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

Abstract

The invention provides a method and device for predicting the concentration of dissolved gas in transformer oil, and the method comprises the steps: determining to-be-detected gas in response to an obtained historical time sequence data set of the concentration of the dissolved gas in the transformer oil, and determining at least one model input gas according to the to-be-detected gas; the historical time sequence data set of the dissolved gas concentration comprises historical time sequence data corresponding to various gases; performing multivariate variational mode decomposition on the historical time sequence data corresponding to the at least one model input gas to obtain a mode component corresponding to each model input gas; and inputting the decomposed modal component into a pre-trained concentration prediction model, and taking the output of the concentration prediction model as the predicted concentration of the to-be-detected gas at the next moment.

Description

Technical Field

[0001] The invention relates to the field of transformers, and in particular to a method and a device for predicting dissolved gas concentration in transformer oil. Background Art

[0002] During operation, transformers and fiber insulation materials age and decompose due to the catalytic effects of moisture, oxygen, heat, and materials such as copper and iron, which generate gases. Most of these gases dissolve in the oil. Under normal circumstances, the rate of gas generation is relatively slow. However, when an initial fault occurs or a new fault condition is formed inside the transformer, the rate and amount of gas generation will increase significantly, and most initial defects will show early signs. Therefore, through appropriate analysis of the gases generated in the transformer oil, the concentration prediction of dissolved gases in the transformer oil is of great significance for transformer fault early warning.

[0003] In the related art, the gas concentration is predicted by a prediction method based on a mathematical statistics model or a machine learning model. Among them, the mathematical statistics method performs statistical analysis on a large amount of historical data on the concentration of dissolved gas in oil, establishes a model using statistical characteristics and laws, and then predicts the change of the concentration of dissolved gas in transformer oil based on these models. The mathematical statistics method is simple and easy to use, and has a fast calculation speed, but it is difficult to predict nonlinear data with strong randomness. The machine learning method uses a large amount of historical data and the complex relationship between features to learn and analyze, and establish a nonlinear prediction model. Although this method can fit nonlinear data well, it generally cannot take into account the correlation between time series data, and the prediction accuracy needs to be improved. Summary of the invention

[0004] In view of this, the present invention provides a method and device for predicting the concentration of dissolved gas in transformer oil, which can solve the deficiencies existing in the related art.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] According to a first aspect of the present invention, a method for predicting dissolved gas concentration in transformer oil is proposed, the method comprising:

[0007] In response to the acquired historical time series data set of dissolved gas concentration in transformer oil, a gas to be measured is determined, and at least one model input gas is determined according to the gas to be measured; the historical time series data set of dissolved gas concentration includes historical time series data corresponding to a plurality of gases;

[0008] Performing multivariate variational modal decomposition on the historical time series data corresponding to the at least one model input gas to obtain a modal component corresponding to each model input gas;

[0009] The modal components obtained by decomposition are input into a pre-trained concentration prediction model, and the output of the concentration prediction model is used as the predicted concentration of the gas to be measured at the next moment; wherein the concentration prediction model includes an input layer, multiple residual blocks, and an output layer, the input layer is used to extract the time series characteristics of the modal components obtained by decomposition according to dilated causal convolution, each residual block is used to perform weight normalization, function activation, and random inactivation processing on the output of the previous residual block, and the output layer is used to superimpose and reconstruct multiple modal components corresponding to the gas to be measured to obtain the predicted concentration of the gas to be measured at the next moment.

[0010] According to a second aspect of the present invention, a device for predicting dissolved gas concentration in transformer oil is provided, the device comprising:

[0011] A determination unit: in response to the acquired historical time series data set of dissolved gas concentration in transformer oil, determines a gas to be measured, and determines at least one model input gas according to the gas to be measured; the historical time series data set of dissolved gas concentration includes historical time series data corresponding to a plurality of gases;

[0012] Decomposition unit: performing multivariate variational modal decomposition on the historical time series data corresponding to the at least one model input gas to obtain a modal component corresponding to each model input gas;

[0013] Prediction unit: input the modal components obtained by decomposition into a pre-trained concentration prediction model, and use the output of the concentration prediction model as the predicted concentration of the gas to be measured at the next moment; wherein the concentration prediction model includes an input layer, multiple residual blocks, and an output layer, the input layer is used to extract the time series characteristics of the modal components obtained by the decomposition according to the dilated causal convolution, each residual block is used to perform weight normalization, function activation, and random inactivation processing on the output of the previous residual block, and the output layer is used to superimpose and reconstruct multiple modal components corresponding to the gas to be measured to obtain the predicted concentration of the gas to be measured at the next moment.

[0014] According to a third aspect of the present invention, an electronic device is provided, comprising:

[0015] processor;

[0016] a memory for storing processor-executable instructions;

[0017] The processor implements the steps of the method described in the first aspect by running the executable instructions.

[0018] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the method described in the first aspect are implemented.

[0019] It can be seen from the above technical solutions that the method for predicting dissolved gas concentration in transformer oil provided by the present invention, on the one hand, uses the modal components generated based on the time series data of dissolved gas concentration as the model input data of the concentration prediction model, taking into account the fit with historical data and the correlation with time series data, thereby improving the accuracy of gas concentration prediction; on the other hand, it is not limited to using a single gas for concentration prediction, but reduces the complexity of the sequence through multivariate variational modal decomposition, and overcomes the frequency mismatch problem after decomposition of different input characteristic gas sequences, thereby realizing the synchronous decomposition of multiple model input gases, fully mining the key information contained between gases, and improving the prediction accuracy and prediction stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of a method for predicting dissolved gas concentration in transformer oil provided by an exemplary embodiment.

[0021] Figure 2 It is a schematic diagram of correlation coefficients of various gases in a dissolved gas provided by an exemplary embodiment.

[0022] Figure 3 It is a schematic diagram of a concentration prediction model provided by an exemplary embodiment.

[0023] Figure 4 is a schematic diagram of an improved fireworks algorithm provided by an exemplary embodiment.

[0024] Figure 5 It is a schematic structural diagram of a device provided by an exemplary embodiment.

[0025] Figure 6 It is a block diagram of a device for predicting dissolved gas concentration in transformer oil provided by an exemplary embodiment. DETAILED DESCRIPTION

[0026] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0027] It should be noted that: in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in the present invention. In some other embodiments, the steps included in the method may be more or less than those described in the present invention. In addition, a single step described in the present invention may be decomposed into multiple steps for description in other embodiments; and multiple steps described in the present invention may be combined into a single step for description in other embodiments.

[0028] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0029] To further illustrate the present invention, the following examples are provided:

[0030] During operation, transformers and fiber insulation materials age and decompose due to the catalytic effects of moisture, oxygen, heat, and materials such as copper and iron, which generate gases. Most of these gases dissolve in the oil. Under normal circumstances, the rate of gas generation is relatively slow. However, when an initial fault occurs or a new fault condition is formed inside the transformer, the rate and amount of gas generation will increase significantly, and most initial defects will show early signs. Therefore, through appropriate analysis of the gases generated in the transformer oil, the concentration prediction of dissolved gases in the transformer oil is of great significance for transformer fault early warning.

[0031] In the related art, the gas concentration is predicted by a prediction method based on a mathematical statistics model or a machine learning model. Among them, the mathematical statistics method performs statistical analysis on a large amount of historical data on the concentration of dissolved gas in oil, establishes a model using statistical characteristics and laws, and then predicts the change of the concentration of dissolved gas in transformer oil based on these models. The mathematical statistics method is simple and easy to use, and has a fast calculation speed, but it is difficult to predict nonlinear data with strong randomness. The machine learning method uses a large amount of historical data and the complex relationship between features to learn and analyze, and establish a nonlinear prediction model. Although this method can fit nonlinear data well, it generally cannot take into account the correlation between time series data, and the prediction accuracy needs to be improved.

[0032] In order to solve the deficiencies existing in the related art, the present invention proposes a method for predicting the concentration of dissolved gas in transformer oil.

[0033] Figure 1 FIG. 1 is a flow chart of a method for predicting dissolved gas concentration in transformer oil provided by an exemplary embodiment. Figure 1 As shown, the method may include the following steps:

[0034] Step 102, in response to the acquired historical time series data set of dissolved gas concentration in transformer oil, determine a gas to be measured, and determine at least one model input gas according to the gas to be measured; the historical time series data set of dissolved gas concentration includes historical time series data corresponding to a plurality of gases.

[0035] Transformers are core equipment of power grids, and their operational reliability directly affects the safety and stability of power grids. Most transformer failures gradually evolve from small internal defects, which are difficult to monitor directly through external measurement methods. Discharge or overheating caused by internal defects of transformers can cause chemical reactions between insulating oil and insulating paper to generate specific types of gases, such as hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO), carbon dioxide (CO2), etc. These gases can dissolve in transformer oil and become dissolved gases.

[0036] The historical time series data set of dissolved gas concentration includes historical time series data corresponding to gas concentrations of multiple gases, and the historical time series data of each gas is composed of gas concentration data of the gas at multiple historical moments.

[0037] The gas to be measured can be one of the multiple gases contained in the dissolved gas, and is the gas whose gas concentration is to be predicted at the next moment. The model input gas is the gas corresponding to the model input data to be input into the model, and the model input gas belongs to the multiple gases of the dissolved gas. The model input gas at least includes the gas to be measured itself.

[0038] In one embodiment, determining at least one model input gas based on the gas to be tested includes: performing grey correlation analysis on various gases contained in the dissolved gas based on the historical time series data set; and determining a gas among the various gases whose correlation coefficient with the gas to be tested is greater than a first preset threshold as a model input gas.

[0039] Grey Relational Analysis (GRA) is a systematic analysis method for dealing with small sample and poor information problems. Its core idea is to measure the degree of correlation between factors by comparing the geometric similarity between data sequences. Assume that the gas to be tested is the parent sequence, denoted as x0, and the other gases in the dissolved gas are subsequences, denoted as x1, x2, x3, ..., x i . Calculate the minimum value a and maximum value b of the absolute difference matrix between the parent sequence and the subsequence. Further, calculate the grey correlation degree Y(x0,x i). The specific formula is as follows:

[0040] a=min i min k |x0(k)-x i (k)|;

[0041] b=max i max k |x0(k)-x i (k)|;

[0042]

[0043] Here, ρ is the resolution, which is generally set to 0.5.

[0044] Figure 2 FIG. 1 is a schematic diagram of the correlation coefficients of various gases in a dissolved gas provided by an exemplary embodiment. Figure 2 As shown, taking H2 as an example, the correlation coefficients between various gases (H2, CH4, C2H6, C2H4, C2H2, CO, CO2) contained in the dissolved gas and H2 are 1, 0.66, 0.74, 0.59, 0.72, 0.85, and 0.81 respectively. If the first preset threshold is set to 0.8, then the gas with a correlation coefficient greater than 0.8 with the gas to be predicted is selected as the model input gas of the prediction model, and the model input gas includes H2, CO, and CO2.

[0045] In this embodiment, the higher the correlation coefficient between other gases and the gas to be measured, the greater the grey correlation between the gas and the gas to be measured. Selecting highly correlated gases to jointly predict the gas to be measured can achieve the prediction of gas concentration by combining multiple gas data, fully tapping the key information contained between gases, and thus improving the accuracy of gas concentration prediction.

[0046] Step 104: Perform multivariate variational modal decomposition on the historical time series data corresponding to the at least one model input gas to obtain a modal component corresponding to each model input gas.

[0047] Multivariate Variational Mode Decomposition (MVMD) is a signal processing method that is mainly used to analyze and process multivariate signals, such as multivariate time series signals. This method reveals the intrinsic laws and structures of signals by decomposing multivariate signals into a set of Intrinsic Mode Functions (IMFs). The core of MVMD is that it can adaptively realize the frequency domain decomposition of signals and the effective separation of components. This method is particularly effective when processing multi-channel or multi-dimensional data, and can better understand and analyze multiple modal components in the data.

[0048] Specifically, if the model input gases are three types, the data to be input into the model is three-dimensional data, and the multi-dimensional data is simultaneously decomposed to obtain a series of intrinsic mode functions (IMF) components with the same component frequency, that is, modal components.

[0049] Combined with the formula, if the C-dimensional input signal x(t) is extracted into k multi-modulation oscillations u k (t).

[0050]

[0051] u k (t)=[u1(t),u2(t),...,u c (t)] T ;

[0052] Decomposition needs to meet two conditions: 1. The mode u obtained by decomposition k (t) is equal to the original input signal x(t); 2. The sum of the modal bandwidths obtained by decomposition is the smallest. These two conditions are converted into an unconstrained variational problem and solved by alternating updates using the multiplier alternating direction method. After MVMD decomposition, the number of IMFs obtained by decomposing the characteristic gas in each dimension is the same, and the IMF components of the same level have the same frequency scale.

[0053] In this embodiment, the sequence complexity is reduced by multivariate variational mode decomposition, and the frequency mismatch problem after decomposition of different input characteristic gas sequences is overcome, thereby achieving simultaneous decomposition of multiple model input gases, fully mining the key information contained between gases, and improving prediction accuracy and prediction stability.

[0054] Step 106, input the modal components obtained by decomposition into a pre-trained concentration prediction model, and use the output of the concentration prediction model as the predicted concentration of the gas to be measured at the next moment; wherein the concentration prediction model includes an input layer, multiple residual blocks, and an output layer, the input layer is used to extract the time series characteristics of the modal components obtained by the decomposition according to the dilated causal convolution, each residual block is used to perform weight normalization, function activation, and random inactivation processing on the output of the previous residual block, and the output layer is used to superimpose and reconstruct multiple modal components corresponding to the gas to be measured to obtain the predicted concentration of the gas to be measured at the next moment.

[0055] The concentration prediction model is a deep learning model built on the temporal convolutional network. The temporal convolutional network (TCN) is a deep learning architecture for sequence modeling and prediction. It consists of dilated, causal 1D convolutional layers with the same input and output lengths. TCN uses dilated causal convolutions to process time series data. This convolution ensures that the output depends only on the current and past inputs, maintaining the causality of the time series. Through the combination of layers with different expansion coefficients, TCN is able to capture features at different time scales. In TCN, the input and output lengths of each layer are the same, which enables the network to maintain the original time steps of the time series. In addition, TCN draws on the design of the residual network (ResNet) and solves the gradient vanishing problem in deep networks through residual connections, thereby enhancing the training efficiency and performance of the model.

[0056] The concentration prediction model can predict the gas concentration of the gas to be tested at the next moment based on the time series data of the historical moment. Use TCN to predict each time series data: TCN is mainly composed of dilated causal convolution and residual connection. Its structure includes input layer, multiple residual blocks and output layer, such as Figure 3 As shown, the input is X = (x0, x1, ..., x T ), extract univariate time series features with multiple time steps through dilated causal convolution to obtain F(T). F(T) is processed by weight normalization, ReLU activation function, and Dropout random inactivation layer. Due to the presence of multiple convolutional layers, a residual structure is introduced to prevent gradient disappearance. Each TCN block takes the output of the previous layer as the input of the next layer and repeats the above process to better capture the features and patterns in the time series data. After calculating multiple TCN blocks, the data passes through the fully connected layer to obtain the output y of the last layer T .

[0057] In this embodiment, on the one hand, the modal components generated based on the time series data of dissolved gas concentration are used as model input data of the concentration prediction model, which takes into account both the fit with historical data and the correlation with time series data, thereby improving the accuracy of gas concentration prediction; on the other hand, it is not limited to using a single gas for concentration prediction, but reduces the complexity of the sequence through multivariate variational modal decomposition, and overcomes the frequency mismatch problem after decomposition of different input characteristic gas sequences, thereby achieving simultaneous decomposition of multiple model input gases, fully mining the key information contained between gases, and improving prediction accuracy and prediction stability.

[0058] In one embodiment, the training process of the concentration prediction model includes: obtaining a training sample set, each training sample in the training sample set includes a modal component obtained by decomposing the time series data of a group of model input gases, and the actual concentration of the corresponding gas to be tested in the group of model input gases at the next moment; inputting the training sample set into the model to be trained constructed based on the time domain convolutional network, so that the model to be trained outputs the predicted concentration of the corresponding gas to be tested at the next moment; comparing the actual concentration and the predicted concentration corresponding to each group of training sample sets, and optimizing the parameters of the model to be trained according to the comparison results until the prediction effect of the model to be trained reaches a preset effect.

[0059] Furthermore, the optimizing the parameters of the model to be trained according to the comparison result comprises: optimizing the parameters of the model to be trained according to an improved fireworks algorithm; wherein the improved fireworks algorithm is a combination of an elite selection strategy and an improved explosion mechanism, the elite selection strategy selects elite individuals for explosion operations according to the fitness values ​​of fireworks individuals, and the improved explosion mechanism generates randomly generated displacements with different fireworks explosion step lengths in each dimension based on a taboo search algorithm.

[0060] The Fireworks Algorithm (FWA) is an optimization algorithm based on swarm intelligence. It searches for the optimal solution by simulating the process of fireworks exploding and generating sparks. The core idea of ​​the Fireworks Algorithm is to simulate the explosion of fireworks to generate sparks and find the optimal solution among these sparks. Each firework represents a potential solution, and the sparks generated by the explosion represent new potential solutions near the solution. The improved Fireworks Algorithm (IFA) combines the elite selection strategy and the explosion mechanism, updates the elite selection strategy from the traditional roulette selection strategy to the best selection based on fitness value, and introduces the taboo search algorithm in the explosion mechanism to generate random steps. The specific algorithm flow is as follows: Figure 4 shown.

[0061] After the fireworks population is initialized, the elite selection strategy calculates the fitness values ​​of the fireworks individuals and sorts them, and selects the elite individuals for the explosion operation. After the explosion operation, the elite selection strategy operation is performed on the generated sparks, and the fireworks individuals with better fitness values ​​are selected as elite individuals to enter the next generation of the population.

[0062] The improvement of the explosion mechanism allows fireworks to produce explosion displacements in all dimensions, reducing the risk of the algorithm falling into a local optimal solution. In addition, the explosion step length of fireworks is improved. Based on the taboo search algorithm, each dimension randomly generates displacements of different step lengths to increase the diversity of individuals.

[0063] In one embodiment, the parameters of the model to be trained are optimized according to the improved fireworks algorithm, including: initializing the parameters and the fireworks population, and randomly generating multiple fireworks; selecting elite individuals with better individual fitness values ​​of fireworks according to the elite selection strategy for normalization processing, and performing explosion operations with different displacements on each dimension of the elite individuals, and the explosion step length of the fireworks with different displacements is generated based on the taboo search algorithm; restoring the sparks generated by the explosion to the original parameter range, and performing iterative operations on this until the number of iterations reaches a second preset threshold.

[0064] The following is an introduction based on the formula:

[0065] First, normalize the variables of different dimensions to be in the range of [0, 1]; set a standard step vector h for all individuals i , which represents the uniform standard step length of n individuals to generate the i-th generation, and its component vector expression is as follows:

[0066]

[0067] Then, the mth individual x m The step length y of exploding the i-th generation in feature direction j j It is expressed as:

[0068] y j =x j +rp h h i ;

[0069] p j =(b j -a j ) / 2,j=1,2,...,n;

[0070] Among them, x j For the parent x m The jth dimension of ; r is a random number in the interval [-1, 1]; p j is the component of the coefficient vector p; b j and a j They represent the upper and lower bounds of the j-th feature respectively.

[0071] Then, the step size of all sparks is restored to the original parameter range. If it exceeds the limit, the mapping rule shown below is adopted:

[0072]

[0073] in, represents the displacement of the i-th individual in the k-th dimension that is out of bounds; and They represent the upper and lower boundaries of the k-th dimension respectively.

[0074] In this embodiment, the improved fireworks algorithm is applied to the training process of the concentration prediction model. The displacement of different fireworks explosion step lengths is randomly generated in each dimension based on the taboo search algorithm, which increases the diversity of individual fireworks and makes the optimization of parameters in the model training process more accurate, thereby increasing the prediction accuracy of the model.

[0075] In one embodiment, it also includes: determining multiple evaluation indicators, and when the comparison result reaches the multiple evaluation indicators, determining that the prediction effect of the model to be trained reaches the preset effect; wherein the multiple evaluation indicators include: mean absolute percentage error, root mean square error, mean absolute error, and determination coefficient, and the prediction effect of the concentration prediction model can be characterized by the above four evaluation indicators. Of course, the evaluation indicators are not limited to the four mentioned above, for example, they can also include variance, average variance, etc., and the present invention is not limited to this.

[0076] Figure 5 is a schematic structural diagram of a device provided by an exemplary embodiment. Figure 5 At the hardware level, the device includes a processor 502, an internal bus 504, a network interface 506, a memory 509, and a non-volatile memory 510, and may also include hardware required for other functions. The present invention can be implemented based on software, such as the processor 502 reading the corresponding computer program from the non-volatile memory 510 into the memory 508 and then running it. Of course, in addition to the software implementation, the present invention does not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0077] Please refer to Figure 6 , a device for predicting dissolved gas concentration in transformer oil can be applied to Figure 6 In the device shown, to implement the technical solution of the present invention, the device may include:

[0078] The determination unit 602 is used to determine the gas to be measured in response to the acquired historical time series data set of dissolved gas concentration in transformer oil, and determine at least one model input gas according to the gas to be measured; the historical time series data set of dissolved gas concentration includes historical time series data corresponding to multiple gases;

[0079] A decomposition unit 604 is used to perform multivariate variational modal decomposition on the historical time series data corresponding to the at least one model input gas to obtain a modal component corresponding to each model input gas;

[0080] The prediction unit 606 is used to input the modal components obtained by decomposition into a pre-trained concentration prediction model, and use the output of the concentration prediction model as the predicted concentration of the gas to be measured at the next moment; wherein the concentration prediction model includes an input layer, multiple residual blocks, and an output layer, the input layer is used to extract the time series characteristics of the modal components obtained by decomposition according to the dilated causal convolution, each residual block is used to perform weight normalization, function activation, and random inactivation processing on the output of the previous residual block, and the output layer is used to superimpose and reconstruct multiple modal components corresponding to the gas to be measured to obtain the predicted concentration of the gas to be measured at the next moment.

[0081] Optionally, the determining unit 602 is specifically configured to:

[0082] Performing grey correlation analysis on various gases contained in the dissolved gas according to the historical time series data set;

[0083] A gas among the various gases whose correlation coefficient with the gas to be measured is greater than a first preset threshold is determined as a model input gas.

[0084] Optionally, the determining unit 602 is specifically configured to:

[0085] The variational problem of multivariate variational mode decomposition is set as follows: the sum of the modal components obtained by decomposition is equal to the input multidimensional historical time series data, and the sum of the bandwidths of the modal components obtained by decomposition is minimized;

[0086] The variational problem is solved according to the alternating direction method of multipliers to obtain the modal components corresponding to each model input gas.

[0087] Optionally, the training process of the concentration prediction model includes:

[0088] An acquisition unit 608 is used to acquire a training sample set, wherein each training sample in the training sample set includes a modal component obtained by decomposing a set of time series data of a model input gas, and an actual concentration of a corresponding gas to be measured in the set of model input gases at the next moment;

[0089] An input unit 610 is used to input the training sample set into a to-be-trained model constructed based on a time-domain convolutional network, so that the to-be-trained model outputs the predicted concentration of the corresponding gas to be tested at the next moment;

[0090] The optimization unit 612 is used to compare the actual concentration and the predicted concentration corresponding to each group of training sample sets, and optimize the parameters of the model to be trained according to the comparison results until the prediction effect of the model to be trained reaches a preset effect.

[0091] Furthermore, the optimization unit 612 is specifically used for:

[0092] The parameters of the model to be trained are optimized according to an improved fireworks algorithm; wherein the improved fireworks algorithm is a combination of an elite selection strategy and an improved explosion mechanism, wherein the elite selection strategy selects elite individuals for explosion operations according to the fitness values ​​of fireworks individuals, and the improved explosion mechanism generates displacements with different fireworks explosion step lengths randomly in each dimension based on a taboo search algorithm.

[0093] Furthermore, the optimization unit 612 is specifically used for:

[0094] Initialize the parameters and fireworks population, and randomly generate multiple fireworks;

[0095] According to the elite selection strategy, elite individuals with better firework individual fitness values ​​are selected for normalization processing, and explosion operations with different displacements are performed on each dimension of the elite individuals, and the explosion step length of the fireworks with the displacement is generated based on the taboo search algorithm;

[0096] The sparks generated by the explosion are restored to the original parameter range, and the iterative operation is performed until the number of iterations reaches a second preset threshold.

[0097] Furthermore, it also includes:

[0098] The evaluation unit 614 is used to determine multiple evaluation indicators, and when the comparison result reaches the multiple evaluation indicators, determine that the prediction effect of the model to be trained reaches a preset effect; wherein the multiple evaluation indicators include: mean absolute percentage error, root mean square error, mean absolute error, and determination coefficient.

[0099] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which may be in the form of a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email transceiver, a game console, a tablet computer, a wearable device or a combination of any of these devices.

[0100] In a typical configuration, a computer includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0101] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0102] Computer-readable media include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, disk storage, quantum memory, graphene-based storage media or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0103] With respect to the computer-readable medium (or computer-readable storage medium) as described above or in any other form, computer instructions may be stored thereon, and when the instructions are executed by a processor, one or more of the above-mentioned embodiments are implemented, thereby realizing the technical solution of the present invention.

[0104] The present invention also proposes a computer program, which, when executed by a processor, implements one or more of the above embodiments, thereby realizing the technical solution of the present invention. The computer program may be specifically recorded in the above or any other form of computer-readable medium, and the present invention is not limited thereto.

[0105] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0106] The above describes specific embodiments of the present invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0107] The terms used in the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "the" and "the" used in the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0108] It should be understood that although the terms first, second, third, etc. may be used in the present invention to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting dissolved gas concentration in transformer oil, characterized in that: The method comprises: In response to the acquired historical time series data set of dissolved gas concentration in transformer oil, a gas to be measured is determined, and at least one model input gas is determined according to the gas to be measured; the historical time series data set of dissolved gas concentration includes historical time series data corresponding to a plurality of gases; Performing multivariate variational modal decomposition on the historical time series data corresponding to the at least one model input gas to obtain a modal component corresponding to each model input gas; The modal components obtained by decomposition are input into a pre-trained concentration prediction model, and the output of the concentration prediction model is used as the predicted concentration of the gas to be measured at the next moment; wherein the concentration prediction model includes an input layer, multiple residual blocks, and an output layer, the input layer is used to extract the time series characteristics of the modal components obtained by decomposition according to dilated causal convolution, each residual block is used to perform weight normalization, function activation, and random inactivation processing on the output of the previous residual block, and the output layer is used to superimpose and reconstruct multiple modal components corresponding to the gas to be measured to obtain the predicted concentration of the gas to be measured at the next moment.

2. The method according to claim 1, characterized in that The step of determining at least one model input gas according to the gas to be tested comprises: Performing grey correlation analysis on various gases contained in the dissolved gas according to the historical time series data set; A gas among the various gases whose correlation coefficient with the gas to be measured is greater than a first preset threshold is determined as a model input gas.

3. The method according to claim 1, characterized in that The performing multivariate variational mode decomposition on the historical time series data corresponding to the at least one model input gas comprises: The variational problem of multivariate variational mode decomposition is set as follows: the sum of the modal components obtained by decomposition is equal to the input multidimensional historical time series data, and the sum of the bandwidths of the modal components obtained by decomposition is minimized; The variational problem is solved according to the alternating direction method of multipliers to obtain the modal components corresponding to each model input gas.

4. The method according to claim 1, characterized in that The training process of the concentration prediction model includes: Acquire a training sample set, wherein each training sample in the training sample set includes a modal component obtained by decomposing a set of time series data of a model input gas, and an actual concentration of a corresponding gas to be measured in the set of model input gases at the next moment; Inputting the training sample set into a to-be-trained model constructed based on a time-domain convolutional network, so that the to-be-trained model outputs the predicted concentration of the corresponding gas to be tested at the next moment; The actual concentration and predicted concentration corresponding to each group of training sample sets are compared, and the parameters of the model to be trained are optimized according to the comparison results until the prediction effect of the model to be trained reaches the preset effect.

5. The method according to claim 4, characterized in that Optimizing the parameters of the model to be trained according to the comparison result includes: The parameters of the model to be trained are optimized according to an improved fireworks algorithm; wherein the improved fireworks algorithm is a combination of an elite selection strategy and an improved explosion mechanism, wherein the elite selection strategy selects elite individuals for explosion operations according to the fitness values ​​of fireworks individuals, and the improved explosion mechanism generates displacements with different fireworks explosion step lengths randomly in each dimension based on a taboo search algorithm.

6. The method according to claim 5, characterized in that The step of optimizing the parameters of the model to be trained according to the improved fireworks algorithm includes: Initialize the parameters and fireworks population, and randomly generate multiple fireworks; According to the elite selection strategy, elite individuals with better firework individual fitness values ​​are selected for normalization processing, and explosion operations with different displacements are performed on each dimension of the elite individuals, and the explosion step length of the fireworks with the displacement is generated based on the taboo search algorithm; The sparks generated by the explosion are restored to the original parameter range, and the iterative operation is performed until the number of iterations reaches a second preset threshold.

7. The method according to claim 4, characterized in that Also includes: Determine multiple evaluation indicators, and when the comparison result reaches the multiple evaluation indicators, determine that the prediction effect of the model to be trained reaches a preset effect; wherein the multiple evaluation indicators include: mean absolute percentage error, root mean square error, mean absolute error, and determination coefficient.

8. A device for predicting dissolved gas concentration in transformer oil, characterized in that: The device comprises: A determination unit: in response to the acquired historical time series data set of dissolved gas concentration in transformer oil, determines a gas to be measured, and determines at least one model input gas according to the gas to be measured; the historical time series data set of dissolved gas concentration includes historical time series data corresponding to a plurality of gases; Decomposition unit: performing multivariate variational modal decomposition on the historical time series data corresponding to the at least one model input gas to obtain a modal component corresponding to each model input gas; Prediction unit: input the modal components obtained by decomposition into a pre-trained concentration prediction model, and use the output of the concentration prediction model as the predicted concentration of the gas to be measured at the next moment; wherein the concentration prediction model includes an input layer, multiple residual blocks, and an output layer, the input layer is used to extract the time series characteristics of the modal components obtained by the decomposition according to the dilated causal convolution, each residual block is used to perform weight normalization, function activation, and random inactivation processing on the output of the previous residual block, and the output layer is used to superimpose and reconstruct multiple modal components corresponding to the gas to be measured to obtain the predicted concentration of the gas to be measured at the next moment.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor implements the steps of the method according to any one of claims 1 to 7 by running the executable instructions.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images