A method, device and monitoring system for predicting oil temperature of power transformer

Abstract

The present application provides an oil temperature prediction method, prediction device and monitoring system for a power transformer, which relates to transformer oil temperature monitoring. The method includes: inputting first target data into an oil temperature prediction model through an input channel; controlling a normalization layer to perform instance normalization according to the mean and variance of the first target data to obtain second target data; controlling a centralized aggregation and unified allocation module to process the second target data in sequence through a multi-channel transient feature representation layer, a multi-channel feature aggregation layer and a channel feature fusion layer to obtain fourth target data; controlling a prediction layer to perform prediction according to the fourth target data to obtain oil temperature time series data; controlling a reverse normalization layer to perform reverse normalization according to the oil temperature time series data to obtain oil temperature prediction data. The method solves the problem in the prior art of lacking a fast prediction model that integrates the correlation between time series data and multi-channel data.

Description

Technical Field

[0001] The present invention relates to the field of transformer oil temperature monitoring, and in particular to an oil temperature prediction method, a prediction device, a computer-readable storage medium and a power transformer monitoring system. Background Art

[0002] In the field of power systems, power transformers, as core equipment, play a key role in the transmission and distribution of electric energy. However, during operation, transformers are subject to multiple stresses such as thermal, electrical, and mechanical stresses, which may cause transformer failures and affect the stability and reliability of the power system. In particular, abnormal increases in transformer oil temperature are often a precursor to transformer failure. Therefore, accurate prediction of power transformer oil temperature is of great significance for preventing failures and maintaining safe system operation.

[0003] At present, the research on oil temperature prediction of power transformers mainly focuses on mathematical model methods and data-driven machine learning methods. The mathematical model method predicts oil temperature by establishing a dynamic model of transformer winding hot spot temperature and load based on IEEE and IEC standards. The limitation of this method is that it requires accurate physical parameters and assumptions, which are often difficult to fully meet in practical applications. With the development of machine learning technology, especially the application of deep learning technology in the field of time series prediction, data-based oil temperature prediction methods have gradually shown their advantages in dealing with complex nonlinear problems.

[0004] In the field of deep learning, the Transformer model has achieved remarkable results in multiple tasks with its unique attention mechanism, but its distributed attention allocation method may increase the complexity of the model and affect the computational efficiency. In addition, when processing multi-channel time series data, existing technologies often lack effective mining of interactions between different channels, which limits the model's ability to accurately predict transformer oil temperature changes. How to effectively perceive the long-term periodic trend and short-term nonlinear change characteristics of time series data has become a key issue for various prediction models to predict transformer oil temperature. Summary of the invention

[0005] The main purpose of the present application is to provide a method for predicting the oil temperature of a power transformer, a prediction device, a computer-readable storage medium and a power transformer monitoring system, so as to at least solve the problem in the prior art of lacking a fast prediction model that integrates the correlation of time series data and multi-channel data.

[0006] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for predicting the oil temperature of a power transformer is provided, wherein the oil temperature prediction model includes multiple input channels, a standardization layer, multiple centralized aggregation and unified allocation modules, a prediction layer and an inverse normalization layer, each of the centralized aggregation and unified allocation modules includes a multi-channel transient feature representation layer, a multi-channel feature aggregation layer and a channel feature fusion layer, and each of the centralized aggregation and unified allocation modules is connected in series. The oil temperature prediction method includes: obtaining multiple first target data, inputting each of the first target data into the oil temperature prediction model through the input channel respectively, the multiple first target data include different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within a first preset time period; controlling the standardization layer to receive the first target data input by each of the input channels, performing instance normalization according to the mean and variance corresponding to the first target data, and obtaining a second target data data; control the centralized aggregation and unified allocation module to receive the second target data output by the standardization layer, extract features from the second target data through the multi-channel transient feature representation layer to obtain third target data, integrate the third target data through the multi-channel feature aggregation layer to obtain global common features, and fuse the global common features and each of the third target data through the channel feature fusion layer to obtain fourth target data; control the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, perform prediction based on the fourth target data, and obtain oil temperature time series data, wherein the oil temperature time series data represents the oil temperature change of the power transformer in the second preset time period, and the start time of the second preset time period is the current time; control the inverse standardization layer to receive the oil temperature time series data output by the prediction layer, and perform inverse standardization based on the oil temperature time series data to obtain oil temperature prediction data.

[0007] Optionally, performing instance normalization according to the mean and variance corresponding to the first target data to obtain the second target data includes: calculating the mean according to the first target data to obtain a target mean, and calculating the variance according to the first target data to obtain a target variance: ; ;in, is the first target data input by channel c at time t, M is the total number of time steps of the first target data, is the target mean of channel c, is the target variance of channel c; the first target data is instance-normalized according to the target mean and the target variance to obtain the second target data: ;in, For the second target data, and is the preset coefficient, is a random error.

[0008] Optionally, extracting features from the second target data through the multi-channel transient feature representation layer includes: rewriting data dimensions of each of the second target data based on linear mapping to obtain third target data: ;in, is the third target data, is a mapping function constructed based on the linear mapping, is the historical observation value of all the input channels at time t in the previous L time steps, , , is the observed value of the second target data inputted into the cth channel at time step t, where , the data dimension of the third target data is d, and the data dimension of the second target data is L.

[0009] Optionally, the multi-channel feature aggregation layer includes a TKAN neural network, a random pooling layer and an information aggregation layer, and the third target data is integrated through the multi-channel feature aggregation layer to obtain global common features, including: based on the KANs neural network and the random pooling layer, feature extraction is performed based on the third target data: ; ;in, For inputting the third target data of the kth centralized aggregation and unified allocation module, the is the function expression of the TKAN neural network, A is the output of the TKAN neural network, and after passing through the TKAN neural network structure, the data dimension of the third target data changes from d to , is a random pooling operation, is the global common feature, include: ; ; ; ; ; ; ;in, The input at time t is In the case of R-layer KAN layer output, the TKAN neural network includes R-layer KAN layer, is the input of the TKAN neural network at time t, is the input of the ith node of the lth KAN layer at time t, The activation function matrix of the first KAN layer, is the output of the forget gate at time t, is the hidden state of the i-th node in the l-th KAN layer at time t-1, is the activation function, and is the weight parameter of the forget gate, is the bias parameter of the forget gate, and is the weight parameter of the input gate, is the bias parameter of the input gate, is the output of the forget gate at time t, is the output gate output at time t, is the cell state at time t, is the intermediate variable of the cell state at time t, and are weight parameters updated for the intermediate variables of the cell state, The bias parameter updated for the intermediate variable of the cell state, is the hidden state at time t, Represents matrix compound operations and addition operations, include: ; ;in, for The corresponding probability value is is the feature vector of the jth dimension of the i-th input channel in A, is the global common feature after random pooling of the j-th dimension, .

[0010] Optionally, fusing the global common feature and each of the third target data through the channel feature fusion layer to obtain the fourth target data includes: based on a unified allocation method, copying and splicing the global common feature with the third target data corresponding to each of the input channels to obtain the fifth target data: ; ;in, The third target data input into the kth centralized aggregation and unified allocation module, To input the global common features of the kth centralized aggregation and unified allocation module, For the copy operation, For the splicing operation, The output of the replication operation performed by the k-th centralized aggregation and unified allocation module, The kth centralized aggregation and unified distribution module performs the splicing operation on the output of the fifth target data; and performs a fusion operation with the third target data according to each of the fifth target data: ;in, is the output of the kth centralized aggregation and unified allocation module, and As the input of the k+1th centralized aggregation and unified allocation module , the corresponding data dimension is d, is a function expression of the TKAN neural network, which is used to transform the data dimension from Change to ; Introduce residual connection to correct the data obtained by the fusion operation to obtain the fourth target data: ; Wherein, F is the fourth target data, is the output of the Nth centralized aggregation and unified allocation module, and N is the total number of the centralized aggregation and unified allocation modules.

[0011] Optionally, the prediction layer includes a bidirectional gated recurrent network and a fully connected layer, and prediction is performed according to the fourth target data to obtain oil temperature time series data, including: inputting the fourth target data into the bidirectional gated recurrent network to obtain a target hidden state: ; ; ;in, is the fourth target data at time t, and is the hidden state of the fourth target data in both positive and negative directions, is the target hidden state after the hidden states are merged, and are respectively the gated recurrent units of the hidden states in the positive and negative directions, include: ;in, and are the update gate and reset gate of the gated recurrent unit, respectively. is the candidate value of the hidden state, Output value for the hidden state, and To update the gate weight parameters, To update the bias parameters of the gate, and To reset the gate weight parameters, To reset the gate bias parameters, and The weight parameter for updating the hidden state output value, The bias parameter for updating the hidden state output value, is the activation function, is a sine function, is a dot product operation; each target hidden state is input into the fully connected layer to obtain the oil temperature time series data: ;in, is the function expression of the fully connected layer, is the oil temperature time series data.

[0012] Optionally, performing inverse normalization on the oil temperature time series data to obtain the oil temperature prediction data includes: performing inverse normalization on the oil temperature time series data to obtain the oil temperature prediction data: ;in, is the oil temperature prediction data.

[0013] According to another aspect of the present application, an oil temperature prediction device for a power transformer is provided, wherein the oil temperature prediction model comprises a plurality of input channels, a standardization layer, a centralized aggregation and unified allocation module, a prediction layer and an inverse standardization layer, wherein the centralized aggregation and unified allocation module comprises a multi-channel transient feature representation layer, a multi-channel feature aggregation layer and a channel feature fusion layer, and the oil temperature prediction device comprises: an acquisition unit, for acquiring a plurality of first target data, and inputting each of the first target data into the oil temperature prediction model through the input channel respectively, wherein the plurality of the first target data comprises different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within a first preset time period; a first calculation unit, for controlling the standardization layer to receive the first target data input by each of the input channels, and performing instance standardization according to the mean and variance corresponding to the first target data to obtain second target data; a second calculation unit, for controlling the centralized aggregation and unified allocation module, and the prediction layer and the inverse standardization layer. The allocation module receives the second target data output by the standardization layer, extracts features from the second target data through the multi-channel transient feature representation layer to obtain third target data, integrates the third target data through the multi-channel feature aggregation layer to obtain global common features, and fuses the global common features and each of the third target data through the channel feature fusion layer to obtain fourth target data; a third calculation unit is used to control the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, and perform prediction based on the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data characterizes the oil temperature change of the power transformer in the second preset time period, and the start time of the second preset time period is the current time; a fourth calculation unit is used to control the inverse standardization layer to receive the oil temperature time series data output by the prediction layer, and perform inverse standardization based on the oil temperature time series data to obtain oil temperature prediction data.

[0014] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the method described in any one of the devices where the computer-readable storage medium is located is controlled.

[0015] According to another aspect of the present application, a power transformer monitoring system is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of the methods described.

[0016] By applying the technical solution of the present application, in the above-mentioned method for predicting the oil temperature of a power transformer, first, a plurality of first target data are obtained, and each of the above-mentioned first target data is respectively input into the above-mentioned oil temperature prediction model through the above-mentioned input channel, wherein the plurality of the above-mentioned first target data include different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within a first preset time period; then, the above-mentioned standardization layer is controlled to receive the above-mentioned first target data input by each of the above-mentioned input channels, and the instance is normalized according to the mean and variance corresponding to the above-mentioned first target data to obtain the second target data; thereafter, the above-mentioned centralized aggregation and unified allocation module is controlled to receive the above-mentioned second target data output by the above-mentioned standardization layer, and the above-mentioned second target data is normalized through the above-mentioned multi-channel transient feature representation layer. Feature extraction is performed to obtain the third target data, the third target data is integrated through the multi-channel feature aggregation layer to obtain the global common features, and the global common features and each of the third target data are respectively fused through the channel feature fusion layer to obtain the fourth target data; thereafter, the prediction layer is controlled to receive the fourth target data output by the centralized aggregation and unified allocation module, and prediction is performed according to the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data characterizes the oil temperature change of the power transformer in the second preset time period, and the starting time of the second preset time period is the current time; finally, the inverse normalization layer is controlled to receive the oil temperature time series data output by the prediction layer, and inverse normalization is performed according to the oil temperature time series data to obtain oil temperature prediction data. The present application proposes an oil temperature prediction model based on centralized aggregation and unified allocation, wherein the centralized aggregation and unified allocation module extracts features of different channels in an aggregated manner through channel information interaction, fuses global common features, and uniformly allocates them based on the global common features, fuses them with the features of each channel to form individual features, and uses the individual features as prediction input. This method solves the problem in the prior art of the lack of a fast prediction model that integrates the correlation between time series data and multi-channel data. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A hardware structure block diagram of a mobile terminal according to a method for predicting oil temperature of a power transformer provided in an embodiment of the present application is shown;

[0018] Figure 2A schematic flow chart of a method for predicting oil temperature of a power transformer provided according to an embodiment of the present application is shown;

[0019] Figure 3 A schematic structural diagram of an oil temperature prediction model for a power transformer provided according to an embodiment of the present application is shown;

[0020] Figure 4 A structural block diagram of an oil temperature prediction device for a power transformer provided according to an embodiment of the present application is shown.

[0021] The above drawings include the following reference numerals:

[0022] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0023] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] For the convenience of description, some nouns or terms involved in the embodiments of the present application are explained below:

[0027] KANs neural network: Kolmogorov-Arnold Neural Network (KAN) is a neural network structure built on the Kolmogorov-Arnold representation theorem. The network is designed to capture and learn complex mapping relationships between multidimensional data.

[0028] As introduced in the background technology, the data-driven intelligent algorithm in the prior art mainly adopts BP neural network or improved neural network to predict transformer temperature. The traditional artificial neural network cannot solve the timing problem. In order to solve the problem of lack of a fast prediction model that integrates the correlation of timing data and multi-channel data, the embodiments of the present application provide a method for predicting the oil temperature of a power transformer, a prediction device, a computer-readable storage medium and a power transformer monitoring system.

[0029] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0030] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 FIG. 1 is a hardware structure block diagram of a mobile terminal of a method for predicting oil temperature of a power transformer according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is for illustration only and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0031] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the display method of device information in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0032] In this embodiment, a method for predicting the oil temperature of an electric transformer running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0033] Figure 2 It is a flow chart of a method for predicting oil temperature of a power transformer according to an embodiment of the present application. Figure 3 It is a structural schematic diagram of an oil temperature prediction model for a power transformer according to an embodiment of the present application.

[0034] like Figure 3 As shown, the first target data is firstly input into the multi-channel input part of the model, which includes multiple input channels (that is, the channel C time series is input from different input channels respectively), and the first target data of each input channel is standardized through the standardization part.

[0035] Furthermore, the standardized data is input into a centralized aggregation and uniform distribution module, e.g. Figure 3As shown in FIG. 1 , the model includes a total of N centralized aggregation and unified allocation modules. The above standardized data is input into the first centralized aggregation and unified allocation module, processed by the N centralized aggregation and unified allocation modules in series, and then output to the prediction layer for oil temperature prediction. The above prediction layer includes a bidirectional gated recurrent network (i.e., a bidirectional gated recurrent unit network) and a fully connected layer.

[0036] Finally, the data output by the prediction layer is inverse normalized through the inverse normalization part to obtain the final output transformer oil temperature prediction value, that is, the above-mentioned oil temperature prediction data.

[0037] like Figure 2 As shown, the method comprises the following steps:

[0038] Step S201, obtaining a plurality of first target data, and inputting each of the first target data into the oil temperature prediction model through the input channel, wherein the plurality of first target data include different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within a first preset time period;

[0039] Specifically, the transformer oil temperature prediction system collects the operating parameter data of the power transformer in a preset specific period of time as the first target data, and inputs the collected first target data into the oil temperature prediction model through the input channel to perform oil temperature prediction. Assume that the original input of all channels is , where the input of each channel is , where C is the number of channels and M is the total number of time steps in the time series. A time series with a single time step is represented as .

[0040] In a specific embodiment, the present application sets each channel to input the observation value of L historical time steps, which is expressed as follows: , predict the oil temperature data of the power transformer at the next S moments: .

[0041] It can be understood that the above operating parameter data at least include input voltage, output voltage, operating frequency, winding hot spot temperature, transformer top oil temperature and input and output efficiency, and may also include operating capacity, no-load loss, operating loss, insulation resistance and temperature rise limit.

[0042] In one embodiment, before making predictions based on the model, the above-mentioned input voltage, output voltage, operating frequency, input-output efficiency, and winding hot spot temperature can be configured as data input into different input channels of the model, and the top oil temperature of the transformer can be configured as the final output data of the model. Then, the above-mentioned operating parameter data in the historical data recorded during the operation of the transformer are collected and input into the model for prediction, and the loss function is calculated based on the prediction results and the actual top oil temperature of the transformer, and the model is optimized until the loss function is less than a preset value or the number of optimizations reaches a preset value, and it is determined that the model training is completed and can be used for prediction.

[0043] Step S202, controlling the standardization layer to receive the first target data inputted by each of the input channels, and performing instance standardization according to the mean and variance corresponding to the first target data to obtain second target data;

[0044] Specifically, the control standardization layer receives the first target data input by the input channel and performs instance standardization. First, the standardization layer is responsible for receiving various operating parameters of the power transformer from each input channel, such as time series data of current, voltage, temperature, etc. For the first target data obtained for each input channel, its mean and variance are calculated, and the input data of each channel is instance standardized using the calculated mean and variance. After instance standardization, the original first target data is converted into second target data.

[0045] Step S203, controlling the centralized aggregation and unified distribution module to receive the second target data output by the standardization layer, extracting features from the second target data through the multi-channel transient feature representation layer to obtain third target data, integrating the third target data through the multi-channel feature aggregation layer to obtain global common features, and fusing the global common features and each of the third target data through the channel feature fusion layer to obtain fourth target data;

[0046] Specifically, the centralized aggregation and unified allocation module receives the second target data output by the standardization layer, which has a unified data distribution characteristic, and uses the multi-channel transient feature representation layer to convert the input time series data into a feature representation suitable for the prediction model. After feature extraction, the third target data is obtained. Next, the third target data is integrated through the multi-channel feature aggregation layer to process the time series features of different channels and aggregate the global common features. Then, the global common features are fused with the third target data of each channel through the channel feature fusion layer. The fused result generates the fourth target data. These data contain both global common features and local features of each channel. These features will be used for the next step of oil temperature prediction.

[0047] Step S204, controlling the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, and performing prediction according to the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data represents the oil temperature change of the power transformer in the second preset time period, and the start time of the second preset time period is the current time;

[0048] Specifically, the prediction layer first receives the fourth target data from the centralized aggregation and unified allocation module. These data include global common features and fused local features. The trained model is used to make predictions based on the fourth target data. The model outputs the oil temperature prediction results to form oil temperature time series data, and predicts the oil temperature changes of the power transformer within a preset period of time in the future.

[0049] Step S205, controlling the above-mentioned inverse normalization layer to receive the above-mentioned oil temperature time series data output by the above-mentioned prediction layer, and performing inverse normalization according to the above-mentioned oil temperature time series data to obtain oil temperature prediction data.

[0050] Specifically, the denormalization layer first receives the oil temperature time series data from the prediction layer, and then performs denormalization calculations on the predicted values ​​at each time point to ensure that each point in the entire time series data is converted from the standardized form back to the original form, until the denormalization of the entire data set is completed, and the denormalized oil temperature prediction data is output to obtain the oil temperature prediction data.

[0051] Through this embodiment, first, a plurality of first target data are obtained, and each of the first target data is input into the oil temperature prediction model through the input channel, wherein the plurality of first target data include different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within the first preset time period; then, the standardization layer is controlled to receive the first target data input by each of the input channels, and the instance is normalized according to the mean and variance corresponding to the first target data to obtain the second target data; thereafter, the centralized aggregation and unified allocation module is controlled to receive the second target data output by the standardization layer, and the feature of the second target data is extracted through the multi-channel transient feature representation layer to obtain the third target data. According to the above-mentioned third target data, the above-mentioned multi-channel feature aggregation layer is used to integrate the above-mentioned third target data to obtain global common features, and the above-mentioned global common features and each of the above-mentioned third target data are respectively fused through the above-mentioned channel feature fusion layer to obtain fourth target data; thereafter, the above-mentioned prediction layer is controlled to receive the above-mentioned fourth target data output by the above-mentioned centralized aggregation and unified allocation module, and prediction is performed according to the above-mentioned fourth target data to obtain oil temperature time series data, and the above-mentioned oil temperature time series data characterizes the oil temperature change of the above-mentioned power transformer in the second preset time period, and the starting time of the above-mentioned second preset time period is the current time; finally, the above-mentioned inverse normalization layer is controlled to receive the above-mentioned oil temperature time series data output by the above-mentioned prediction layer, and inverse normalization is performed according to the above-mentioned oil temperature time series data to obtain oil temperature prediction data. The present application proposes an oil temperature prediction model based on centralized aggregation and unified allocation, wherein the centralized aggregation and unified allocation module extracts features of different channels in an aggregated manner through channel information interaction, fuses global common features, and uniformly allocates them based on the global common features, fuses them with the features of each channel to form individual features, and uses the individual features as prediction input. This method solves the problem in the prior art of the lack of a fast prediction model that integrates the correlation between time series data and multi-channel data.

[0052] In order to obtain the second target data, in an optional implementation, the above step S202 includes:

[0053] Step S2021, calculating the mean according to the first target data to obtain the target mean, and calculating the variance according to the first target data to obtain the target variance:

[0054] ;

[0055] ;

[0056] in, is the first target data input by channel c at time t, M is the total number of time steps of the first target data, is the above target mean of channel c, is the above target variance of channel c;

[0057] Step S2022, performing instance standardization on the first target data according to the target mean and the target variance to obtain the second target data:

[0058] ;

[0059] in, is the second target data mentioned above, and is the preset coefficient, is a random error.

[0060] Furthermore, by performing instance standardization on the first target data, non-stationary information in the data can be removed and the problem of data distribution drift can be solved.

[0061] In order to perform feature extraction, in an optional implementation, the above step S203 includes:

[0062] Step S20301, rewrite the data dimensions of each of the second target data based on linear mapping to obtain third target data:

[0063] ;

[0064] in, For the third target data, is a mapping function constructed based on the linear mapping, is the historical observation value of all the above input channels at time t in the previous L time steps, , , is the observed value of the second target data inputted into the cth channel at time step t, where , the data dimension of the third target data is d, and the data dimension of the second target data is L.

[0065] It can be understood that in the present application, the above-mentioned mapping function is used to map the input time series data, that is, the above-mentioned second target data, to the data space, and in the specific mapping process, the data dimension is changed to convert the input data with a data dimension of L into the input data with a data dimension of d required by the model.

[0066] In a specific implementation, one optional method is to intercept the second target data with a data dimension of L at a preset position, and only retain a portion with a length of d to obtain the third target data.

[0067] It should be noted that the above embodiment is only an optional implementation method, and the present application does not limit how to convert the data dimension of the second target data.

[0068] In order to obtain global common features, in an optional implementation, the above step S203 further includes:

[0069] Step S20311, based on the KANs neural network and the random pooling layer, feature extraction is performed based on the third target data:

[0070] ;

[0071] ;

[0072] in, The third target data is input into the kth centralized aggregation and unified allocation module. is the function expression of the TKAN neural network, A is the output of the TKAN neural network, and after passing through the TKAN neural network structure, the data dimension of the third target data changes from d to , is a random pooling operation, For the above global common characteristics, include:

[0073] ;

[0074] ;

[0075] ;

[0076] ;

[0077] ;

[0078] ;

[0079] ;

[0080] in, The input at time t is In the case of R-layer KAN layer output, the TKAN neural network includes R-layer KAN layer, is the input of the above TKAN neural network at time t, is the input of the ith node of the lth KAN layer at time t, The activation function matrix of the first KAN layer, is the output of the forget gate at time t, is the hidden state of the i-th node in the l-th KAN layer at time t-1, is the activation function, and is the weight parameter of the forget gate, is the bias parameter of the forget gate, and is the weight parameter of the input gate, is the bias parameter of the input gate, is the output of the forget gate at time t, is the output gate output at time t, is the cell state at time t, is the intermediate variable of the cell state at time t, and are weight parameters updated for the intermediate variables of the cell state, The bias parameter updated for the intermediate variable of the cell state, is the hidden state at time t, Represents matrix compound operations and addition operations, include:

[0081] ;

[0082] ;

[0083] in, for The corresponding probability value is is the feature vector of the jth dimension of the i-th input channel in A, is the above global common feature after random pooling of the jth dimension, .

[0084] In order to obtain the fourth target data, in an optional implementation, the above step S203 includes:

[0085] Step S20321, based on a unified allocation method, the global common features are respectively copied and concatenated with the third target data corresponding to each of the input channels to obtain fifth target data:

[0086] ;

[0087] ;

[0088] in, The third target data is input into the kth centralized aggregation and unified allocation module. To input the global common features of the kth centralized aggregation and unified allocation module, For the above copy operation, For the above splicing operation, The output of the above-mentioned replication operation performed by the k-th centralized aggregation and unified allocation module, The fifth target data output by the kth centralized aggregation and unified distribution module after the splicing operation;

[0089] Step S20322, performing a fusion operation based on the fifth target data and the third target data:

[0090] ;

[0091] in, is the output of the kth centralized aggregation and unified allocation module, and As the input of the k+1th centralized aggregation and unified allocation module , the corresponding data dimension is d, is the function expression of the TKAN neural network, which is used to transform the above data dimensions from Change to ;

[0092] Step S20323, introduce a residual connection to correct the data obtained by the fusion operation to obtain the fourth target data:

[0093] ;

[0094] Wherein, F is the fourth target data mentioned above, is the output of the Nth centralized aggregation and unified allocation module, and N is the total number of the centralized aggregation and unified allocation modules.

[0095] In order to obtain the oil temperature time series data, in an optional implementation, the above step S204 includes:

[0096] Step S2041, input the fourth target data into the bidirectional gated recurrent network to obtain the target hidden state:

[0097] ;

[0098] ;

[0099] ;

[0100] in, is the fourth target data at time t, and is the hidden state of the fourth target data in both positive and negative directions, is the target hidden state after the above hidden states are merged, and are the gated recurrent units of the above hidden states in the positive and negative directions respectively, include:

[0101] ;

[0102] in, and They are the update gate and reset gate of the above gated recurrent unit, respectively. is the candidate value of the hidden state, Output value for the above hidden state, and To update the gate weight parameters, To update the bias parameters of the gate, and To reset the gate weight parameters, To reset the gate bias parameters, and The weight parameter updated for the above hidden state output value, The bias parameter updated for the above hidden state output value, is the activation function, is a sine function, is the dot product operation;

[0103] Step S2042, input each of the above target hidden states into the above fully connected layer to obtain the above oil temperature time series data:

[0104] ;

[0105] in, is the function expression of the above fully connected layer, is the above oil temperature time series data.

[0106] In order to obtain oil temperature prediction data, in an optional implementation, the above step S205 includes:

[0107] Step S2051, perform inverse normalization on the oil temperature time series data to obtain oil temperature prediction data:

[0108] ;

[0109] in, is the oil temperature prediction data mentioned above.

[0110] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0111] The embodiment of the present application also provides an oil temperature prediction device for a power transformer. It should be noted that the oil temperature prediction device for a power transformer in the embodiment of the present application can be used to execute the oil temperature prediction method for a power transformer provided in the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.

[0112] The following is an introduction to the oil temperature prediction device for the power transformer provided in an embodiment of the present application.

[0113] Figure 4 1 is a structural block diagram of an oil temperature prediction device for a power transformer according to an embodiment of the present application. Figure 4 As shown, the device comprises:

[0114] An acquisition unit 10 is used to acquire a plurality of first target data, and input each of the first target data into the oil temperature prediction model through the input channel, wherein the plurality of first target data include different parameter time series data of the power transformer, and the different parameter time series data represent changes of different operating parameters within a first preset time period;

[0115] Specifically, the transformer oil temperature prediction system collects the operating parameter data of the power transformer in a preset specific period of time as the first target data, and inputs the collected first target data into the oil temperature prediction model through the input channel to perform oil temperature prediction. Assume that the original input of all channels is , where the input of each channel is , where C is the number of channels and M is the total number of time steps in the time series. A time series with a single time step is represented as .

[0116] In a specific embodiment, the present application sets each channel to input the observation value of L historical time steps, which is expressed as follows: , predict the oil temperature data of the power transformer at the next S moments: .

[0117] A first calculation unit 20 is used to control the standardization layer to receive the first target data input by each of the input channels, perform instance standardization according to the mean and variance corresponding to the first target data, and obtain second target data;

[0118] Specifically, the control standardization layer receives the first target data input by the input channel and performs instance standardization. First, the standardization layer is responsible for receiving various operating parameters of the power transformer from each input channel, such as time series data of current, voltage, temperature, etc. For the first target data obtained for each input channel, its mean and variance are calculated, and the input data of each channel is instance standardized using the calculated mean and variance. After instance standardization, the original first target data is converted into second target data.

[0119] The second computing unit 30 is used to control the centralized aggregation and unified allocation module to receive the second target data output by the standardization layer, extract features from the second target data through the multi-channel transient feature representation layer to obtain third target data, integrate the third target data through the multi-channel feature aggregation layer to obtain global common features, and fuse the global common features and each of the third target data through the channel feature fusion layer to obtain fourth target data;

[0120] Specifically, the centralized aggregation and unified allocation module receives the second target data output by the standardization layer, which has a unified data distribution characteristic, and uses the multi-channel transient feature representation layer to convert the input time series data into a feature representation suitable for the prediction model. After feature extraction, the third target data is obtained. Next, the third target data is integrated through the multi-channel feature aggregation layer to process the time series features of different channels and aggregate the global common features. Then, the global common features are fused with the third target data of each channel through the channel feature fusion layer. The fused result generates the fourth target data. These data contain both global common features and local features of each channel. These features will be used for the next step of oil temperature prediction.

[0121] The third calculation unit 40 is used to control the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, and to perform prediction according to the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data represents the oil temperature change of the power transformer in the second preset time period, and the start time of the second preset time period is the current time;

[0122] Specifically, the prediction layer first receives the fourth target data from the centralized aggregation and unified allocation module. These data include global common features and fused local features. The trained model is used to make predictions based on the fourth target data. The model outputs the oil temperature prediction results to form oil temperature time series data, and predicts the oil temperature changes of the power transformer within a preset period of time in the future.

[0123] The fourth calculation unit 50 is used to control the above-mentioned inverse normalization layer to receive the above-mentioned oil temperature time series data output by the above-mentioned prediction layer, and perform inverse normalization on the above-mentioned oil temperature time series data to obtain oil temperature prediction data.

[0124] Specifically, the denormalization layer first receives the oil temperature time series data from the prediction layer, and then performs denormalization calculations on the predicted values ​​at each time point to ensure that each point in the entire time series data is converted from the standardized form back to the original form, until the denormalization of the entire data set is completed, and the denormalized oil temperature prediction data is output to obtain the oil temperature prediction data.

[0125] Through this embodiment, the acquisition unit acquires multiple first target data, and inputs each of the above first target data into the above oil temperature prediction model through the above input channel respectively, and the multiple first target data include different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within the first preset time period; the first calculation unit controls the above standardization layer to receive the above first target data input by each of the above input channels, and performs instance standardization according to the mean and variance corresponding to the above first target data to obtain the second target data; the second calculation unit controls the above centralized aggregation and unified allocation module to receive the above second target data output by the above standardization layer, and performs feature extraction on the above second target data through the above multi-channel transient feature representation layer to obtain the third target data. data, the third target data is integrated through the multi-channel feature aggregation layer to obtain global common features, and the global common features and each of the third target data are respectively fused through the channel feature fusion layer to obtain fourth target data; the third computing unit controls the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, and performs prediction according to the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data represents the oil temperature change of the power transformer in the second preset time period, and the starting time of the second preset time period is the current time; the fourth computing unit controls the inverse normalization layer to receive the oil temperature time series data output by the prediction layer, and performs inverse normalization according to the oil temperature time series data to obtain oil temperature prediction data. The present application proposes an oil temperature prediction model based on centralized aggregation and unified allocation, wherein the centralized aggregation and unified allocation module extracts features of different channels in an aggregated manner through channel information interaction, fuses global common features, and uniformly allocates them based on the global common features, fuses them with the features of each channel to form individual features, and uses the individual features as prediction input. This method solves the problem in the prior art of the lack of a fast prediction model that integrates the correlation between time series data and multi-channel data.

[0126] In order to obtain the second target data, in an optional implementation manner, the first calculation unit includes:

[0127] The first calculation subunit is used to calculate the mean according to the first target data to obtain the target mean, and calculate the variance according to the first target data to obtain the target variance:

[0128] ;

[0129] ;

[0130] in, is the first target data input by channel c at time t, M is the total number of time steps of the first target data, is the above target mean of channel c, is the above target variance of channel c;

[0131] The second calculation subunit is used to perform instance normalization on the first target data according to the target mean and the target variance to obtain the second target data:

[0132] ;

[0133] in, is the second target data mentioned above, and is the preset coefficient, is a random error.

[0134] Furthermore, by performing instance standardization on the first target data, non-stationary information in the data can be removed and the problem of data distribution drift can be solved.

[0135] In order to perform feature extraction, in an optional implementation, the third computing unit includes:

[0136] The third calculation subunit is used to rewrite the data dimension of each of the second target data based on linear mapping to obtain third target data:

[0137] ;

[0138] in, For the third target data, is a mapping function constructed based on the linear mapping, is the historical observation value of all the above input channels at time t in the previous L time steps, , , is the observed value of the second target data inputted into the cth channel at time step t, where , the data dimension of the third target data is d, and the data dimension of the second target data is L.

[0139] In order to obtain the global common features, in an optional implementation, the third calculation unit includes:

[0140] The fourth computing sub-unit is used to perform feature extraction based on the third target data based on the KANs neural network and the random pooling layer:

[0141] ;

[0142] ;

[0143] in, The third target data is input into the kth centralized aggregation and unified allocation module. is the function expression of the TKAN neural network, A is the output of the TKAN neural network, and after passing through the TKAN neural network structure, the data dimension of the third target data changes from d to , is a random pooling operation, For the above global common characteristics, include:

[0144] ;

[0145] ;

[0146] ;

[0147] ;

[0148] ;

[0149] ;

[0150] ;

[0151] in, The input at time t is In the case of R-layer KAN layer output, the TKAN neural network includes R-layer KAN layer, is the input of the above TKAN neural network at time t, is the input of the ith node of the lth KAN layer at time t, The activation function matrix of the first KAN layer, is the output of the forget gate, is the hidden state of the i-th node in the l-th KAN layer at time t-1, is the activation function, and is the weight parameter of the forget gate, is the bias parameter of the forget gate, and is the weight parameter of the input gate, is the bias parameter of the input gate, is the output of the forget gate at time t, is the output gate output at time t, is the cell state at time t, is the intermediate variable of the cell state at time t, and are weight parameters updated for the intermediate variables of the cell state, The bias parameter updated for the intermediate variable of the cell state, is the hidden state at time t, Represents matrix compound operations and addition operations, include:

[0152] ;

[0153] ;

[0154] in, for The corresponding probability value is is the feature vector of the jth dimension of the i-th input channel in A, is the above global common feature after random pooling of the jth dimension, .

[0155] In order to obtain the fourth target data, in an optional implementation manner, the second calculation unit includes:

[0156] The fourth calculation subunit is used to copy and concatenate the global common features with the third target data corresponding to each of the input channels based on a unified allocation method to obtain fifth target data:

[0157] ;

[0158] ;

[0159] in, The third target data is input into the kth centralized aggregation and unified allocation module. To input the global common features of the kth centralized aggregation and unified allocation module, For the above copy operation, For the above splicing operation, The output of the above-mentioned replication operation performed by the k-th centralized aggregation and unified allocation module, The fifth target data output by the kth centralized aggregation and unified distribution module after the splicing operation;

[0160] The fifth calculation subunit is used to perform a fusion operation based on the fifth target data and the third target data:

[0161] ;

[0162] in, is the output of the kth centralized aggregation and unified allocation module, and As the input of the k+1th centralized aggregation and unified allocation module , the corresponding data dimension is d, is the function expression of the TKAN neural network, which is used to transform the above data dimensions from Change to ;

[0163] The sixth computing subunit is used to introduce a residual connection to correct the data obtained by the fusion operation to obtain the fourth target data:

[0164] ;

[0165] Wherein, F is the fourth target data mentioned above, is the output of the Nth centralized aggregation and unified allocation module, and N is the total number of the centralized aggregation and unified allocation modules.

[0166] In order to obtain the oil temperature time series data, in an optional implementation, the third calculation unit includes:

[0167] The seventh computing subunit is used to input the fourth target data into the bidirectional gated recurrent network to obtain a target hidden state:

[0168] ;

[0169] ;

[0170] ;

[0171] in, is the fourth target data at time t, and is the hidden state of the fourth target data in both positive and negative directions, is the target hidden state after the above hidden states are merged, and are the gated recurrent units of the above hidden states in the positive and negative directions respectively, include:

[0172] ;

[0173] in, and They are the update gate and reset gate of the above gated recurrent unit, respectively. is the candidate value of the hidden state, Output value for the above hidden state, and To update the gate weight parameters, To update the bias parameters of the gate, and To reset the gate weight parameters, To reset the gate bias parameters, and The weight parameter updated for the above hidden state output value, The bias parameter updated for the above hidden state output value, is the activation function, is a sine function, is the dot product operation;

[0174] Step S2042, input each of the above target hidden states into the above fully connected layer to obtain the above oil temperature time series data:

[0175] ;

[0176] in, is the function expression of the above fully connected layer, is the above oil temperature time series data.

[0177] In order to obtain oil temperature prediction data, in an optional implementation manner, the fourth calculation unit includes:

[0178] The eighth calculation subunit is used to perform inverse normalization according to the above oil temperature time series data to obtain oil temperature prediction data:

[0179] ;

[0180] in, is the oil temperature prediction data mentioned above.

[0181] The oil temperature prediction device for the power transformer includes a processor and a memory. The acquisition unit, the first calculation unit, the second calculation unit, the third calculation unit, the fourth calculation unit, etc. are all stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions. The modules are all located in the same processor; or, the modules are located in different processors in any combination.

[0182] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the accuracy of oil temperature prediction can be improved by adjusting kernel parameters.

[0183] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0184] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the oil temperature prediction method for the power transformer.

[0185] Specifically, the oil temperature prediction method of the power transformer includes:

[0186] Step S201, obtaining a plurality of first target data, and inputting each of the first target data into the oil temperature prediction model through the input channel, wherein the plurality of first target data include different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within a first preset time period;

[0187] Step S202, controlling the standardization layer to receive the first target data inputted by each of the input channels, and performing instance standardization according to the mean and variance corresponding to the first target data to obtain second target data;

[0188] Step S203, controlling the centralized aggregation and unified distribution module to receive the second target data output by the standardization layer, extracting features from the second target data through the multi-channel transient feature representation layer to obtain third target data, integrating the third target data through the multi-channel feature aggregation layer to obtain global common features, and fusing the global common features and each of the third target data through the channel feature fusion layer to obtain fourth target data;

[0189] Step S204, controlling the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, and performing prediction according to the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data represents the oil temperature change of the power transformer in the second preset time period, and the start time of the second preset time period is the current time;

[0190] Step S205, controlling the above-mentioned inverse normalization layer to receive the above-mentioned oil temperature time series data output by the above-mentioned prediction layer, and performing inverse normalization according to the above-mentioned oil temperature time series data to obtain oil temperature prediction data.

[0191] An embodiment of the present invention provides a processor, and the processor is used to run a program, wherein the oil temperature prediction method of the power transformer is executed when the program is run.

[0192] Specifically, the oil temperature prediction method of the power transformer includes:

[0193] Step S201, obtaining a plurality of first target data, and inputting each of the first target data into the oil temperature prediction model through the input channel, wherein the plurality of first target data include different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within a first preset time period;

[0194] Step S202, controlling the standardization layer to receive the first target data inputted by each of the input channels, and performing instance standardization according to the mean and variance corresponding to the first target data to obtain second target data;

[0195] Step S203, controlling the centralized aggregation and unified distribution module to receive the second target data output by the standardization layer, extracting features from the second target data through the multi-channel transient feature representation layer to obtain third target data, integrating the third target data through the multi-channel feature aggregation layer to obtain global common features, and fusing the global common features and each of the third target data through the channel feature fusion layer to obtain fourth target data;

[0196] Step S204, controlling the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, and performing prediction according to the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data represents the oil temperature change of the power transformer in the second preset time period, and the start time of the second preset time period is the current time;

[0197] Step S205, controlling the above-mentioned inverse normalization layer to receive the above-mentioned oil temperature time series data output by the above-mentioned prediction layer, and performing inverse normalization according to the above-mentioned oil temperature time series data to obtain oil temperature prediction data.

[0198] An embodiment of the present invention provides a power transformer monitoring system, the power transformer monitoring system comprising a processor, a memory, and a program stored in the memory and executable on the processor, and when the processor executes the program, at least the following steps are implemented:

[0199] Step S201, obtaining a plurality of first target data, and inputting each of the first target data into the oil temperature prediction model through the input channel, wherein the plurality of first target data include different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within a first preset time period;

[0200] Step S202, controlling the standardization layer to receive the first target data inputted by each of the input channels, and performing instance standardization according to the mean and variance corresponding to the first target data to obtain second target data;

[0201] Step S203, controlling the centralized aggregation and unified distribution module to receive the second target data output by the standardization layer, extracting features from the second target data through the multi-channel transient feature representation layer to obtain third target data, integrating the third target data through the multi-channel feature aggregation layer to obtain global common features, and fusing the global common features and each of the third target data through the channel feature fusion layer to obtain fourth target data;

[0202] Step S204, controlling the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, and performing prediction according to the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data represents the oil temperature change of the power transformer in the second preset time period, and the start time of the second preset time period is the current time;

[0203] Step S205, controlling the above-mentioned inverse normalization layer to receive the above-mentioned oil temperature time series data output by the above-mentioned prediction layer, and performing inverse normalization according to the above-mentioned oil temperature time series data to obtain oil temperature prediction data.

[0204] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program for initializing at least the following method steps:

[0205] Step S201, obtaining a plurality of first target data, and inputting each of the first target data into the oil temperature prediction model through the input channel, wherein the plurality of first target data include different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within a first preset time period;

[0206] Step S202, controlling the standardization layer to receive the first target data inputted by each of the input channels, and performing instance standardization according to the mean and variance corresponding to the first target data to obtain second target data;

[0207] Step S203, controlling the centralized aggregation and unified distribution module to receive the second target data output by the standardization layer, extracting features from the second target data through the multi-channel transient feature representation layer to obtain third target data, integrating the third target data through the multi-channel feature aggregation layer to obtain global common features, and fusing the global common features and each of the third target data through the channel feature fusion layer to obtain fourth target data;

[0208] Step S204, controlling the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, and performing prediction according to the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data represents the oil temperature change of the power transformer in the second preset time period, and the start time of the second preset time period is the current time;

[0209] Step S205, controlling the above-mentioned inverse normalization layer to receive the above-mentioned oil temperature time series data output by the above-mentioned prediction layer, and performing inverse normalization according to the above-mentioned oil temperature time series data to obtain oil temperature prediction data.

[0210] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0211] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0212] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0213] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0214] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0215] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0216] The memory may include non-permanent memory 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.

[0217] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. 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 disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage 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.

[0218] 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.

[0219] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0220] 1) The oil temperature prediction method of the power transformer of the present application firstly obtains a plurality of first target data, and inputs each of the first target data into the oil temperature prediction model through the input channel, wherein the plurality of the first target data include different parameter time series data of the power transformer, and the different parameter time series data represent the changes of different operating parameters within the first preset time period; then, the standardization layer is controlled to receive the first target data input by each of the input channels, and the instance is normalized according to the mean and variance corresponding to the first target data to obtain the second target data; then, the centralized aggregation and unified distribution module is controlled to receive the second target data output by the standardization layer, and the feature of the second target data is extracted through the multi-channel transient feature representation layer. The third target data is obtained, and the third target data is integrated through the multi-channel feature aggregation layer to obtain global common features, and the global common features and each of the third target data are respectively fused through the channel feature fusion layer to obtain fourth target data; thereafter, the prediction layer is controlled to receive the fourth target data output by the centralized aggregation and unified allocation module, and prediction is performed according to the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data characterizes the oil temperature change of the power transformer in the second preset time period, and the starting time of the second preset time period is the current time; finally, the inverse normalization layer is controlled to receive the oil temperature time series data output by the prediction layer, and inverse normalization is performed according to the oil temperature time series data to obtain oil temperature prediction data. The present application proposes an oil temperature prediction model based on centralized aggregation and unified allocation, wherein the centralized aggregation and unified allocation module extracts features of different channels in an aggregated manner through channel information interaction, fuses global common features, and uniformly allocates them based on the global common features, fuses them with the features of each channel to form individual features, and uses the individual features as prediction input. This method solves the problem in the prior art of the lack of a fast prediction model that integrates the correlation between time series data and multi-channel data.

[0221] 2) The oil temperature prediction device for the power transformer of the present application, the acquisition unit acquires a plurality of first target data, and inputs each of the first target data into the oil temperature prediction model through the input channel, wherein the plurality of first target data include different parameter time series data of the power transformer, and the different parameter time series data characterize the changes of different operating parameters within the first preset time period; the first calculation unit controls the standardization layer to receive the first target data input by each of the input channels, and performs instance standardization according to the mean and variance corresponding to the first target data to obtain the second target data; the second calculation unit controls the centralized aggregation and unified distribution module to receive the second target data output by the standardization layer, and performs feature extraction on the second target data through the multi-channel transient feature representation layer. The third target data is obtained, and the third target data is integrated through the multi-channel feature aggregation layer to obtain global common features, and the global common features and each of the third target data are respectively fused through the channel feature fusion layer to obtain fourth target data; the third calculation unit controls the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, and performs prediction according to the fourth target data to obtain oil temperature time series data, wherein the oil temperature time series data characterizes the oil temperature change of the power transformer in the second preset time period, and the starting time of the second preset time period is the current time; the fourth calculation unit controls the inverse normalization layer to receive the oil temperature time series data output by the prediction layer, and performs inverse normalization according to the oil temperature time series data to obtain oil temperature prediction data. The present application proposes an oil temperature prediction model based on centralized aggregation and unified allocation, wherein the centralized aggregation and unified allocation module extracts features of different channels in an aggregated manner through channel information interaction, fuses global common features, and uniformly allocates them based on the global common features, fuses them with the features of each channel to form individual features, and uses the individual features as prediction input. This method solves the problem in the prior art of the lack of a fast prediction model that integrates the correlation between time series data and multi-channel data.

[0222] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting oil temperature of a power transformer, characterized in that: The oil temperature prediction model includes multiple input channels, a normalization layer, multiple centralized aggregation and unified allocation modules, a prediction layer and an inverse normalization layer. Each of the centralized aggregation and unified allocation modules includes a multi-channel transient feature representation layer, a multi-channel feature aggregation layer and a channel feature fusion layer. Each of the centralized aggregation and unified allocation modules is connected in series. The oil temperature prediction method includes: Acquire a plurality of first target data, and input each of the first target data into the oil temperature prediction model through the input channel, wherein the plurality of first target data include different parameter time series data of the power transformer, and the different parameter time series data represent changes of different operating parameters within a first preset time period; Control the standardization layer to receive the first target data input by each of the input channels, perform instance standardization according to the mean and variance corresponding to the first target data, and obtain second target data; Control the centralized aggregation and unified distribution module to receive the second target data output by the standardization layer, extract features from the second target data through the multi-channel transient feature representation layer to obtain third target data, integrate the third target data through the multi-channel feature aggregation layer to obtain global common features, and fuse the global common features and each of the third target data through the channel feature fusion layer to obtain fourth target data; Control the prediction layer to receive the fourth target data output by the centralized aggregation and unified allocation module, perform prediction according to the fourth target data, and obtain oil temperature time series data, wherein the oil temperature time series data represents the oil temperature change of the power transformer in the second preset time period, and the start time of the second preset time period is the current time; Control the inverse normalization layer to receive the oil temperature time series data output by the prediction layer, and perform inverse normalization according to the oil temperature time series data to obtain oil temperature prediction data; The step of extracting features from the second target data through the multi-channel transient feature representation layer includes: The data dimensions of each of the second target data are rewritten based on linear mapping to obtain the third target data: ； in, is the third target data, is a mapping function constructed based on the linear mapping, is the historical observation value of all the input channels at time t in the previous L time steps, , , is the observed value of the second target data inputted into the cth channel at time step t, where , the data dimension of the third target data is d, and the data dimension of the second target data is L; The multi-channel feature aggregation layer includes a TKAN neural network, a random pooling layer and an information aggregation layer. The third target data is integrated through the multi-channel feature aggregation layer to obtain global common features, including: Based on the KANs neural network and the random pooling layer, feature extraction is performed based on the third target data: ； ； in, The third target data input into the kth centralized aggregation and unified allocation module, is the function expression of the TKAN neural network, A is the output of the TKAN neural network, and after passing through the TKAN neural network structure, the data dimension of the third target data changes from d to , is a random pooling operation, B is the global common feature, include: ; ; ; ; ; ; ; in, The input at time t is In the case of R-layer KAN layer output, the TKAN neural network includes R-layer KAN layer, is the input of the TKAN neural network at time t, is the input of the ith node of the lth KAN layer at time t, The activation function matrix of the first KAN layer, is the output of the forget gate at time t, is the hidden state of the i-th node in the l-th KAN layer at time t-1, is the activation function, and is the weight parameter of the forget gate, is the bias parameter of the forget gate, and is the weight parameter of the input gate, is the bias parameter of the input gate, is the output of the forget gate at time t, is the output gate output at time t, is the cell state at time t, is the intermediate variable of the cell state at time t, and are weight parameters updated for the intermediate variables of the cell state, The bias parameter updated for the intermediate variable of the cell state, is the hidden state at time t, Represents matrix compound operations and addition operations, It means that the elements between vectors are multiplied according to the corresponding positions. include: ; ; in, for The corresponding probability value is is the feature vector of the jth dimension of the i-th input channel in A, is the global common feature after random pooling of the j-th dimension, ; The global common features and each of the third target data are respectively fused through the channel feature fusion layer to obtain the fourth target data, including: Based on a unified allocation method, the global common features are respectively copied and concatenated with the third target data corresponding to each input channel to obtain the fifth target data: ; ; in, The third target data input into the kth centralized aggregation and unified allocation module, To input the global common features of the kth centralized aggregation and unified allocation module, For the copy operation, For the splicing operation, The output of the replication operation performed by the k-th centralized aggregation and unified allocation module, The fifth target data output by the kth centralized aggregation and unified distribution module performing the splicing operation; Perform a fusion operation based on each of the fifth target data and the third target data: ; in, is the output of the kth centralized aggregation and unified allocation module, and As the input of the k+1th centralized aggregation and unified allocation module , the corresponding data dimension is d, is a function expression of the TKAN neural network, which is used to transform the data dimension from Change to ; The residual connection is introduced to correct the data obtained by the fusion operation to obtain the fourth target data: ； Wherein, F is the fourth target data, is the output of the Nth centralized aggregation and unified allocation module, and N is the total number of the centralized aggregation and unified allocation modules.

2. The method according to claim 1, characterized in that Performing instance normalization according to the mean and variance corresponding to the first target data to obtain second target data includes: The mean is calculated according to the first target data to obtain the target mean, and the variance is calculated according to the first target data to obtain the target variance: ; ; in, is the first target data input by channel c at time t, M is the total number of time steps of the first target data, is the target mean of channel c, is the target variance of channel c; The first target data is instance-normalized according to the target mean and the target variance to obtain the second target data: ; in, For the second target data, and is the preset coefficient, is a random error.

3. The method according to claim 1, characterized in that The prediction layer includes a bidirectional gated recurrent network and a fully connected layer, and performs prediction according to the fourth target data to obtain oil temperature time series data, including: The fourth target data is input into the bidirectional gated recurrent network to obtain the target hidden state: ； ； ； in, is the fourth target data at time t, and is the hidden state of the fourth target data in both positive and negative directions, is the target hidden state after the hidden states are merged, and are respectively the gated recurrent units of the hidden states in the positive and negative directions, include: ； in, and are the update gate and reset gate of the gated recurrent unit, respectively. is the candidate value of the hidden state, Output value for hidden state, and To update the gate weight parameters, To update the bias parameters of the gate, and To reset the gate weight parameters, To reset the gate bias parameters, and The weight parameter for updating the hidden state output value, The bias parameter for updating the hidden state output value, is the activation function, is a sine function, is the dot product operation; Input each target hidden state into the fully connected layer to obtain the oil temperature time series data: ； in, is the function expression of the fully connected layer, is the oil temperature time series data.

4. The method according to claim 3, characterized in that: The oil temperature prediction data obtained by inverse normalization of the oil temperature time series data includes: According to the oil temperature time series data, inverse normalization is performed to obtain the oil temperature prediction data: ； in, is the oil temperature prediction data.

5. An oil temperature prediction device for a power transformer, characterized in that: The oil temperature prediction model includes multiple input channels, a standardization layer, a centralized aggregation and unified allocation module, a prediction layer and an inverse normalization layer. The centralized aggregation and unified allocation module includes a multi-channel transient feature representation layer, a multi-channel feature aggregation layer and a channel feature fusion layer. The oil temperature prediction device is used to execute the method described in any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 4.

7. A power transformer monitoring system, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of claims 1 to 4.

Images