DC voltage transformer error prediction method and system based on improved Inception model
Through the improved Inception model, combined with polynomial features and densely connected network, the accuracy of DC resistance capacitive voltage transformer error prediction is solved, the operating efficiency and reliability of the power system are improved, and the incidence of system failures is reduced.
Patent Information
- Application Number
- CN202411124990.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-16
AI Technical Summary
The prior art is difficult to accurately predict the error of DC resistance-capacitive voltage transformers, which affects the operating efficiency and reliability of the power system, and the error affects the measurement accuracy significantly as the harmonic content increases.
Using the improved Inception model, polynomial feature construction and data normalization are carried out by collecting influencing factors data, combining densely connected networks and channel attention mechanisms to build a DC voltage transformer error prediction model, enhance feature extraction and multiplexing capabilities, and reduce Information loss and overfitting.
It improves the expression ability and prediction accuracy of the model, reduces information loss and overfitting, enhances the understanding and expression ability of the input data, and achieves more accurate error prediction.
Smart Images

Figure CN118656706B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of online monitoring of electric power metering, and in particular to a method and system for predicting a direct current voltage transformer error based on an improved Inception model. Background Art
[0002] DC RC voltage transformer is a commonly used voltage sensor in power system, and its accuracy is crucial to the safe and stable operation of the system. However, due to the influence of external environmental factors and the characteristics of the device itself, errors may occur in the DC RC voltage transformer during use. Therefore, accurate prediction and timely correction of its errors have important research significance and practical application value. Through in-depth research on the errors of DC RC voltage transformers, it can help improve the operating efficiency and reliability of the power system, reduce the occurrence of system failures, and ensure the safe and stable operation of the power grid; it can provide more accurate data support for system operators, helping them to find problems in time and take effective measures to solve them.
[0003] In the prior art "Modeling and Measurement Characteristics Analysis of DC RC Voltage Transformer for Flexible DC Transmission" (Zheng Xin, Tang Dengping, Lei Ming, et al. Modeling and Measurement Characteristics Analysis of DC RC Voltage Transformer for Flexible DC Transmission [J]. Journal of Wuhan University (Engineering Edition), 2021, 54(08): 747-754. DOI: 10.14188 / j.1671-8844.2021-08-010.), the authors analyzed the influence of resistance and capacitance deviation, ambient temperature and stray capacitance on the measurement error, and further combined with the simulation results to show that when the voltage in flexible DC transmission contains harmonics, there are many factors that will cause the DC RC transformer to have a measurement error, and the greater the harmonic content, the greater the corresponding error, which has a significant impact on the measurement accuracy.
[0004] Based on the prior art, the present invention utilizes influencing factors such as resistance and capacitance deviation, ambient temperature and stray capacitance to further establish a neural network model to predict measurement errors. Summary of the invention
[0005] In order to solve the above problems existing in the prior art, the present invention proposes a DC voltage transformer error prediction method and system based on an improved Inception model.
[0006] The technical solution of the present invention is as follows:
[0007] On the one hand, the present invention proposes a DC voltage transformer error prediction method based on an improved Inception model, and the specific steps include:
[0008] Collecting the measurement error of the DC RC voltage transformer within a preset time range, and collecting the data of the influencing factors affecting the measurement error within the preset time range;
[0009] The data values of each influencing factor data are normalized, and the data type and data volume of the influencing factor data are expanded through the polynomial feature construction method;
[0010] The embedded Inception module is built based on the Inception V1 module in the GoogLeNet neural network model. A densely connected network is used inside the Inception V1 module to directly connect each layer on each link in the module to all previous layers. A channel-wise splicing method is used to build a connection channel between two connection layers, and a channel attention mechanism is used at the splicing of each splicing channel.
[0011] The embedded Inception module is embedded into the 5x5 convolutional layer of the original Inception V1 module to form an improved Inception module.
[0012] A DC voltage transformer error prediction model is constructed based on the improved Inception module and Inception structure, including a feature extraction module formed by connecting three consecutive improved Inception modules, a maximum pooling layer, and a feature complex extraction module formed by connecting two consecutive improved Inception modules, an average pooling layer and a fully connected layer, wherein the feature extraction module, the maximum pooling layer, the feature complex extraction module, the average pooling layer and the fully connected layer are connected in sequence;
[0013] The influencing factor data of the expanded voltage transformer measurement error within a preset time range is used as input, and the measurement error data within the corresponding time range is used as output to train the DC voltage transformer error prediction model, and the trained DC voltage transformer error prediction model is used to perform error prediction.
[0014] As a preferred implementation manner, the influencing factors in the influencing factor data affecting the measurement error within the preset time range include resistance deviation X1, capacitance deviation X2, ambient temperature X3, and stray capacitance data X4.
[0015] As a preferred implementation, the steps of expanding the existing influencing factor types by using the existing influencing factor data through the polynomial feature construction method are specifically as follows:
[0016] Based on the original influencing factors of resistance deviation X1, capacitance deviation X2, ambient temperature X3, and stray capacitance data X4, a new influencing factor X1 is constructed using the quadratic polynomial feature construction method.2 , X1X2, X1X3, X1X4, X2 2 , X2X3, X2X4, X3 2 ,X3X4,X4 2 ;
[0017] The new features formed by the superposition of the same influencing factors are removed to obtain the remaining new influencing factors X1X2, X1X3, X1X4, X2X3, X2X4, and X3X4. The remaining new influencing factors and the original influencing factors are combined to form the expanded influencing factor type.
[0018] As a preferred implementation, the step of normalizing the data of each influencing factor is completed by using z-score standardization, specifically:
[0019]
[0020] In the formula, is the normalized value of each influencing factor, For each influencing factor, is the mean value of each influencing factor, is the standard deviation of each influencing factor.
[0021] On the other hand, the present invention proposes a DC voltage transformer error prediction system based on an improved Inception model, comprising:
[0022] An influencing factor data acquisition module collects the measurement error of the DC RC voltage transformer within a preset time range, and collects the influencing factor data that affects the measurement error within the preset time range;
[0023] The influencing factor data expansion module normalizes the data values of each influencing factor data, and expands the data type and data volume of the influencing factor data through a polynomial feature construction method;
[0024] The embedded Inception building module is built based on the Inception V1 module in the GoogLeNet neural network model. A densely connected network is used inside the Inception V1 module to directly connect each layer on each link in the module to all previous layers. At the same time, the channel-wise splicing method is used to build the connection channel between the two connection layers, and the channel attention mechanism is used at the splicing of each splicing channel.
[0025] Improve the Inception building module, embed the embedded Inception module into the 5x5 convolution layer of the original Inception V1 module to form an improved Inception module;
[0026] The error prediction model construction module constructs a DC voltage transformer error prediction model based on the improved Inception module and Inception structure, including a feature extraction module formed by connecting three consecutive improved Inception modules, a maximum pooling layer, and a feature complex extraction module formed by connecting two consecutive improved Inception modules, an average pooling layer, and a fully connected layer. The feature extraction module, the maximum pooling layer, the feature complex extraction module, the average pooling layer, and the fully connected layer are connected in sequence;
[0027] The model training and prediction module uses the influencing factor data of the expanded voltage transformer measurement error within a preset time range as input, and the measurement error data within the corresponding time range as output to train the DC voltage transformer error prediction model, and uses the trained DC voltage transformer error prediction model to perform error prediction.
[0028] As a preferred implementation, the influencing factors in the influencing factor data that are collected within the preset time range and that affect the measurement error include resistance deviation X1, capacitance deviation X2, ambient temperature X3, and stray capacitance data X4.
[0029] As a preferred implementation, the steps of expanding the existing influencing factor types by using the existing influencing factor data through the polynomial feature construction method are specifically as follows:
[0030] Based on the original influencing factors of resistance deviation X1, capacitance deviation X2, ambient temperature X3, and stray capacitance data X4, a new influencing factor X1 is constructed using the quadratic polynomial feature construction method. 2 , X1X2, X1X3, X1X4, X2 2 , X2X3, X2X4, X3 2 ,X3X4,X4 2 ;
[0031] The new features formed by the superposition of the same influencing factors are removed to obtain the remaining new influencing factors X1X2, X1X3, X1X4, X2X3, X2X4, and X3X4. The remaining new influencing factors and the original influencing factors are combined to form the expanded influencing factor type.
[0032] As a preferred implementation, the step of normalizing the data of each influencing factor is completed by using z-score standardization, specifically:
[0033]
[0034] In the formula, is the normalized value of each influencing factor, For each influencing factor, is the mean value of each influencing factor, is the standard deviation of each influencing factor.
[0035] On the other hand, the present invention proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the DC voltage transformer error prediction method based on the improved Inception model described in any embodiment of the present invention is implemented.
[0036] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the DC voltage transformer error prediction method based on the improved Inception model described in any embodiment of the present invention.
[0037] Compared with the prior art, the present invention has the following beneficial effects:
[0038] 1. The present invention uses a polynomial feature construction method to expand the original feature data and generate high-order features to fit the nonlinear relationship of the data, thereby improving the expression ability and prediction accuracy of the model.
[0039] 2. To address the shortcoming that the 5x5 convolution kernel is too large and causes insufficient extraction of local information, the present invention embeds an embedded Inception module after the 5x5 convolution kernel, and then performs feature extraction at different scales to enhance the expression ability of the network and reduce the risk of information loss.
[0040] 3. The embedded Inception module of the present invention adopts a densely connected network to achieve feature reuse by connecting features on channels, and ensures that the features of each layer can be directly transmitted to all subsequent layers, avoiding the problem of gradual loss of information in the network and improving the reusability and propagation efficiency of features.
[0041] 4. The embedded Inception module of the present invention uses a channel attention mechanism at the splicing point of each splicing channel to dynamically learn the importance weight of each channel, reduce the network's attention to redundant information, and the network can learn more discriminative feature representations, thereby improving the network's ability to understand and express input data and preventing overfitting due to too many features. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 is a flow chart of the steps of the present invention;
[0043] Figure 2This is a schematic diagram of data conversion;
[0044] Figure 3 Schematic diagram of the channel attention mechanism;
[0045] Figure 4 This is the structure diagram of the improved Inception module;
[0046] Figure 5 Schematic diagram of the complete working steps of the prediction model. DETAILED DESCRIPTION
[0047] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0048] The present invention provides the following technical solution: a direct current voltage transformer error evaluation method.
[0049] Embodiment 1:
[0050] See also Figure 1 , a DC voltage transformer error prediction method based on the improved Inception model, the specific steps include:
[0051] Step 1: collecting the measurement error of the DC RC voltage transformer within a preset time range, and collecting the data of the influencing factors that affect the measurement error within the preset time range;
[0052] Step 2: Normalize the data values of each influencing factor data, and expand the data type and data volume of the influencing factor data through the polynomial feature construction method;
[0053] In this embodiment, polynomial feature construction is a commonly used data conversion technology. By combining existing features with polynomials, high-order features are generated to fit the nonlinear relationship of the data, thereby improving the expressiveness and prediction accuracy of the model. However, when there are too many high-order features, it is easy to cause dimensional disaster, increase model complexity and computational cost. Therefore, it is necessary to carefully select the polynomial order to avoid overfitting.
[0054] This embodiment uses a quadratic polynomial feature construction method to expand the original four influencing factor data to generate new features. , the following new features can be generated To avoid the problem of overfitting of the subsequent model, only These six new features are denoted as These 6 new features are combined with the original 4 features for subsequent modeling.
[0055] In this embodiment, it is specified that the influencing characteristic data of the first 30 moments are used to predict the measurement error of the next moment, and if the length of the last segment of data is less than 30, it is discarded. The data is divided into a training set and a test set in a ratio of 7:3.
[0056] In order to make the size of the input data more suitable for the Inception module, this solution converts the original input two-dimensional data into three-dimensional data of size 10*10*3. The specific structure is as follows Figure 2 shown.
[0057] Step 3: Based on the Inception V1 module in the GoogLeNet neural network model, an embedded Inception module is constructed. A densely connected network is used inside the Inception V1 module to directly connect each layer on each link in the module to all previous layers. At the same time, a channel-wise splicing method is used to construct a connection channel between two connection layers, and a channel attention mechanism is used at the splicing of each splicing channel.
[0058] In this embodiment, the channel attention mechanism is as follows Figure 3 As shown in the figure, F1 represents feature compression, which converts each two-dimensional feature map into a real number; F2 represents the excitation operation, which obtains the importance of each channel; F3 represents the application of the weight to each original feature channel.
[0059] The dense connection network DenseNet introduces a dense connection structure, that is, each layer is directly connected to all previous layers, and feature reuse is achieved through the connection of features on the channel to improve efficiency. The use of Densnet dense connection method inside the Inception V1 module ensures that the features of each layer can be directly passed to all subsequent layers, avoiding the problem of gradual loss of information in the network and improving the reusability and propagation efficiency of features.
[0060] The DensNet structure uses channel-wise concatenation, which means that different feature maps are concatenated according to the channel dimension. The channel attention mechanism is used at each channel concatenation in the embedded Inception V1 module to dynamically learn the importance weight of each channel, reduce the network's attention to redundant information, and the network can learn more discriminative feature representations, improve the network's ability to understand and express input data, and prevent overfitting due to too many features.
[0061] Step 4: After embedding the embedded Inception module into the 5x5 convolutional layer of the original Inception V1 module, an improved Inception module is constructed;
[0062] In this embodiment, since the 5×5 convolution kernel is relatively large, it may lead to insufficient capture of local information, especially for the recognition of some detail features or local structures, which may not be as good as using multiple small convolution kernels. Therefore, an improved Inception module is connected after the 5×5 convolution. The improved Inception module can perform feature features of different scales on the output results of the 5×5 convolution again, reduce the risk of information loss, enhance the network's expression ability, and enable the network to better adapt to different types of features; the improved Inception module is as follows: Figure 4 shown.
[0063] Step 5: Construct a DC voltage transformer error prediction model based on the improved Inception module and Inception structure, including a feature extraction module composed of three consecutive improved Inception modules, a maximum pooling layer, and a feature complex extraction module composed of two consecutive improved Inception modules, an average pooling layer and a fully connected layer, wherein the feature extraction module, the maximum pooling layer, the feature complex extraction module, the average pooling layer and the fully connected layer are connected in sequence;
[0064] Step 6: Take the influencing factor data of the voltage transformer measurement error expanded within the preset time range as input, and the measurement error data within the corresponding time range as output to train the DC voltage transformer error prediction model, and use the trained DC voltage transformer error prediction model to perform error prediction.
[0065] In this embodiment, assuming that the input data size of the model is n*10*10*3, the corresponding measurement error value is , the output data of the Inception model is . use and Calculate the loss function and back-propagate to update the Inception model parameters until the model is stable and the model training is completed.
[0066] The complete process steps are as follows Figure 5 As shown, the first step is to process the data: including the collection of influencing factor data, normalization of the data of each influencing factor, data feature expansion, data slicing, division of training set and test set, and conversion of two-dimensional data into three-dimensional data;
[0067] Secondly, the processed data is input into the prediction model: the three-dimensional data is first input into three consecutive improved Inception modules; then the maximum pooling is performed to extract the most significant features in the feature map and reduce the feature dimension; then it is input into two improved Inception modules and an average pooling layer to reduce the feature dimension while retaining the statistical information of the overall features; finally, the fully connected layer is connected to determine the output size of the model so that all features are mapped to one-dimensional data.
[0068] As a preferred implementation of this embodiment, the influencing factors in the influencing factor data affecting the measurement error within the preset time range include resistance deviation X1, capacitance deviation X2, ambient temperature X3, and stray capacitance data X4.
[0069] As a preferred implementation of this embodiment, the steps of expanding the existing influencing factor types by using the existing influencing factor data through the polynomial feature construction method are specifically as follows:
[0070] Based on the original influencing factors of resistance deviation X1, capacitance deviation X2, ambient temperature X3, and stray capacitance data X4, a new influencing factor X1 is constructed using the quadratic polynomial feature construction method. 2 , X1X2, X1X3, X1X4, X2 2 , X2X3, X2X4, X3 2 ,X3X4,X4 2 ;
[0071] The new features formed by the superposition of the same influencing factors are removed to obtain the remaining new influencing factors X1X2, X1X3, X1X4, X2X3, X2X4, and X3X4. The remaining new influencing factors and the original influencing factors are combined to form the expanded influencing factor type.
[0072] As a preferred implementation of this embodiment, the step of normalizing the data of each influencing factor is completed by using z-score standardization, specifically:
[0073]
[0074] In the formula, is the normalized value of each influencing factor, For each influencing factor, is the mean value of each influencing factor, is the standard deviation of each influencing factor.
[0075] Embodiment 2:
[0076] The DC voltage transformer error prediction system based on the improved Inception model includes:
[0077] An influencing factor data acquisition module is used to collect the measurement error of the DC RC voltage transformer within a preset time range, and to collect the influencing factor data that affects the measurement error within the preset time range; this module is used to implement the function of step 1 in embodiment 1, and will not be described in detail here;
[0078] The influencing factor data expansion module normalizes the data values of each influencing factor data, and expands the data type and data volume of the influencing factor data through a polynomial feature construction method; this module is used to implement the function of step 2 in embodiment 1, and will not be described in detail here;
[0079] An embedded Inception building module is constructed based on the Inception V1 module in the GoogLeNet neural network model. A densely connected network connection method is used inside the Inception V1 module to directly connect each layer on each link in the module to all previous layers. At the same time, a channel-wise splicing method is used to construct a connection channel between two connection layers, and a channel attention mechanism is used at the splicing of each splicing channel. This module is used to implement the function of step three in Example 1, which will not be repeated here.
[0080] Improve the Inception construction module, embed the embedded Inception module into the 5x5 convolution layer of the original Inception V1 module to form an improved Inception module; this module is used to implement the function of step 4 in embodiment 1, which will not be repeated here;
[0081] The error prediction model construction module constructs a DC voltage transformer error prediction model based on the improved Inception module and Inception structure, including a feature extraction module composed of three consecutive improved Inception modules, a maximum pooling layer, and a feature complex extraction module composed of two consecutive improved Inception modules, an average pooling layer and a fully connected layer, wherein the feature extraction module, the maximum pooling layer, the feature complex extraction module, the average pooling layer and the fully connected layer are connected in sequence; the module is used to implement the function of step five in embodiment one, which will not be repeated here;
[0082] The model training and prediction module uses the influencing factor data of the voltage transformer measurement error expanded within a preset time range as input, and the measurement error data within the corresponding time range as output to train the DC voltage transformer error prediction model, and uses the trained DC voltage transformer error prediction model to perform error prediction; this module is used to implement the function of step 6 in embodiment 1, which will not be repeated here;
[0083] As a preferred implementation of this embodiment, the influencing factors in the influencing factor data affecting the measurement error within the preset time range include resistance deviation X1, capacitance deviation X2, ambient temperature X3, and stray capacitance data X4.
[0084] As a preferred implementation of this embodiment, the steps of expanding the existing influencing factor types by using the existing influencing factor data through the polynomial feature construction method are specifically as follows:
[0085] Based on the original influencing factors of resistance deviation X1, capacitance deviation X2, ambient temperature X3, and stray capacitance data X4, a new influencing factor X1 is constructed using the quadratic polynomial feature construction method. 2 , X1X2, X1X3, X1X4, X2 2 , X2X3, X2X4, X3 2 ,X3X4,X4 2 ;
[0086] The new features formed by the superposition of the same influencing factors are removed to obtain the remaining new influencing factors X1X2, X1X3, X1X4, X2X3, X2X4, and X3X4. The remaining new influencing factors and the original influencing factors are combined to form the expanded influencing factor type.
[0087] As a preferred implementation of this embodiment, the step of normalizing the data of each influencing factor is completed by using z-score standardization, specifically:
[0088]
[0089] In the formula, is the normalized value of each influencing factor, For each influencing factor, is the mean value of each influencing factor, is the standard deviation of each influencing factor.
[0090] As a preferred implementation of this embodiment, the fully connected Inception V1 module is characterized in that Densnet dense connection is adopted inside the module, that is, all layers on each link in the module are directly connected to each other, and the connection method adopts channel-wise splicing; and a channel attention mechanism is used at the splicing of each splicing channel.
[0091] Embodiment three:
[0092] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the DC voltage transformer error prediction method based on the improved Inception model described in any embodiment of the present invention is implemented.
[0093] Embodiment 4:
[0094] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the DC voltage transformer error prediction method based on an improved Inception model described in any embodiment of the present invention.
[0095] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A DC voltage transformer error prediction method based on an improved Inception model is characterized by: The specific steps include: Collecting the measurement error of the DC RC voltage transformer within a preset time range, and collecting the influencing factor data affecting the measurement error within the preset time range; wherein the influencing factors in the influencing factor data affecting the measurement error within the preset time range include resistance deviation X1, capacitance deviation X2, ambient temperature X3 and stray capacitance data X4; The data values of each influencing factor data are normalized, and the data type and data volume of the influencing factor data are expanded through the polynomial feature construction method; The embedded Inception module is built based on the Inception V1 module in the GoogLeNet neural network model. A densely connected network is used inside the Inception V1 module to directly connect each layer on each link in the module to all previous layers. A channel-wise splicing method is used to construct the splicing channel between two connection layers, and a channel attention mechanism is used at the splicing point of each splicing channel. The embedded Inception module is embedded into the 5x5 convolutional layer of the original Inception V1 module to form an improved Inception module. A DC voltage transformer error prediction model is constructed based on the improved Inception module and Inception structure, including a feature extraction module and a maximum pooling layer formed by connecting three consecutive improved Inception modules, and a feature complex extraction module, an average pooling layer and a fully connected layer formed by connecting two consecutive improved Inception modules. The feature extraction module, the maximum pooling layer, the feature complex extraction module, the average pooling layer and the fully connected layer are connected in sequence. The influencing factor data of the expanded voltage transformer measurement error within a preset time range is used as input, and the measurement error data within the corresponding time range is used as output to train the DC voltage transformer error prediction model, and the trained DC voltage transformer error prediction model is used to perform error prediction.
2. The DC voltage transformer error prediction method based on the improved Inception model according to claim 1 is characterized in that: The step of expanding the data type of the influencing factor data by the polynomial feature construction method is specifically as follows: According to the original influencing factors of resistance deviation X1, capacitance deviation X2, ambient temperature X3 and stray capacitance data X4, a new influencing factor X1 is constructed using the quadratic polynomial feature construction method. 2 , X1X2, X1X3, X1X4, X2 2 , X2X3, X2X4, X3 2 ,X3X4,X4 2 ; The new features formed by the superposition of the same influencing factors are removed to obtain the remaining new influencing factors X1X2, X1X3, X1X4, X2X3, X2X4, and X3X4. The remaining new influencing factors and the original influencing factors are combined to form the expanded influencing factor type.
3. The DC voltage transformer error prediction method based on the improved Inception model according to claim 1 is characterized in that: The step of normalizing the data values of each influencing factor data is completed by using z-score standardization, specifically: In the formula, is the normalized value of each influencing factor data, For the data of each influencing factor, is the mean of the data of each influencing factor, is the standard deviation of the data of each influencing factor.
4. A DC voltage transformer error prediction system based on an improved Inception model is characterized by: include: An influencing factor data acquisition module collects the measurement error of the DC RC voltage transformer within a preset time range, and collects the influencing factor data that affects the measurement error within the preset time range; wherein the influencing factors in the influencing factor data that affects the measurement error within the preset time range include resistance deviation X1, capacitance deviation X2, ambient temperature X3 and stray capacitance data X4; The influencing factor data expansion module normalizes the data values of each influencing factor data, and expands the data type and data volume of the influencing factor data through a polynomial feature construction method; The embedded Inception building module is built based on the Inception V1 module in the GoogLeNet neural network model. A densely connected network is used inside the Inception V1 module to directly connect each layer on each link in the module to all previous layers. At the same time, the channel-wise splicing method is used to construct the splicing channel between two connection layers, and the channel attention mechanism is used at the splicing of each splicing channel. Improve the Inception building module, embed the embedded Inception module into the 5x5 convolution layer of the original Inception V1 module to form an improved Inception module; The error prediction model construction module constructs a DC voltage transformer error prediction model based on the improved Inception module and the Inception structure, including a feature extraction module and a maximum pooling layer formed by connecting three consecutive improved Inception modules, and a feature complex extraction module, an average pooling layer, and a fully connected layer formed by connecting two consecutive improved Inception modules. The feature extraction module, the maximum pooling layer, the feature complex extraction module, the average pooling layer, and the fully connected layer are connected in sequence; The model training and prediction module uses the influencing factor data of the expanded voltage transformer measurement error within a preset time range as input, and the measurement error data within the corresponding time range as output to train the DC voltage transformer error prediction model, and uses the trained DC voltage transformer error prediction model to perform error prediction.
5. The DC voltage transformer error prediction system based on the improved Inception model according to claim 4 is characterized in that: The step of expanding the data type of the influencing factor data by the polynomial feature construction method is specifically as follows: According to the original influencing factors of resistance deviation X1, capacitance deviation X2, ambient temperature X3 and stray capacitance data X4, a new influencing factor X1 is constructed using the quadratic polynomial feature construction method. 2 , X1X2, X1X3, X1X4, X2 2 , X2X3, X2X4, X3 2 ,X3X4,X4 2 ; The new features formed by the superposition of the same influencing factors are removed to obtain the remaining new influencing factors X1X2, X1X3, X1X4, X2X3, X2X4, and X3X4. The remaining new influencing factors and the original influencing factors are combined to form the expanded influencing factor type.
6. The DC voltage transformer error prediction system based on the improved Inception model according to claim 4 is characterized in that: The step of normalizing the data values of each influencing factor data is completed by using z-score standardization, specifically: In the formula, is the normalized value of each influencing factor data, For the data of each influencing factor, is the mean of the data of each influencing factor, is the standard deviation of the data of each influencing factor.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the DC voltage transformer error prediction method based on the improved Inception model as described in any one of claims 1 to 3 is implemented.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the DC voltage transformer error prediction method based on the improved Inception model as described in any one of claims 1 to 3 is implemented.
