ensure that the device continuously provides reliable prediction results under different operating conditions, improve the fault warning ability and the safety of system operation, this application provides an embodiment of an electronic device for implementing all or part of the content in the prediction method of the dissolved gas content in transformer oil. The electronic device specifically includes the following content:

[0173] A processor, a memory, a communications interface, and a bus; wherein, the processor, the memory, and the communications interface complete communication with each other through the bus; the communications interface is used to realize information transmission between the prediction device of the dissolved gas content in transformer oil and related devices such as the core business system, the user terminal, and the relevant database, etc. This logic controller can be a desktop computer, a tablet computer, a mobile terminal, etc., and this embodiment is not limited thereto. In this embodiment, this logic controller can be implemented with reference to the embodiments of the prediction method of the dissolved gas content in transformer oil and the embodiments of the prediction device of the dissolved gas content in transformer oil, and the content thereof is incorporated herein, and the repeated parts will not be elaborated.

[0174] It can be understood that the user terminal may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. Among them, the smart wearable device may include smart glasses, a smart watch, a smart bracelet, etc.

[0175] In practical applications, part of the prediction method of the dissolved gas content in transformer oil can be executed on the electronic device side as described above, or all operations can be completed in the client device. Specifically, it can be selected according to the processing capacity of the client device and the limitations of the user usage scenario, etc. This application does not make any limitations in this regard. If all operations are completed in the client device, the client device may further include a processor.

[0176] The above-mentioned client device may have a communication module (i.e., communication unit), which can communicate with a remote server to achieve data transmission with the server. The server may include a server on the side of the task scheduling center, and in other implementation scenarios, it may also include a server of an intermediate platform, such as a server of a third-party server platform with a communication link to the task scheduling center server. The server may include a single computer device, or a server cluster composed of multiple servers, or a server structure of a distributed device.

[0177] Figure 17 It is a schematic block diagram of the system composition of the electronic device 9600 according to an embodiment of the present application. As Figure 17 shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It should be noted that this Figure 17 is exemplary; other types of structures can also be used to supplement or replace this structure to achieve telecommunication functions or other functions.

[0178] In one embodiment, the function of the prediction method for the content of dissolved gases in transformer oil can be integrated into the central processing unit 9100. Among them, the central processing unit 9100 can be configured to perform the following controls:

[0179] Step S101: Input the data to be measured of the dissolved gases in the transformer oil into a pre-trained combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM) to obtain the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil. Among them, the combined model decomposes the oil chromatogram sequence of the historical data of the dissolved gases in the transformer oil into multiple mode functions with different frequencies through VMD, and captures the temporal characteristics and non-linear relationships of multi-dimensional data through LSTM for optimization.

[0180] Step S102: Combine the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil to obtain the corresponding prediction results.

[0181] As can be seen from the above description, the electronic device provided by the embodiment of the present application improves the prediction accuracy of non-linear and non-stationary time series based on a combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), realizes real-time monitoring and fault warning of dissolved gas data, enhances the stability of the model in dynamic scenarios, ensures that the device continuously provides reliable prediction results under different operating conditions, and improves the fault warning ability and the safety of system operation.

[0182] In another embodiment, the prediction device for the content of dissolved gases in transformer oil can be separately configured from the central processing unit 9100. For example, the prediction device for the content of dissolved gases in transformer oil can be configured as a chip connected to the central processing unit 9100, and the function of the prediction method for the content of dissolved gases in transformer oil can be realized through the control of the central processing unit.

[0183] As Figure 17 shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It should be noted that the electronic device 9600 does not necessarily have to include Figure 17 all the components shown in Figure 17 ; in addition, the electronic device 9600 may further include Figure 17 components not shown in

[0184] As Figure 17 shown, the central processing unit 9100, sometimes also referred to as a controller or operation control, may include a microprocessor or other processor devices and / or logic devices. The central processing unit 9100 receives inputs and controls the operations of the various components of the electronic device 9600.

[0185] Among them, the memory 9140, for example, may be one or more of a buffer, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory, or other suitable devices. The above information related to failures can be stored, and in addition, programs for executing relevant information can also be stored. And the central processing unit 9100 can execute the program stored in the memory 9140 to implement information storage or processing, etc.

[0186] The input unit 9120 provides inputs to the central processing unit 9100. The input unit 9120 is, for example, a key or a touch input device. The power supply 9170 is used to supply power to the electronic device 9600. The display 9160 is used to display display objects such as images and texts. The display may be, for example, an LCD display, but is not limited thereto.

[0187] The memory 9140 may be a solid-state memory. For example, a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It may also be such a memory that stores information even when powered off, can be selectively erased and has more data stored. Examples of such a memory are sometimes referred to as EPROMs, etc. The memory 9140 may also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142, which is used to store application programs and function programs or the processes for operating the electronic device 9600 through the central processing unit 9100.

[0188] The memory 9140 may further include a data storage unit 9143 for storing data such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers of the electronic device for communication functions and / or for performing other functions of the electronic device (such as a messaging application, an address book application, etc.).

[0189] The communication module 9110 is a transmitter / receiver 9110 that transmits and receives signals via the antenna 9111. The communication module (transmitter / receiver) 9110 is coupled to the central processor 9100 to provide input signals and receive output signals, which may be the same as in the case of a conventional mobile communication terminal.

[0190] Based on different communication technologies, multiple communication modules 9110 may be provided in the same electronic device, such as a cellular network module, a Bluetooth module, and / or a wireless local area network module, etc. The communication module (transmitter / receiver) 9110 is also coupled to the speaker 9131 and the microphone 9132 via the audio processor 9130 to provide an audio output via the speaker 9131 and receive an audio input from the microphone 9132, thereby implementing normal telecommunication functions. The audio processor 9130 may include any suitable buffers, decoders, amplifiers, etc. In addition, the audio processor 9130 is also coupled to the central processor 9100, so that it is possible to record on the local machine through the microphone 9132 and play the sound stored on the local machine through the speaker 9131.

[0191] Embodiments of the present application also provide a computer-readable storage medium capable of implementing all steps in the method for predicting the content of dissolved gases in transformer oil with the execution subject being a server or a client in the above embodiments. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, all steps in the method for predicting the content of dissolved gases in transformer oil with the execution subject being a server or a client in the above embodiments are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0192] Step S101: Input the data to be measured of the dissolved gases in the transformer oil into a combined model of a pre-trained variational mode decomposition (VMD) and a long short-term memory network (LSTM) to obtain the prediction results of the intrinsic mode functions of each type of dissolved gas in the oil, where the combined model is obtained by decomposing the oil chromatogram sequence of the historical data of the dissolved gases in the transformer oil into multiple mode functions with different frequencies through VMD, and capturing the temporal characteristics and non-linear relationships of multi-dimensional data through LSTM for optimization.

[0193] Step S102: Combine the prediction results of the intrinsic mode functions of the dissolved gases in each type of oil to obtain the corresponding prediction results.

[0194] As can be seen from the above description, the computer-readable storage medium provided by the embodiments of the present application improves the prediction accuracy of non-linear and non-stationary time series based on a combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), realizes real-time monitoring and fault warning of dissolved gas data, enhances the stability of the model in dynamic scenarios, ensures that the device continuously provides reliable prediction results under different operating conditions, and improves the fault warning ability and the safety of system operation.

[0195] An embodiment of the present application also provides a computer program product capable of implementing all steps in the prediction method for the content of dissolved gases in transformer oil with the execution subject being a server or a client in the above embodiments. When the computer program / instructions are executed by a processor, the steps of the prediction method for the content of dissolved gases in transformer oil are implemented. For example, the computer program / instructions implement the following steps:

[0196] Step S101: Input the data to be measured of the dissolved gases in transformer oil into a pre-trained combined model of variational mode decomposition VMD and long short-term memory network LSTM to obtain the prediction results of the intrinsic mode functions of the dissolved gases in each type of oil. Among them, the combined model is obtained by decomposing the oil chromatogram sequence of the historical data of the dissolved gases in transformer oil into multiple mode functions with different frequencies through VMD, and capturing the time series characteristics and non-linear relationships of multi-dimensional data through LSTM for optimization.

[0197] Step S102: Combine the prediction results of the intrinsic mode functions of the dissolved gases in each type of oil to obtain the corresponding prediction results.

[0198] As can be seen from the above description, the computer program product provided by the embodiments of the present application improves the prediction accuracy of non-linear and non-stationary time series based on a combined model of variational mode decomposition (VMD) and long short-term memory network (LSTM), realizes real-time monitoring and fault warning of dissolved gas data, enhances the stability of the model in dynamic scenarios, ensures that the device continuously provides reliable prediction results under different operating conditions, and improves the fault warning ability and the safety of system operation.

[0199] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, an apparatus, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can 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.) that contain computer-usable program code.

[0200] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (apparatuses), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0201] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0202] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, so that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process. Therefore, the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0203] Specific embodiments are applied in the present application to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for predicting dissolved gas content in transformer oil, characterized in that: The method comprises: The data to be tested of the dissolved gas in the transformer oil is input into a pre-trained combination model of variational mode decomposition (VMD) and long short-term memory (LSTM) network to obtain the prediction results of the intrinsic mode function of each type of dissolved gas in the oil, wherein the combination model is obtained by decomposing the oil chromatogram sequence of the historical data of the dissolved gas in the transformer oil into a plurality of mode functions with different frequencies through VMD, and capturing the time series characteristics and nonlinear relationship of the multidimensional data through LSTM for optimization; The prediction results of the intrinsic mode functions of the dissolved gas in each type of oil are combined to obtain corresponding prediction results.

2. The method for predicting dissolved gas content in transformer oil according to claim 1, characterized in that: The training process of the pre-trained combined model of variational mode decomposition VMD and long short-term memory network LSTM includes: Collect historical data of dissolved gas in transformer oil within a preset time range, and perform missing value replacement processing on the historical data based on the forward method; By using VDM, the sequences of different gases in the historical data are decomposed synchronously in time and frequency to obtain corresponding multiple frequency-aligned multivariate intrinsic mode functions IMFs and residuals, wherein the frequency alignment indicates that the multiple IMFs obtained by decomposing each gas with other influencing factor sequences are internally aligned; The multiple IMFs are subjected to Z-score standardization processing, and corresponding multiple multi-step LSTM network models are constructed according to the multiple IMFs after the standardization processing.

3. The method for predicting dissolved gas content in transformer oil according to claim 2, characterized in that: The VDM is used to perform time-frequency synchronous decomposition on the sequences of different gases in the historical data to obtain corresponding multiple frequency-aligned multivariate intrinsic mode functions IMFs and residuals, including: Hilbert transform is performed on each dissolved gas sequence to extract the one-sided spectrum information; The spectrum of each dissolved gas component is moved to the baseband region by modulation and the center frequency is aligned; Regularization is performed on the modulated signal gradient, and a Gaussian smoothing method is applied to obtain the frequency bandwidth of each component; Each processed dissolved gas sequence is input into the VMD model and decomposed into multiple IMFs through an iterative optimization method. Each IMF represents an independent signal component in a frequency range while retaining the residual information.

4. The method for predicting dissolved gas content in transformer oil according to claim 2, characterized in that: The Z-score standardization process of the multiple IMFs includes: Calculate the mean and standard deviation of each IMFs to obtain its data distribution characteristics; The mean is subtracted from each data point and the data is divided by the standard deviation to adjust it to a standard normal distribution.

5. The method for predicting dissolved gas content in transformer oil according to claim 2, characterized in that: The method of constructing corresponding multiple multi-step LSTM network models according to the multiple IMFs after the standardized processing includes: The standardized multiple IMFs sequences are used as input data, and the IMFs sequence corresponding to the next week is used as the target output. An LSTM network model is independently constructed for each IMFs component, and the weight parameters and bias of the input gate, the forget gate, and the output gate are initialized to prepare for model training; Each IMFs sequence is input into the corresponding LSTM model. The model learns the relationship between the current sequence and the future trend, and gradually adjusts the parameters using an iterative optimization method to generate the multiple multi-step LSTM network models that can accurately capture the characteristics of the time series.

6. The method for predicting dissolved gas content in transformer oil according to claim 2, characterized in that: The prediction results of the intrinsic mode functions of the dissolved gas in each type of oil are combined to obtain corresponding prediction results, including: Using the reverse normalization technique, the standardized IMFs results predicted by the LSTM model are restored to the original value range through weighted calculation; The normalized prediction values ​​of each IMFs are gradually added up according to the gas type to which they belong to obtain the corresponding prediction results.

7. A device for predicting dissolved gas content in transformer oil, characterized in that: The device comprises: A data input module is used to: input the data to be tested of the dissolved gas in the transformer oil into a pre-trained combination model of variational mode decomposition VMD and long short-term memory network LSTM, to obtain the prediction result of the intrinsic mode function of each type of dissolved gas in the oil, wherein the combination model is obtained by decomposing the oil chromatogram sequence of the historical data of the dissolved gas in the transformer oil into a plurality of mode functions with different frequencies through VMD, and capturing the time series characteristics and nonlinear relationship of the multidimensional data through LSTM for optimization; The result determination module is used to combine the prediction results of the intrinsic mode functions of the dissolved gas in each type of oil to obtain corresponding prediction results.

8. The device for predicting dissolved gas content in transformer oil according to claim 7, characterized in that: The training process of the pre-trained combined model of variational mode decomposition VMD and long short-term memory network LSTM includes: The data acquisition module is used to collect historical data of dissolved gas in transformer oil within a preset time range, and perform missing value replacement processing on the historical data based on the forward method; A data decomposition module is used to: perform time-frequency synchronous decomposition of the sequences of different gases in the historical data through VDM to obtain corresponding multiple frequency-aligned multivariate intrinsic mode functions IMFs and residuals, wherein the frequency alignment indicates that the multiple IMFs obtained by decomposing each gas with other influencing factor sequences are internally aligned; The model building module is used to: perform Z-score standardization on the multiple IMFs, and build corresponding multiple multi-step LSTM network models according to the multiple IMFs after the standardization.

9. The device for predicting dissolved gas content in transformer oil according to claim 8, characterized in that: The data decomposition module comprises: A transformation unit, used for: performing Hilbert transformation on each dissolved gas sequence to extract single-side spectrum information; An alignment unit, used to: move the spectrum of each dissolved gas component to the baseband region by modulation and align the center frequency; A bandwidth unit is used to: perform regularization processing on the modulated signal gradient and apply a Gaussian smoothing device to obtain the frequency bandwidth of each component; The iterative unit is used for: inputting each dissolved gas sequence after processing into the VMD model, decomposing it into a plurality of IMFs through an iterative optimization device, each IMF representing an independent signal component in a frequency range, and retaining residual information at the same time.

10. The device for predicting dissolved gas content in transformer oil according to claim 8, characterized in that: The model building module performs Z-score standardization processing on the multiple IMFs, including: A feature acquisition unit is used to: calculate the mean value and standard deviation of each IMFs to obtain its data distribution characteristics; The distribution adjustment unit is used to: subtract the mean value from each data point and divide the data after the mean value is removed by the standard deviation to adjust the data to a standard normal distribution state.

11. The device for predicting dissolved gas content in transformer oil according to claim 8, characterized in that: The model building module builds corresponding multiple multi-step LSTM network models according to the multiple IMFs after standardization, including: A model building unit is used to: use the standardized multiple IMFs sequences as input data, use the IMFs sequence corresponding to the next week as the target output, independently build an LSTM network model for each IMFs component, and initialize the weight parameters and bias of the input gate, the forget gate and the output gate to perform model training; The parameter adjustment unit is used to: input each IMFs sequence into the corresponding LSTM model, and the model gradually adjusts the parameters by using an iterative optimization device by learning the relationship between the current sequence and the future trend, so as to generate the multiple multi-step LSTM network models that can accurately capture the characteristics of the time series.

12. The device for predicting dissolved gas content in transformer oil according to claim 8, characterized in that: The result determination module combines the prediction results of the intrinsic mode functions of the dissolved gas in each type of oil to obtain corresponding prediction results, including: The data restoration unit is used to: use the reverse normalization technology to restore the standardized IMFs results predicted by the LSTM model to the original value range through weighted calculation; The data merging unit is used to gradually add up the normalized prediction values ​​of each IMFs according to the gas type to which it belongs, so as to obtain the corresponding prediction result.

13. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method for predicting the dissolved gas content in transformer oil according to any one of claims 1 to 6 are implemented.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the dissolved gas content in transformer oil according to any one of claims 1 to 6 are implemented.

15. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method for predicting the dissolved gas content in transformer oil according to any one of claims 1 to 6 are implemented.

Images