Method for predicting parameters of soluble gas in transformer oil and related device

Abstract

The invention discloses a method for predicting parameters of soluble gas in transformer oil and a related device, and the method comprises the steps: collecting the parameters of the soluble gas in the transformer oil through an FPGA (Field Programmable Gate Array), carrying out data time sequence segmentation and framing, obtaining a parameter packet, sending the parameter packet to a processor, and carrying out the frame decoding of the parameter packet through the processor, so as to obtain gas parameters; training a detection model according to the gas parameters, wherein the detection model is a long-short-term memory neural network with a transformation mechanism; the training effectiveness of the detection model is evaluated through a target function, and the target function comprises mean absolute error and mean square error loss; time-dependent information is fully extracted to avoid gradient disappearance in the training process, and an objective function contains a mean absolute error and a mean square error loss item to evaluate errors of different amplitudes; and the influence of mutation and slow change information in each sequence on the target sequence is adaptively captured by adopting a second-order difference technology in related items of the target function, so that the prediction accuracy and reliability can be greatly improved.

Description

Technical Field

[0001] The invention relates to the technical field of power system overhaul and maintenance, and in particular to a method for predicting parameters of dissolved gas in transformer oil and a related device. Background Art

[0002] The oil used in power transformers is transformer oil, which is mainly used for internal insulation and cooling of transformers. During the operation of the transformer, a small amount of gas will be dissolved in the transformer due to aging, electrical and thermal failures, etc. The volume fraction of various components of the gas in the oil and the proportional relationship between different components are closely related to the operating conditions of the transformer. Therefore, the analysis of dissolved gas in oil is one of the important means of monitoring the health status of power transformers. By analyzing the various gas components dissolved in the transformer oil and their concentrations, it is possible to effectively monitor whether there are local overheating, discharge and other fault phenomena inside the transformer. However, the traditional method for predicting dissolved gas parameters in oil mainly involves regular testing of transformer oil, which has problems such as long detection cycle, low accuracy, and difficulty in achieving real-time online monitoring. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides a method for predicting parameters of dissolved gases in transformer oil.

[0004] In a first aspect, the present invention provides a method for predicting parameters of dissolved gases in transformer oil, comprising:

[0005] Collecting the dissolved gas parameters of transformer oil through FPGA, and performing data time series segmentation and framing on the dissolved gas parameters, obtaining a parameter package and sending it to a processor;

[0006] De-framing the parameter packet by a processor to obtain gas parameters;

[0007] A detection model is trained according to the gas parameters, wherein the detection model is a long short-term memory neural network with a transformation mechanism; the training effectiveness of the detection model is evaluated by an objective function, wherein the objective function includes a mean absolute error and a mean square error loss;

[0008] After the detection model training is completed, the detection model is used to detect and predict the dissolved gas parameters of transformer oil.

[0009] In a second aspect, the present invention provides a system for predicting parameters of dissolved gases in transformer oil, comprising:

[0010] A real-time data acquisition module is used to collect dissolved gas parameters of transformer oil through FPGA, and perform data time series segmentation and framing on the dissolved gas parameters to obtain a parameter package and send it to a processor;

[0011] A preprocessing and management module, used for performing frame-decoding processing on the parameter packet through a processor to obtain gas parameters, wherein the gas parameters are time series;

[0012] A network model training module is used to train a detection model using the time series, wherein the detection model is a long short-term memory neural network with a transformation mechanism; the training effectiveness of the detection model is evaluated by an objective function, wherein the objective function includes mean absolute error and mean square error loss, and a second-order difference technique is used in the relevant terms of the objective function to adaptively capture the impact of sudden changes and slowly changing information in each time series on the time series;

[0013] The gas prediction module is used to predict the dissolved gas parameters of transformer oil by using the detection model after the detection model training is completed.

[0014] In a third aspect, the present invention provides an electronic device, the electronic device comprising:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for predicting dissolved gas parameters in transformer oil described in the first aspect of the present invention.

[0018] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a processor to implement the method for predicting dissolved gas parameters in transformer oil described in the first aspect of the present invention when executed.

[0019] The method for predicting dissolved gas parameters in transformer oil provided by the embodiment of the present invention collects dissolved gas parameters of transformer oil through FPGA, and performs data time series segmentation and framing on dissolved gas parameters to obtain parameter packages and send them to a processor, and the processor performs frame-unframing on the parameter packages to obtain gas parameters; the detection model is trained according to the gas parameters, and the detection model is a long short-term memory neural network with a transformation mechanism; the training effectiveness of the detection model is evaluated by an objective function, and the objective function includes mean absolute error and mean square error loss; after the detection model training is completed, the detection model is used to detect and predict the dissolved gas parameters of transformer oil. A long short-term memory network is proposed to fully extract time-dependent information to avoid gradient disappearance during training, and mean absolute error and mean square error loss terms are included in the objective function to evaluate errors of different amplitudes, and the second-order difference technology is used in the relevant terms of the objective function to adaptively capture the influence of mutations and slowly changing information in each sequence on the target sequence, which can significantly improve the accuracy and reliability of the prediction.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 It is a flow chart of a method for predicting dissolved gas parameters in transformer oil provided by an embodiment of the present invention;

[0023] Figure 2 It is a communication framework between FPGA and processor provided by an embodiment of the present invention;

[0024] Figure 3 It is a schematic diagram of a long short-term memory network structure with a transformation mechanism provided by an embodiment of the present invention;

[0025] Figure 4 A comparison curve between a predicted value and an actual value of CO concentration provided by an embodiment of the present invention;

[0026] Figure 5 It is a structural schematic diagram of a system for predicting parameters of dissolved gases in transformer oil provided by an embodiment of the present invention;

[0027] Figure 6It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0029] Embodiment 1

[0030] Figure 1 The flowchart of a method for predicting dissolved gas parameters in transformer oil provided in the first embodiment of the present invention is applicable to the case where dissolved gas parameters in transformer oil are predicted. The method can be executed by a system for predicting dissolved gas parameters in transformer oil. The system for predicting dissolved gas parameters in transformer oil can be implemented in the form of hardware and / or software. The system for predicting dissolved gas parameters in transformer oil can be configured in an electronic device. Figure 1 As shown, the method for predicting parameters of dissolved gas in transformer oil includes:

[0031] S101, collecting dissolved gas parameters of transformer oil through FPGA, and performing data time series segmentation and framing on the dissolved gas parameters to obtain a parameter package and send it to a processor.

[0032] Gases in oil may include ethane C2H6, hydrogen H2, methane CH4, ethylene C2H4, carbon monoxide CO, carbon dioxide CO2, total hydrocarbons, etc.

[0033] FPGA (Field Programmable Gate Array) can be an FPGA in the prior art, and an example can be a V7690T FPGA. A multi-parameter coupled oil-immersed transformer intelligent sensor device based on a micro-electromechanical system is provided in the transformer, and the transformer intelligent sensor device is connected to the FPGA. The FPGA is used for high-speed multi-channel parallel acquisition to obtain the dissolved gas parameters in the oil from the transformer intelligent sensor device, and the data time series segmentation and framing preprocessing are completed in the FPGA. Due to its high parallel processing capability and flexibility, FPGA has unique advantages in real-time signal processing, data acquisition, etc., and is particularly suitable for processing complex data structures in the prediction of dissolved gas in oil. Neural networks can automatically extract features and efficiently classify these features, thereby improving the accuracy and robustness of predictions.

[0034] Data cutting refers to cutting time series data into certain time window sizes. The time window size can be adjusted according to actual needs. Usually, choosing a suitable window size can preserve the time series characteristics of the data. For example, if the time series data is sampled once per second, the window size can be selected as 3 seconds, that is, each window contains 3 data points.

[0035] Framing: Reorganize the cut data into the form of specified channels according to certain rules.

[0036] S102: De-frame the parameter packet through a processor to obtain gas parameters, which are time series.

[0037] Figure 2 It is a communication framework between FPGA and processor, such as Figure 2 As shown, the processor is connected to the FPGA, and the processor is D2000 by way of example. The parameter packet is transmitted to D2000 at high speed for de-framing through the PCIE 3.0 high-speed bus between the FPGA and the processor, thereby achieving real-time and efficient data processing.

[0038] As a high-performance processor platform, D2000 has powerful data processing capabilities and is suitable for building efficient online monitoring systems. The powerful computing power of the D2000 platform can be used to run complex neural network algorithms, conduct in-depth analysis and prediction, and improve prediction accuracy. In addition, FPGA can be used to achieve real-time signal acquisition and preliminary processing, providing reliable basic data support for subsequent data analysis.

[0039] According to the monitoring requirements and data characteristics, the time window size is defined as 1 second in the FPGA, which means that each time window will contain all dissolved gas parameter data collected within 1 second. As the data continues to flow in, the data in the current time window will be updated in real time inside the FPGA. When a new data point arrives, it is added to the end of the current window, and a check is made to see if it exceeds the preset time window length. If it is reached, the current window is marked as completed, and a new time window is prepared to be opened. At the same time, the data within 1s is framed, as shown in Table 1. The frame format is: 0x55aaaa55+packet number+0x77888877+timestamp information+4095*16 bytes of multi-parameter data packet of dissolved gas in oil.

[0040] Table 1. Frame format composition table.

[0041]

[0042] FPGA uses XDMA to handshake the framed data with D2000 through the PCIE 3.0 bus. Specifically, after FPGA completes a frame of data, it pulls up the dma_pkg_ready signal. After D2000 detects dma_pkg_ready, D2000 first pulls up the D2000_ready signal and then pulls it down. After FPGA detects the rising edge of D2000_ready, it starts to transmit data. After D2000 finishes reading a frame of data, it pulls up dma_end_flag. After FPGA detects that dma_end_flag is pulled high, it pulls down dma_pkg_ready to end the transmission of a frame.

[0043] After receiving the data frame, D2000 deframes it and inputs the multi-parameter data sequence of dissolved gas in oil into the detection model (neural network) deployed on D2000 for training.

[0044] S103. A detection model is trained using time series. The detection model is a long short-term memory neural network with a transformation mechanism. The training effectiveness of the detection model is evaluated by an objective function, which includes mean absolute error and mean square error loss. The second-order difference technique is used in the relevant terms of the objective function to adaptively capture the impact of sudden changes and slowly changing information in each time series on the time series.

[0045] The LSTM network has a transformation mechanism that can fully extract time-dependent information to avoid gradient vanishing during training. Errors of different magnitudes are evaluated by adding mean absolute error and mean square error loss terms to the objective function, and the second-order difference technique is used in the relevant terms of the objective function to adaptively capture the impact of mutations and slowly changing information in each sequence on the target sequence.

[0046] Specifically, the long short-term memory neural network includes an input gate, an output gate, a forget gate, a memory unit and a plurality of hidden layers. In the experiment, the number of units of the long short-term memory network with a transformation mechanism is set to 128, which means that there are 128 memory units in each layer, which helps to capture the long-term dependencies in the time series data. The long short-term memory network processes the time series data through the input gate, the output gate, the forget gate and the memory unit state update mechanism. In order to overcome the problem that the traditional activation function easily enters the saturation region, the present invention proposes a transformation mechanism to adjust the output values ​​of the input gate and the forget gate.

[0047] Among them, the output values ​​of the input gate and the forget gate are adjusted by the following formula:

[0048] T it =1-tanh(i t );

[0049] Tft =1-tanh(f t );

[0050] Among them, i t and f t are the original output values ​​of the input gate and the forget gate at time t, T it and T ft is the adjusted output value;

[0051] The adjusted input gate and forget gate output values ​​are used to update the state vector of the memory unit. The formula is as follows;

[0052]

[0053] Among them, c t is the state vector at the current time, c t-1 is the state vector at the previous time, g t is the candidate state vector at the current time, and Represent element-wise multiplication and addition operations, respectively.

[0054] Figure 3 Schematic diagram of the long short-term memory network structure with transformation mechanism. Figure 3 The meaning of the parameters in can be obtained by referring to the above formula.

[0055] In this way, the LSTM network with transformation mechanism can effectively learn both sudden changes and slowly changing information, ensuring that the model can accurately capture various change patterns in the input time series, thereby improving the accuracy and reliability of the prediction.

[0056] Furthermore, the LSTM network further includes a mechanism for adjusting the output value range of the output gate using a hyperbolic tangent function to capture sudden changes in the input time series data stream. The output value of the output gate is adjusted by the following formula:

[0057] o t =σ(W o x t +U o h t-1 +b o );

[0058] Among them, t is the output value of the output gate, σ is the activation function, W o is the input weight matrix, x t is the input vector at the current time, U o is the weight matrix from hidden layer to hidden layer, h t-1 is the output of the hidden layer at the previous time, b ois the bias vector;

[0059] The output of the hidden layer h t The calculation formula is:

[0060]

[0061] Among them, h t is the output of the hidden layer at the current time, Represents element-wise multiplication operation, c t is the state vector at the current time.

[0062] In this way, the LSTM network with transformation mechanism can not only capture the sudden change information in the time series, but also ensure that the model avoids the problem of gradient vanishing during training, so that the model has higher accuracy and robustness when processing multivariate time series data. In addition, by introducing an adjustment mechanism in the output gate, the model's sensitivity to subtle changes in the time series can be further enhanced, ensuring more accurate prediction results.

[0063] The gas parameters are multivariate time series. In order to evaluate the effectiveness of the detection model during training, an objective function is constructed. The objective function formula is as follows:

[0064]

[0065]

[0066] Among them, f t (θ t ) is the objective function, MSE(θ) is the mean square error function, MAE(θ) is the mean absolute error function, Ω(t) is the time correlation function, C is the proportional factor of the fusion of square error and absolute error, λ is the preset smoothing parameter, x k,t 、u k,t ,θ k,t are the predicted value, true value, and model parameter of the kth time series at time t, respectively. K is the total number of time series, and T is the window length of the time series. represents the second-order difference operation, corr(.) represents the correlation operation between two variables, u m,t 、u n,t Represent the true values ​​of the mth and nth time series at time t respectively.

[0067] The smoothing parameter is a trade-off hyperparameter, and λ and C are used to adjust the importance of different loss terms in the objective function. The specific values ​​are as follows:

[0068] When λ=0.2, the test RMSE value reaches the optimal value in the multivariate time series forecasting task.

[0069] When the λ value is greater than 0.2, as λ increases, the test RMSE on different data sets increases to varying degrees.

[0070] This shows that a larger λ value significantly reduces the prediction performance. The hyperparameter C is used to adjust the fusion ratio of the two classic loss functions. According to the experimental results, the optimal C value is 0.1.

[0071] The mean square error (MSE) is the expected value of the square of the difference between the true value and the estimated value. At the same time, the mean absolute error (MAE) can better reflect the actual situation of the prediction error. The essence of the objective function is the loss function. Taking these two errors as the mixed loss of the objective function not only considers the amplitude error between the predicted value and the true value, but also considers the time change information on the time axis.

[0072] In the training process of the detection model, for multivariate time series data of different scales, the calculation formula for the time variation amplitude of each time series is:

[0073]

[0074] When 2≤t≤T-1,

[0075] Among them, λ is the preset smoothing parameter, represents the second-order difference operation, u k,t is the true value of the kth time series at time t, T is the window length of the time series, m+1 represents the order of difference, m=0,1.

[0076] According to most time series analysis studies, the continuous lag range, time dependence, and time-varying information in the data are usually most obvious within a few time intervals. Therefore, we focus on the amplitude of the time-varying information between data in a short period of time (t-1, t, t+1), and choose the absolute value method of the second-order difference to process the time-varying information.

[0077] Removing the absolute value operation from the above time variation formula yields:

[0078]

[0079] In the formula, Diif 2 =Diif 1 *Diif 1The expanded form of is obtained. The number of differential calculations for each column of attributes is N-1, and for multivariate time series data with D dimensions, each column of time series needs to be processed D times. Therefore, the total computational overhead is only O((N-1)D). The differential operation keeps the computational complexity of the learning system within an acceptable range. Therefore, the second-order difference with absolute value operation is more suitable for guiding the model to highlight the impact of this change amplitude on the target sequence. It is worth noting that because the local and global mutation amplitude information of different time series data sets is different. In order to simplify the extraction of time change information, the absolute value of the second-order difference is used to adaptively extract the time change amplitude information of each series, instead of setting a mutation threshold to determine when this information exists.

[0080] In this embodiment, the objective function f is defined t (θ t ), the core of the algorithm is to learn the complex time-dependent information rules within the data by fusing historical and current first-order and second-order moment estimation information, thereby effectively processing the sudden and slow-changing information in time series data.

[0081] The objective function includes all model parameters θ of the LSTM network with the transformation mechanism t , initial vector θ0, initialize m0=0, v0=0, t=0. Then, during the algorithm iteration, when θ t When it does not converge, do the following:

[0082] Update time, t=t+1;

[0083] Calculate the current gradient g t ,

[0084] g t Denotes the objective function f t (θ t ) in the current model parameter θ t The gradient below

[0085] Update the learning rate α t ,

[0086] α is the basic learning rate, α t is the learning rate that decreases with time t;

[0087] Update the first-order moment estimate m t ,

[0088] m t =β1m t-1 +(1-β1)g t , β1 is the exponential decay rate of the first-order moment estimate;

[0089] Update the forgetting factor f f t,

[0090] f f t=f f 0×m t , f f 0 is the initial memory factor;

[0091] Calculate the fused first-order moment estimate

[0092] is the square of the first moment estimate for the previous time period.

[0093] The above operations can adaptively forget the first-order moment estimation in the optimization algorithm.

[0094] Furthermore, when θ t When convergence is not achieved, the following operations are performed:

[0095] Update the second-order moment estimate v t ,

[0096] β2 is the exponential decay rate of the second-order moment estimate, v t-1 is the second-order moment estimate for the previous time;

[0097] Compute the fused second-order moment estimate

[0098] is the second-order moment estimate after fusion at the previous time;

[0099] Update model parameters θ t ,

[0100] ∈ is a constant, θ t-1 is the model parameter of the previous time, α t is the learning rate, ∈ is used to prevent the denominator from being zero.

[0101] Through the above operations, the fusion process of the second-order moment estimation is optimized and the Without introducing new hyperparameters, the ratio of estimated information at the current and historical moments is adaptively stored to improve the model’s ability to obtain time-varying information in the error gradient flow.

[0102] S104: After the detection model training is completed, the detection model is used to detect and predict the dissolved gas parameters of the transformer oil.

[0103] After the detection model training is completed,

[0104] Therefore, the present invention aims to utilize the combined advantages of FPGA and D2000, combined with advanced neural network technology, to propose a new multi-parameter prediction method and system for dissolved gas in oil, so as to achieve high-precision, real-time online monitoring of dissolved gas components in power transformer oil. The system can not only quickly respond to changes in the internal state of the transformer, but also provide more accurate fault warning information by integrating advanced data processing algorithms, thereby improving the safety and reliability of the power system.

[0105] The method for predicting dissolved gas parameters in transformer oil provided by the embodiment of the present invention collects dissolved gas parameters of transformer oil through FPGA, and performs data time series segmentation and framing on dissolved gas parameters to obtain parameter packages and send them to a processor, and the processor performs frame-unframing on the parameter packages to obtain gas parameters; the detection model is trained according to the gas parameters, and the detection model is a long short-term memory neural network with a transformation mechanism; the training effectiveness of the detection model is evaluated by an objective function, and the objective function includes mean absolute error and mean square error loss; after the detection model training is completed, the detection model is used to detect and predict the dissolved gas parameters of transformer oil. A long short-term memory network is proposed to fully extract time-dependent information to avoid gradient disappearance during training, and mean absolute error and mean square error loss terms are included in the objective function to evaluate errors of different amplitudes, and the second-order difference technology is used in the relevant terms of the objective function to adaptively capture the influence of mutations and slowly changing information in each sequence on the target sequence, which can significantly improve the accuracy and reliability of the prediction.

[0106] Figure 4 The figure is a comparison curve between the predicted value and the true value of CO concentration. The predicted value is predicted by the detection model of the present invention, and the true value is actually detected by the transformer oil intelligent sensing device. It can be seen from the comparison curve that the predicted CO concentration is very close to the true value, and the error is very small. It can be seen that the method for predicting dissolved gas parameters in transformer oil of the present invention can greatly improve the prediction accuracy of dissolved gas in transformer oil.

[0107] Corresponding to the method for predicting parameters of dissolved gas in transformer oil of the present invention, a system for predicting parameters of dissolved gas in transformer oil is also provided. Figure 5 The present invention provides a schematic diagram of a system for predicting parameters of dissolved gases in transformer oil. Figure 5 As shown, the prediction system for parameters of dissolved gas in transformer oil includes:

[0108] The real-time data acquisition module 501 is used to collect the dissolved gas parameters of the transformer oil through the FPGA, and perform data time series segmentation and framing on the dissolved gas parameters to obtain a parameter package and send it to the processor;

[0109] A preprocessing and management module 502 is used to perform frame-decoding processing on the parameter packet through a processor to obtain gas parameters, where the gas parameters are time series;

[0110] A network model training module 503 is used to train a detection model using the time series, wherein the detection model is a long short-term memory neural network with a transformation mechanism; the training effectiveness of the detection model is evaluated by an objective function, wherein the objective function includes mean absolute error and mean square error loss, and a second-order difference technique is used in the relevant terms of the objective function to adaptively capture the impact of sudden changes and slowly changing information in each time series on the time series;

[0111] The gas prediction module 504 is used to predict the dissolved gas parameters of the transformer oil by using the detection model after the detection model training is completed.

[0112] The system for predicting parameters of dissolved gases in transformer oil provided in the embodiment of the present invention can execute the method for predicting parameters of dissolved gases in transformer oil provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0113] Figure 6 A schematic diagram of an electronic device 40 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0114] like Figure 6 As shown, the electronic device 40 includes at least one processor 41, and a memory connected to the at least one processor 41, such as a read-only memory (ROM) 42, a random access memory (RAM) 43, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 41 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 42 or the computer program loaded from the storage unit 48 to the random access memory (RAM) 43. In the RAM 43, various programs and data required for the operation of the electronic device 40 can also be stored. The processor 41, the ROM 42, and the RAM 43 are connected to each other through a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0115] A number of components in the electronic device 40 are connected to the I / O interface 45, including: an input unit 46, such as a keyboard, a mouse, etc.; an output unit 47, such as various types of displays, speakers, etc.; a storage unit 48, such as a disk, an optical disk, etc.; and a communication unit 49, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 49 allows the electronic device 40 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0116] The processor 41 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 41 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 41 executes the various methods and processes described above, such as a method for predicting dissolved gas parameters in transformer oil.

[0117] In some embodiments, the method for predicting parameters of dissolved gases in transformer oil may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 48. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 40 via the ROM 42 and / or the communication unit 49. When the computer program is loaded into the RAM 43 and executed by the processor 41, one or more steps of the method for predicting parameters of dissolved gases in transformer oil described above may be performed. Alternatively, in other embodiments, the processor 41 may be configured to execute the method for predicting parameters of dissolved gases in transformer oil in any other appropriate manner (e.g., by means of firmware).

[0118] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0119] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0120] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0121] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0122] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0123] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0124] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0125] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting dissolved gas parameters in transformer oil, characterized in that: include: Collecting the dissolved gas parameters of transformer oil through FPGA, and performing data time series segmentation and framing on the dissolved gas parameters, obtaining a parameter package and sending it to a processor; De-framing the parameter packet through a processor to obtain gas parameters, where the gas parameters are time series; The time series is used to train a detection model, which is a long short-term memory neural network with a transformation mechanism; the training effectiveness of the detection model is evaluated by an objective function, which includes mean absolute error and mean square error loss, and a second-order difference technique is used in the relevant terms of the objective function to adaptively capture the impact of sudden changes and slowly changing information in each time series on the time series; After the detection model training is completed, the detection model is used to detect and predict the dissolved gas parameters of transformer oil.

2. The method according to claim 1, characterized in that The long short-term memory neural network includes an input gate, an output gate, a forget gate, a memory unit and a plurality of hidden layers, wherein the output values ​​of the input gate and the forget gate are adjusted by the following formula: T it =1-tanh(i t ); T ft =1-tanh(f t ); Among them, i t and f t are the original output values ​​of the input gate and the forget gate at time t, T it and T ft is the adjusted output value; The adjusted input gate and forget gate output values ​​are used to update the state vector of the memory unit. The formula is as follows; Among them, c t is the state vector at the current time, c t-1 is the state vector at the previous time, g t is the candidate state vector at the current time, and ⊕ represent element-wise multiplication and addition operations, respectively.

3. The method according to claim 2, characterized in that The output value of the output gate is adjusted by the following formula: the t =σ(W o x t +U o h t-1 +b o ); Among them, t is the output value of the output gate, σ is the activation function, W o is the input weight matrix, x t is the input vector at the current time, U o is the weight matrix from hidden layer to hidden layer, h t-1 is the output of the hidden layer at the previous time, b o is the bias vector; The output of the hidden layer h t The calculation formula is: Among them, h t is the output of the hidden layer at the current time, Represents element-wise multiplication operation, c t is the state vector at the current time.

4. The method according to claim 1, characterized in that The gas parameters are multivariate time series, and the objective function formula is as follows: Among them, f t (θ t ) is the objective function, MSE(θ) is the mean square error function, MAE(λ) is the mean absolute error function, Ω(t) is the time correlation function, C is the proportional factor of the fusion of square error and absolute error, λ is the preset smoothing parameter, x k,t 、u k,t ,θ k,t are the predicted value, true value, and model parameter of the kth time series at time t, respectively. K is the total number of time series, and T is the window length of the time series. represents the second-order difference operation, corr(.) represents the correlation operation between two variables, u m,t 、u n,t Respectively represent the true value of the mth and nth time series at time t.

5. The method according to claim 4, characterized in that During the training process of the detection model, the calculation formula for the time variation amplitude of each time series is: When 2≤t≤T-1, Among them, λ is the preset smoothing parameter, represents the second-order difference operation, u k,t is the true value of the kth time series at time t, T is the window length of the time series, m+1 represents the order of difference, m=0,1.

6. The method according to claim 4, characterized in that During the training process of the detection model, when the model parameter θ of the detection model t When it does not converge, do the following: Update time, t=t+1; Calculate the current gradient g t , g t Denotes the objective function f t (θ t ) in the current model parameters θ t The gradient below Update the learning rate α t , α is the basic learning rate, α t is the learning rate that decreases with time t; Update the first-order moment estimate m t , m t =β1m t-1 +(1-β1)g t , β1 is the exponential decay rate of the first-order moment estimate; Update the forgetting factor f f t, f f t=f f 0×m t , f f 0 is the initial memory factor; Calculate the fused first-order moment estimate is the square of the first moment estimate for the previous time period.

7. The method according to claim 6, characterized in that Also includes: Update the second-order moment estimate v t , β2 is the exponential decay rate of the second-order moment estimate, v t-1 is the second-order moment estimate for the previous time; Compute the fused second-order moment estimate is the second-order moment estimate after fusion at the previous time; Update model parameters θ t , ∈ is a constant, θ t-1 is the model parameter of the previous time, α t is the learning rate.

8. A system for predicting parameters of dissolved gases in transformer oil, characterized in that: include: A real-time data acquisition module is used to collect dissolved gas parameters of transformer oil through FPGA, and to perform data time series segmentation and framing on the dissolved gas parameters to obtain a parameter package and send it to a processor; A preprocessing and management module, used for performing frame-decoding processing on the parameter packet through a processor to obtain gas parameters, wherein the gas parameters are time series; A network model training module is used to train a detection model using the time series, wherein the detection model is a long short-term memory neural network with a transformation mechanism; the training effectiveness of the detection model is evaluated by an objective function, wherein the objective function includes mean absolute error and mean square error loss, and a second-order difference technique is used in the relevant terms of the objective function to adaptively capture the impact of sudden changes and slowly changing information in each time series on the time series; The gas prediction module is used to predict the dissolved gas parameters of transformer oil by using the detection model after the detection model training is completed.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for predicting parameters of dissolved gases in transformer oil according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for predicting parameters of dissolved gases in transformer oil according to any one of claims 1 to 7 when executed.

Images