Electric power data automatic classification and grading method and system based on multilayer hybrid model

Abstract

The invention relates to the technical field of electric power informatization, and discloses an electric power data automatic classification and grading method and system based on a multilayer hybrid model, and the method comprises the steps: carrying out the word vector training and representation of electric power data entity phrases through a SkipGram model; inputting the trained word vectors into a multilayer convolutional neural network for feature learning, and extracting semantic information of power text data; extracting context information from the character vector sequence through a BiGRU layer; and performing text depth feature learning by using a CapsNet network to realize automatic classification and grading of the power data. According to the method, the power text data can be well processed and graded, feature representation and grading prediction of the context of the power text data are carried out, an automatic power data grading function is achieved, and the problem of low efficiency of manual grading and classification is solved.

Description

Technical Field

[0001] The present invention relates to the field of electric power information technology, and in particular to a method and system for automatic classification and grading of electric power data based on a multi-layer hybrid model. Background Art

[0002] The classification and grading objects of data can be divided into coarse-grained and fine-grained. The coarse-grained method generally takes the entire database table as the object of classification and grading, while the fine-grained method takes the content of each row and column in the table as the object of classification and grading. Its essence is to classify text according to the sensitive features of the data. Since there are few technical routes for automatic classification and grading of data. In text classification, there are many factors that can affect the accuracy of its results, including but not limited to the selection of models, the expression of word vectors, and the extraction of features. With the development of deep learning, machine learning has gradually been replaced by it. Compared with the methods and technologies based on machine learning, the methods based on deep learning better avoid the design of rules and realize the automatic learning of the potential laws existing in the samples.

[0003] Domestic research on text data classification and grading started later than that of foreign countries. Due to the large differences between Chinese and English in terms of word data division and sentence grammar, Chinese scholars have conducted research on word segmentation, feature extraction and text classification based on the characteristics of Chinese. In terms of word segmentation, since there is no clear space symbol as a delimiter in English vocabulary, rule-based matching and statistical methods are generally used for word segmentation. In terms of feature selection, recursive feature elimination, feature filtering or feature selection embedding are generally used for selection. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an automatic classification and grading method for power data based on a multi-layer hybrid model, which solves the key problems in data processing in the power industry. It improves the accuracy of semantic information extraction, realizes the effective processing of professional terms, and improves the automation and efficiency of data classification and grading, thereby simplifying the operation process, reducing manual intervention, and providing technical support for the intelligent management of power systems.

[0006] To solve the above technical problems, the present invention provides the following technical solutions, a method for automatic classification and grading of power data based on a multi-layer hybrid model, comprising: using the Skip_Gram model to train and represent word vectors for power data entity phrases; inputting the trained word vectors into a multi-layer convolutional neural network for feature learning to extract semantic information of power text data; extracting context information from the character vector sequence through a BiGRU layer; and using a CapsNet network to perform deep feature learning of text to achieve automatic classification and grading of power data.

[0007] As a preferred solution of the method for automatic classification and grading of power data based on a multi-layer hybrid model described in the present invention, the Skip_Gram model includes predicting context vocabulary through professional power vocabulary entities, selecting the Skip-Gram model to train power professional vocabulary, and training power text data, and representing vocabulary appearing in fixed combinations as single vocabulary.

[0008] As a preferred solution of the method for automatic classification and grading of power data based on a multi-layer hybrid model described in the present invention, wherein: the multi-layer convolutional neural network includes, through an input layer, a convolutional layer, a pooling layer and a fully connected layer, effectively extracting semantic information of power text data;

[0009] Input layer, receiving word vectors trained by Skip_Gram model;

[0010] Convolutional layer, uses ReLU function as activation function and sets filters of different sizes for feature extraction;

[0011] The pooling layer uses the maximum pooling method to extract the local optimal solution from the feature map output by the convolutional layer;

[0012] The fully connected layer concatenates the vectors obtained by the pooling layer into a feature matrix to provide input for the subsequent BiGRU layer.

[0013] As a preferred solution of the method for automatic classification and grading of power data based on a multi-layer hybrid model described in the present invention, the input layer includes: taking the trained word vector as input, as x i ∈R m×d , where m is the number of words and d is the vector dimension;

[0014] The convolution layer uses the ReLU function as the activation function, and the deep features are extracted by filters with sizes of 3, 4, and 5 respectively:

[0015] c i =f(Wx i：i+h-1 +b)

[0016] Among them, f is the activation function, W is the weight information, x i：i+h-1 is the local feature matrix of the text in the convolution window, i.e., the sentence vector; h is the convolution kernel size, and b is the bias value;

[0017] The pooling layer performs maximum pooling on the feature map output by the convolutional layer, extracts the maximum value in each local area as the representative feature of the current area, forms the output feature map of the pooling layer, and the feature map finally extracted by max pooling is expressed as:

[0018] O=max(c i )

[0019] The fully connected layer, since the BiGRU layer takes the serialized structure as input, the vectors obtained by the pooling layer are concatenated into a feature matrix H through the fully connected layer:

[0020] H=[O1,O2,...O n ]

[0021] Among them, n represents the number of convolution results, which are finally passed to the BiGRU layer to extract contextual feature relationships.

[0022] As a preferred solution of the method for automatic classification and grading of power data based on a multi-layer hybrid model described in the present invention, the BiGRU layer includes extracting context information from a character vector sequence using BiGRU, and outputting a sequence determined by the common state of GRUs in two directions by splicing forward GRU units and reverse GRU units;

[0023] The calculation formula of the GRU model is expressed as:

[0024] z t =σ(W z ·[c t-1 , x t ])

[0025] r t =σ(W r ·[c t-1 , x t ])

[0026] h t =tanh(W·[r t ·c t-1 , x t ])

[0027] c t =(1-z t )·c t-1 +z t ·h t

[0028] Where σ is the sigmoid function, W is the weight matrix, and x t is the text data input at time t, c t-1 is the text data input at the previous moment, r t is the reset gate, z t is the update gate, h t is the state to be activated, c t is the current output moment, W z represents the weight matrix of the update gate, W r represents the weight matrix of the reset gate;

[0029] The output sequence of the BiGRU layer is input into the upper CapsNet capsule network for data classification inference.

[0030] As a preferred solution of the method for automatic classification and grading of power data based on a multi-layer hybrid model described in the present invention, wherein: the CapsNet network includes predicting entity classification using the CapsNet capsule network structure, the capsule represents the entity classification label, the modulus length of the capsule vector represents the predicted probability of the entity label, and the direction of the capsule vector represents the attribute of the entity, so as to realize power data entity recognition and classification;

[0031] Among them, the capsule network includes a lower-level capsule layer to transmit local features; a dynamic routing layer to realize information transmission between the lower-level capsule layer and the upper-level capsule layer; and an upper-level capsule layer to learn the overall features and output the final classification and grading results.

[0032] As a preferred solution of the method for automatic classification and grading of power data based on a multi-layer hybrid model described in the present invention, the dynamic routing layer includes calculating the prediction vector from the i-th lower-level capsule to the j-th upper-level capsule, determining the weight matrix; calculating the main capsule output vector; updating the connection probability of the i-th capsule connecting to the j-th capsule; calculating the capsule input and the capsule output; repeating the process until the algorithm converges, which is specifically expressed as follows:

[0033] u j|i =W ij ·u i

[0034]

[0035]

[0036] b ij =b ij +u j|i ·v j

[0037] Among them, u j|i represents the prediction vector from the i-th lower-level capsule to the j-th upper-level capsule, W ij represents the current weight matrix, u i represents the main capsule output vector, c ij represents the connection probability of the i-th capsule connecting to the j-th capsule, s j represents capsule input, v j represents the capsule output, b ij Represents the weight of the capsule.

[0038] Another object of the present invention is to provide an automatic classification and grading system for power data based on a multi-layer hybrid model, which can automatically classify and grade power data, improve data processing efficiency and accuracy through data preprocessing, word vector training, feature extraction and CapsNet network learning, and provide intelligent support for power system management.

[0039] As a preferred solution of the automatic classification and grading system of power data based on the multi-layer hybrid model described in the present invention, it includes: a data preprocessing module, a word vector training module, a feature extraction network module, a CapsNet classification and grading module, and a result output and evaluation module;

[0040] The data preprocessing module is responsible for collecting raw power data, performing data cleaning and formatting, and converting power data entity phrases into a form suitable for Skip_Gram model training;

[0041] The word vector training module uses the Skip_Gram model to perform word vector training on power data entity phrases, generates vectors that can represent the semantics of the words, and uses words that appear in a fixed combination as single word representations;

[0042] The feature extraction network module includes a multi-layer convolutional neural network (CNN) and a BiGRU layer. The multi-layer CNN extracts semantic information from the data trained with word vectors through an input layer, a convolution layer, a pooling layer, and a fully connected layer. The BiGRU layer further extracts context information from the character vector sequence to provide deep features for the CapsNet network.

[0043] The CapsNet classification and grading module uses the CapsNet network structure to perform deep feature learning on power data to achieve automatic classification and grading of data;

[0044] The result output and evaluation module receives the output results of the CapsNet classification and grading module, parses and displays the results, and provides an evaluation mechanism to verify the accuracy and reliability of the classification and grading results. At the same time, it feeds back the results to other systems or performs subsequent data analysis and processing as needed.

[0045] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of a method for automatic classification and grading of power data based on a multi-layer hybrid model are implemented.

[0046] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of a method for automatic classification and grading of power data based on a multi-layer hybrid model are implemented.

[0047] The beneficial effects of the present invention are as follows: in order to determine the classification and grading content of electric power text data, a data grading method based on the kip_Gram-CNN-BiGRU-CapsNet model is designed, which can process and grade the electric power text data well, perform feature representation of the context of the electric power text data and predict the grading, and has the function of automatic grading of electric power data, thus eliminating the inefficiency of manual grading and classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0049] Figure 1 An overall input and output flow chart of automatic classification and grading of power data of a method for automatic classification and grading of power data based on a multi-layer hybrid model provided in one embodiment of the present invention.

[0050] Figure 2 A Skip_Gram-CNN-BiGRU-CapsNet model framework diagram of a method for automatic classification and grading of power data based on a multi-layer hybrid model provided in one embodiment of the present invention.

[0051] Figure 3 A Skip-Gram pattern diagram of a method for automatic classification and grading of power data based on a multi-layer hybrid model provided in one embodiment of the present invention.

[0052] Figure 4 A convolutional neural network (CNN) layer structure of a method for automatic classification and grading of power data based on a multi-layer hybrid model is provided in accordance with an embodiment of the present invention.

[0053] Figure 5 A GRU structure diagram of a method for automatic classification and grading of power data based on a multi-layer hybrid model provided in one embodiment of the present invention.

[0054] Figure 6 A CapsNet model structure diagram of a method for automatic classification and grading of power data based on a multi-layer hybrid model provided in one embodiment of the present invention.

[0055] Figure 7 A schematic diagram of neuron operations in a capsule network of a method for automatic classification and grading of power data based on a multi-layer hybrid model provided by an embodiment of the present invention.

[0056] Figure 8A schematic diagram of the relationship between dynamic routing capsules of a method for automatic classification and grading of power data based on a multi-layer hybrid model provided by an embodiment of the present invention.

[0057] Fig. 9 A schematic diagram of working modules of a system for automatic classification and grading of power data based on a multi-layer hybrid model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0059] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0060] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0061] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0062] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0063] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0064] Example 1, reference Figure 1 - Figure 8 , which is the first embodiment of the present invention, provides a method for automatic classification and grading of power data based on a multi-layer hybrid model, comprising:

[0065] S1: Use the Skip_Gram model to train and represent word vectors for power data entity phrases.

[0066] Furthermore, the context words are predicted through professional power vocabulary entities, the Skip-Gram model is selected to train the power professional vocabulary, and the power text data is trained to represent the words that appear in fixed combinations as single words.

[0067] It should be noted that the task of this layer is mainly to predict context words through professional power vocabulary entities, and the Skip-Gram model is selected, and its mathematical expression is:

[0068]

[0069] Among them, w is a single word, and Context(w) is the context word of the word. First, the Word2Vec model is used to train the power text data, and the words that appear in fixed combinations are represented as single words, such as "transformer capacity" and "substation location". The words that frequently appear in the power profession are sampled, and all target data are sampled. Finally, the power vocabulary entity output by this layer is used as the input of the convolutional neural network.

[0070] S2: Input the trained word vector into the multi-layer convolutional neural network for feature learning to extract the semantic information of power text data.

[0071] Furthermore, a multi-layer convolutional neural network is used to effectively extract the semantic information of power text data through the input layer, convolution layer, pooling layer and full connection layer. The main structure is as follows Figure 4 As shown;

[0072] The input layer receives the word vector trained by the Skip_Gram model and takes the trained word vector as input, which is x i ∈R m×d , where m is the number of words and d is the vector dimension;

[0073] The convolution layer uses the ReLU function as the activation function, and the deep features are extracted by filters of size 3, 4, and 5 respectively:

[0074] c i =f(Wx i：i+h-1 +b)

[0075] Among them, f is the activation function, W is the weight information, x i：i+h-1 is the local feature matrix of the text in the convolution window, i.e., the sentence vector; h is the convolution kernel size, b is the bias value, and the output calculation result of the entire convolution layer is:

[0076] c=[c1,c2,…c n-h+1 ]

[0077] Where n is the number of convolution results.

[0078] The pooling layer performs maximum pooling on the feature map output by the convolutional layer, extracts the maximum value in each local area as the representative feature of the current area, forms the output feature map of the pooling layer, and expresses the feature map finally extracted by max pooling as:

[0079] O=max(c i )

[0080] Fully connected layer, since the BiGRU layer takes the serialized structure as input, the vectors obtained by the pooling layer are concatenated into a feature matrix H through the fully connected layer:

[0081] H=[O1,O2,...O n ]

[0082] Among them, n represents the number of convolution results, which are finally passed to the BiGRU layer to extract contextual feature relationships.

[0083] S3: Extract contextual information from the character vector sequence through the BiGRU layer.

[0084] Furthermore, BiGRU is used to extract context information from the character vector sequence. GRU is a unidirectional neural network structure consisting of an update gate and a reset gate. It can save context information to solve the problem of RNN gradient disappearance or explosion. In order to solve the problem of relationship extraction between the current output and the previous and next moments in semantic understanding, BiGRU splices the forward GRU unit and the reverse GRU unit, and the output is determined by the common state of the GRU in the two directions. The specific structure is as follows Figure 5 As shown;

[0085] Use BiGRU to extract context information from the character vector sequence, and output a sequence determined by the common state of the GRUs in both directions by splicing the forward GRU unit and the reverse GRU unit;

[0086] The calculation formula of the GRU model is expressed as:

[0087] z t =v(W z ·[c t-1 , x t ])

[0088] r t =σ(W r ·[c t-1 , x t ])

[0089] h t =tanh(W·[r t ·c t-1 , x t ])

[0090] c t =(1-z t )·c t-1 +z t ·h t

[0091] Where σ is the sigmoid function, W is the weight matrix, and x t is the text data input at time t, c t-1 is the text data input at the previous moment, r t is the reset gate, z t is the update gate, h t is the state to be activated, c t is the current output moment, W z represents the weight matrix of the update gate, W r represents the weight matrix of the reset gate;

[0092] The output sequence of the BiGRU layer is input into the upper CapsNet capsule network for data classification inference.

[0093] S4: Use the CapsNet network to perform deep text feature learning to achieve automatic classification and grading of power data.

[0094] Furthermore, in the process of power data classification, it is not enough to use CNN-BiGRU alone, because the pooling process can only roughly extract features, which may easily lead to the loss of important information. In the process of judging the entity level, it is also necessary to know the relative positions of these entities in order to extract the contribution of the corresponding features. For example, when three entities appear in the combination of "transmission line equipment information", the entity is defined as the "sensitive" level, but when it appears in the combination of "transmission line information equipment", it is defined as the "non-sensitive" level as a whole. The key lies in the special meaning of the entity "information". It is necessary to extract the comprehensive features of the entity, and use the CapsNet capsule network structure to predict the entity classification to realize power data entity recognition and classification. The capsule represents the entity classification label, the modulus length of the capsule vector represents the predicted probability of the entity label, and the direction of the capsule vector represents the attribute of the entity. All feature information in the capsule detection will be encapsulated by the capsule in the form of a vector. Since the capsule uses a vector representation instead of a scalar representation, it has a stronger ability to express entity information. CapsNet models such as Figure 6 As shown;

[0095] The CapsNet capsule network structure is used to predict entity classification. The capsule represents the entity classification label. The modulus length of the capsule vector represents the predicted probability of the entity label. The direction of the capsule vector represents the attribute of the entity, thus realizing the entity recognition and classification of power data.

[0096] Among them, the capsule network includes a lower-level capsule layer to transmit local features; a dynamic routing layer to realize information transmission between the lower-level capsule layer and the upper-level capsule layer; and an upper-level capsule layer to learn the overall features and output the final classification and grading results.

[0097] It should be noted that the dynamic routing process is as follows Figure 8 As shown in the figure, the prediction vector from the i-th lower-level capsule to the j-th upper-level capsule is calculated to determine the weight matrix; the main capsule output vector is calculated; the connection probability of the i-th capsule connecting to the j-th capsule is updated; the capsule input and capsule output are calculated; the process is repeated until the algorithm converges, which is specifically expressed as:

[0098] u j|i =W ij ·u i

[0099]

[0100] b ij =b ij +u j|i ·v j

[0101] Among them, u j|irepresents the prediction vector from the i-th lower-level capsule to the j-th upper-level capsule, W ij represents the current weight matrix, u i represents the main capsule output vector, c ij represents the connection probability of the i-th capsule connecting to the j-th capsule, s j represents capsule input, v j represents the capsule output, b ij Represents the weight of the capsule.

[0102] Embodiment 2, the second embodiment of the present invention, is different from the previous embodiment in that:

[0103] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0104] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0105] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0106] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0107] Example 3, reference Fig. 9 , which is an embodiment of the present invention, provides an automatic classification and grading system for power data based on a multi-layer hybrid model, characterized by: comprising a data preprocessing module 1, a word vector training module 2, a feature extraction network module 3, a CapsNet classification and grading module 4, and a result output and evaluation module 5;

[0108] Data preprocessing module 1 is responsible for collecting raw power data, cleaning and formatting the data, and converting the power data entity phrases into a form suitable for Skip_Gram model training;

[0109] Word vector training module 2 uses the Skip_Gram model to train word vectors for power data entity phrases, generates vectors that can represent word semantics, and uses words that appear in fixed combinations as single word representations;

[0110] Feature extraction network module 3, including multi-layer convolutional neural network CNN and BiGRU layer. Multi-layer CNN extracts semantic information from the data trained with word vectors through input layer, convolution layer, pooling layer and fully connected layer; BiGRU layer further extracts context information from the character vector sequence to provide deep features for CapsNet network;

[0111] CapsNet classification and grading module 4 uses the CapsNet network structure to perform deep feature learning on power data to achieve automatic classification and grading of data;

[0112] The result output and evaluation module 5 receives the output results of the CapsNet classification and grading module, parses and displays the results, and provides an evaluation mechanism to verify the accuracy and reliability of the classification and grading results. At the same time, it feeds back the results to other systems or performs subsequent data analysis and processing as needed.

[0113] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for automatic classification and grading of power data based on a multi-layer hybrid model, characterized by: include, Use Skip_Gram model to train and represent word vectors for power data entity phrases; The trained word vectors are input into a multi-layer convolutional neural network for feature learning to extract semantic information of power text data; Extract contextual information from the character vector sequence through the BiGRU layer; The CapsNet network is used to perform deep text feature learning to achieve automatic classification and grading of power data.

2. The method for automatic classification and grading of power data based on a multi-layer hybrid model according to claim 1, characterized in that: The Skip-Gram model includes predicting context words through professional power vocabulary entities, selecting the Skip-Gram model to train power professional vocabulary, and training power text data to represent words that appear in fixed combinations as single words.

3. The method for automatic classification and grading of power data based on a multi-layer hybrid model according to claim 2, characterized in that: The multi-layer convolutional neural network includes an input layer, a convolutional layer, a pooling layer and a fully connected layer to effectively extract semantic information of power text data; Input layer, receiving word vectors trained by Skip_Gram model; Convolutional layer, uses ReLU function as activation function and sets filters of different sizes for feature extraction; The pooling layer uses the maximum pooling method to extract the local optimal solution from the feature map output by the convolutional layer; The fully connected layer concatenates the vectors obtained by the pooling layer into a feature matrix to provide input for the subsequent BiGRU layer.

4. The method for automatic classification and grading of power data based on a multi-layer hybrid model according to claim 3, characterized in that: The input layer includes taking the trained word vector as input, which is x i ∈R m×d , where m is the number of words and d is the vector dimension; The convolution layer uses the ReLU function as the activation function, and the deep features are extracted by filters with sizes of 3, 4, and 5 respectively: c i =f(Wx i：i+h-1 +b) Among them, f is the activation function, W is the weight information, x i：i+h-1 is the local feature matrix of the text in the convolution window, i.e., the sentence vector; h is the convolution kernel size, and b is the bias value; The pooling layer performs maximum pooling on the feature map output by the convolutional layer, extracts the maximum value in each local area as the representative feature of the current area, forms the output feature map of the pooling layer, and expresses the feature map finally extracted by maxpooling as: O=max(c i ) The fully connected layer, since the BiGRU layer takes the serialized structure as input, the vectors obtained by the pooling layer are concatenated into a feature matrix H through the fully connected layer: H=[O1,O2,...O n ] Among them, n represents the number of convolution results, which are finally passed to the BiGRU layer to extract contextual feature relationships.

5. The method for automatic classification and grading of power data based on a multi-layer hybrid model according to claim 4, characterized in that: The BiGRU layer includes extracting context information from a character vector sequence using BiGRU, and outputting a sequence determined by the common state of the GRUs in two directions by splicing the forward GRU unit and the reverse GRU unit; The calculation formula of the GRU model is expressed as: z t =σ(W z ·[c t-1 ,x t ]) r t =σ(W r ·[c t-1 ,x t ]) h t =tanh(W·[r t ·c t-1 ,x t ]) c t =(1-z t )·c t-1 +z t ·h t Among them, σ is the sigmoid function, W is the weight matrix, x t is the text data input at time t, c t-1 is the text data input at the previous moment, r t is the reset gate, z t is the update gate, h t is the state to be activated, c t is the current output moment, W z represents the weight matrix of the update gate, W r represents the weight matrix of the reset gate; The output sequence of the BiGRU layer is input into the upper CapsNet capsule network for data classification inference.

6. The method for automatic classification and grading of power data based on a multi-layer hybrid model according to claim 5, characterized in that: The CapsNet network includes using the CapsNet capsule network structure to predict entity classification, capsules represent entity classification labels, the modulus length of the capsule vector represents the prediction probability of the entity label, and the direction of the capsule vector represents the attribute of the entity, so as to realize power data entity recognition and classification; Among them, the capsule network includes a lower-level capsule layer to transmit local features; a dynamic routing layer to realize information transmission between the lower-level capsule layer and the upper-level capsule layer; and an upper-level capsule layer to learn the overall features and output the final classification and grading results.

7. The method for automatic classification and grading of power data based on a multi-layer hybrid model according to claim 6, characterized in that: The dynamic routing layer includes calculating the prediction vector from the ith lower capsule to the jth upper capsule and determining the weight matrix; calculating the main capsule output vector; updating the connection probability of the ith capsule connecting to the jth capsule; calculating the capsule input and the capsule output; and repeating the process until the algorithm converges, which is specifically expressed as follows: u j|i =W ij ·u i b ij =b ij +u j|i ·n j Among them, u j|i represents the prediction vector from the i-th lower-level capsule to the j-th upper-level capsule, W ij represents the current weight matrix, u i represents the main capsule output vector, c ij represents the connection probability of the i-th capsule connecting to the j-th capsule, s j represents capsule input, v j represents the capsule output, b ij Represents the weight of the capsule.

8. A system using the method for automatic classification and grading of power data based on a multi-layer hybrid model as claimed in any one of claims 1 to 7, characterized in that: It includes data preprocessing module, word vector training module, feature extraction network module, CapsNet classification module, and result output and evaluation module; The data preprocessing module is responsible for collecting raw power data, performing data cleaning and formatting, and converting power data entity phrases into a form suitable for Skip_Gram model training; The word vector training module uses the Skip_Gram model to perform word vector training on power data entity phrases, generates vectors that can represent the semantics of the words, and uses words that appear in a fixed combination as single word representations; The feature extraction network module includes a multi-layer convolutional neural network (CNN) and a BiGRU layer. The multi-layer CNN extracts semantic information from the data trained with word vectors through an input layer, a convolution layer, a pooling layer, and a fully connected layer. The BiGRU layer further extracts context information from the character vector sequence to provide deep features for the CapsNet network. The CapsNet classification and grading module uses the CapsNet network structure to perform deep feature learning on power data to achieve automatic classification and grading of data; The result output and evaluation module receives the output results of the CapsNet classification and grading module, parses and displays the results, and provides an evaluation mechanism to verify the accuracy and reliability of the classification and grading results. At the same time, it feeds back the results to other systems or performs subsequent data analysis and processing as needed.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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

Images