AI technology-based power grid regulation and control service auxiliary office application system and method

Abstract

The invention discloses a power grid regulation and control service auxiliary office application method and system based on the AI technology, and belongs to the field of artificial intelligence, the system comprises a data preparation and preprocessing module, a model construction module, a model combination module and a model training and optimization module, data is acquired from an OMS system, historical data and annotated data are collected and sorted, and preprocessing operation is carried out; a BERT model is loaded, a DPCNN architecture is defined, and parameters and configurations required by model training are prepared; the output of the BERT model is connected with the input of the DPCNN, and a complete BERT-DPCNN model is constructed; model training, parameter updating, performance monitoring and test evaluation. According to the invention, the complex requirements of each business process of a scheduling domain are met, the working efficiency of full-business regulation and control operation of the power grid is improved, and the digital transformation of enterprises and the construction of digital power grids are further promoted; a plurality of existing production management systems in the domain are dispatched in a linkage and cooperative mode, and regulation and control integrated construction and supplement of the dispatching intelligent anti-misoperation technology are carried out.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an auxiliary office application system and method for power grid control business based on AI technology. Background Art

[0002] During the duty process of power grid control and operation, the controller not only needs to dispatch, monitor and control the power grid operation in real time and uninterruptedly, but also needs to spend a lot of time and energy to complete various auxiliary business processes, such as: preparing dispatch operation tickets according to professional operation opinions on power grid operation mode, logging defects, faults and other abnormal situations that occur during the duty process of control and operation, preparing real-time risk warning notices for power grid operation and corresponding dispatch disposal plans, etc.

[0003] There are three specific problems:

[0004] First, the dispatching operation ticket is flexible and changeable, and manual invoicing affects efficiency. The operation mode of the power grid is flexible and changeable, and the corresponding dispatching operation tickets are also different. Although there is a historical typical invoicing function, it still needs to be manually modified each time according to the current power grid operation mode. In addition, when filling out the dispatching operation ticket, the controller needs to switch back and forth in multiple production management systems to check the information in accordance with the requirements of the "three comparisons" to ensure the logical correctness of the operation ticket. When the maintenance work is more concentrated, the dispatcher takes a long time to fill in and review the operation ticket, which affects the operation efficiency. At the same time, errors or irregularities in manually filling out the operation ticket also occur from time to time.

[0005] Second, there are many daily operation records, and the manual recording task is arduous. The daily operation records of the control mainly include: shift handover logs, power supply reports, equipment defect records, accident and abnormal handling process records, etc. Although various types of daily operation information have corresponding information management modules in the dispatching and operation management system, the degree of intelligence is not high. When recording various types of daily operation information, the on-duty controller still needs to switch multiple systems to collect and integrate information, which is complicated and cumbersome. In addition, manual records are often incomplete or omitted, which poses a safety hazard to the operation of the power grid.

[0006] Third, the formats and specifications of official documents are complex, and document processing is overly time-consuming and laborious. During the operation of power grid regulation, official documents need to be frequently sent to other professionals, departments and users, such as regulation operation information letters, new equipment startup plans, and real-time power grid operation risk notices. These official documents have standardized terminology and format requirements. A large amount of document processing increases the duty burden of the regulators, distracts the regulators' energy, and brings risks to the operation of power grid regulation.

[0007] At present, the following problems still exist in the existing grid control and operation interaction projects: semantic analysis and knowledge graphs are highly dependent on a large amount of high-quality data, and professional knowledge is required to maintain and update the knowledge base, and there may be difficulties in handling and understanding ambiguities; deep learning and neural networks require a large amount of labeled data for training, the model has poor interpretability, and requires high computing resources; natural language processing has limited understanding and processing capabilities for domain-specific languages, and may have difficulties in dealing with ambiguity and contextual relevance; data privacy and security issues of big data technology, data quality and consistency are difficult to ensure, and professional data management and processing skills are required; stability and scalability challenges of artificial intelligence platforms, complex resource allocation and management in a multi-user environment, and high requirements for developer skills; user adaptability and acceptance issues of human-computer interaction technology, the complexity of interaction design, and reliability and stability issues in specific environments; graphical operations and intelligent approvals, challenges in the intuitiveness and ease of use of the user interface, may have limitations in complex operations, and are highly dependent on graphical technology. Summary of the invention

[0008] In view of the above existing problems, the present invention can analyze learning history data, predict and judge auxiliary business needs, and efficiently generate written text to meet complex scheduling needs. This method makes full use of the advantages of the BERT pre-training model in text semantic understanding to generate high-quality text word vectors. On this basis, BERT is combined with the optimized DPCNN (deep convolutional neural network) to deeply optimize the text feature extraction model, not only replacing the original output layer, but also fine-tuning the entire model structure. In particular, in the convolution block design of DPCNN, the Mish activation function is cleverly used to replace the traditional ReLU activation function. This improvement enables the model to capture key information in the text more keenly, while effectively alleviating the problem of gradient explosion and enhancing the stability and robustness of the model. In addition, this paper also introduces a dynamic learning rate mechanism to further optimize the training method of the model, so that the model can adaptively adjust the learning rate during the training process, thereby improving the training efficiency while also ensuring the final performance of the model.

[0009] In order to solve the above technical problems, an auxiliary office application system for power grid control business based on AI technology is proposed, including data preparation and preprocessing module, model building module, model combination module, model training and optimization module;

[0010] The data preparation and preprocessing module obtains data from the system, organizes the collected data, cleans, standardizes and segments the data, and improves the model's ability to understand the text and the effectiveness of subsequent processing;

[0011] The model building module loads the pre-trained BERT model to obtain the context information of the text and generate word vectors, designs the DPCNN convolutional block to extract deep text features, and expands the feature extraction capability of the BERT model;

[0012] The model combines the modules to connect the last hidden state of BERT output with the input of DPCNN to form a coherent data flow. The convolutional blocks and pooling layers of DPCNN process the data in sequence, extract deep features and map the features to classification labels through the fully connected layers for text classification.

[0013] The model training and optimization module uses the word vectors obtained through BERT training to perform forward propagation and back propagation, dynamically adjusts the learning rate and optimizer parameters, improves the training strategy by monitoring the loss function and accuracy, and uses the softmax function to output the probability of each classification to provide a basis for decision-making.

[0014] As a preferred solution of the auxiliary office application system for power grid control business based on AI technology described in the present invention, wherein: the data preparation and preprocessing module includes obtaining three types of data including maintenance application form data, business rule data and annotation data through the system;

[0015] The maintenance application form data includes obtaining maintenance application form data from the OMS system, including operation policy approval opinions, power outage scope, and key information affecting the equipment, and inputting the maintenance application form data into the pre-trained BERT model of the model building module for training and testing;

[0016] The business rule data includes collecting and organizing historical typical operation ticket data, new equipment commissioning and startup plans, and risk warning notices, providing text samples, and inputting the business logic and rules of power grid regulation and operation into the pre-trained BERT model of the model building module for learning;

[0017] The labeled data includes collecting and organizing dispatching procedures and grid-related annotations, and inputting the data into the pre-trained BERT model of the model building module for supervised learning;

[0018] The three types of data obtained are preprocessed, including text cleaning, standardization and word segmentation;

[0019] The text cleaning includes removing irrelevant characters and punctuation marks, the standardization includes unifying text formats and units, and the word segmentation includes splitting the text into words or phrases so that the model can understand the text semantics.

[0020] As a preferred solution of the power grid control business auxiliary office application system based on AI technology described in the present invention, the model construction module includes loading a BERT model from a pre-trained model library, and the model defines 12 Transformer layers, 768 hidden units, 12 self-attention Headers and 110 million parameters;

[0021] Use Tokenizer to segment the original text, map the segmentation results to the predefined embedding matrix to obtain the corresponding embedding vector, and input it into the Transformer encoder of BERT for further processing through the self-attention mechanism;

[0022] Define the convolution blocks of the DPCNN model, each of which contains 250 fixed-size convolution kernels, apply the DPCNN architecture to the output of the BERT model, and use the last hidden state of the BERT model as the input of the DPCNN.

[0023] As a preferred solution of the power grid control business auxiliary office application system based on AI technology described in the present invention, the self-attention mechanism is further processed to generate a word representation containing context information through the self-attention mechanism:

[0024]

[0025] Among them, Q, K and V are matrices composed of input word vectors, d k is the dimension of the key vector, used to scale the dot product results to prevent the gradients from vanishing or exploding.

[0026] As a preferred solution of the power grid control business auxiliary office application system based on AI technology described in the present invention, wherein: the model combination module includes text classification and classification prediction;

[0027] The text classification is to connect the output of the BERT model with the input of the DPCNN, and pass the last hidden state of BERT to the first convolutional block of the DPCNN;

[0028] The convolution blocks and pooling layers in the DPCNN architecture process the input data in sequence to extract deep text features. Each convolution block contains two convolution layers and one pooling layer. The convolution layer uses equal-length convolution, and the pooling layer uses 1 / 2 maximum pooling.

[0029] The output of DPCNN is mapped to the classification label through the fully connected layer to complete the text classification task;

[0030] The classification prediction is that the fully connected layer converts the feature vector extracted by DPCNN into classification probability to perform text classification prediction.

[0031] As a preferred solution of the auxiliary office application system for power grid control business based on AI technology described in the present invention, the model training and optimization module includes: using BERT to convert each word in the text into a word vector at the input layer, and splicing it into a vector matrix, inputting the vector matrix into the BERT model, and performing forward propagation to calculate the loss;

[0032] The gradient is calculated through the back-propagation algorithm, and the parameters of the BERT model are updated. The output of the BERT model is used as the input of the DPCNN network and is jointly sent to the DPCNN network. During the training process, the BertAdam optimizer and learning rate strategy are used to adjust the training strategy.

[0033] The definition of DPCNN network consists of 7 convolutional blocks, each of which contains two convolutional layers and one pooling layer;

[0034] When the convolution process generates features, it uses convolution kernels of equal length for matrix multiplication and uses the Mish activation function for nonlinear transformation;

[0035] The transformed vector is passed into the Transformer encoding layer to extract vector features and dynamically adjusted using the BertAdam algorithm;

[0036] The output layer uses the softmax function to assign a probability value to each classification result, indicating the possibility of each classification;

[0037] The optimal model is tested using the test dataset to evaluate its performance in real application scenarios.

[0038] Another object of the present invention is to provide an auxiliary office application method for power grid control business based on AI technology. The present invention improves work efficiency, reduces manual intervention through automated processing and intelligent analysis, and quickly processes power grid control documents and applications. At the same time, the method aims to enhance decision support, help dispatchers analyze historical data more accurately, make reasonable decisions, and thus reduce risks and losses. Through advanced AI models (such as BERT and DPCNN), the method improves text processing capabilities, realizes intelligent question-and-answer functions, and enables staff to quickly obtain required information. In addition, it supports data-driven business optimization and risk management, helps to identify and assess potential risks, ensure the safety and stability of power grid operation, and thus promote the intelligent transformation and development of the power industry.

[0039] As a preferred solution of the auxiliary office application method for power grid control business based on AI technology described in the present invention, it is characterized by comprising:

[0040] Acquire data from the OMS system, collect and organize historical data and annotated data, and perform preprocessing operations; load the BERT model, define the DPCNN architecture, and prepare the parameters and configurations required for model training; connect the output of the BERT model with the input of the DPCNN, and build a complete BERT-DPCNN model; train and update parameters, monitor performance, and test and evaluate the BERT-DPCNN model.

[0041] As a preferred solution of the auxiliary office application method for power grid control business based on AI technology described in the present invention, the training of the BERT-DPCNN model includes using BERT to convert each word in the text into a word vector at the input layer through the Transformer encoder in the semantic extraction layer, and concatenating each word vector to obtain a vector matrix X, the formula is as follows:

[0042]

[0043] Among them, n is the number of words in the sequence, and the vector obtained by each word after passing through the encoder is x1, x2, ..., x n , Represents a splicing operation;

[0044] The BERT model forward propagates to calculate the loss, calculates the gradient through the back-propagation algorithm, and updates the model parameters;

[0045] The CLS vector output by the BERT model and the vectors of each token in the extracted question and answer sentences are used as the input of the DPCNN network to fine-tune the BERT model. The BertAdam optimizer and learning rate strategy are used to monitor the model's loss function and accuracy to adjust the training strategy.

[0046] The DPCNN network is defined to consist of 7 convolution blocks, each of which contains two convolution layers and one pooling layer. The convolution process generates features using equal-length convolutions, where h is the size of the convolution kernel W and the feature S i It is expressed as:

[0047]

[0048] Among them, S i is the value at position j in the output feature map, W i is the weight of position i in the convolution kernel, X j+i-1 is the element in the input data corresponding to the current position of the convolution kernel, k is the size of the convolution kernel, b is the bias term, and f is the nonlinear transformation function;

[0049] After performing matrix multiplication and obtaining the result, use the activation function:

[0050] f(x)=xtanh(ln(1+e x )

[0051] The model obtains high-level features of the data, passes the transformed vector into the Transformer encoding layer to extract vector features, uses the BertAdam algorithm to dynamically adjust the learning rate to make the model converge quickly in the early stage of training, and adjusts the weights in the later stage of training. The output layer uses the softmax function to assign a probability value to each classification result, indicating the possibility of each classification:

[0052]

[0053] Among them, y i is the output of each node, n is the number of output nodes, that is, the number of classification categories;

[0054] The optimal model is tested using the test dataset to evaluate the performance of the model in real application scenarios.

[0055] 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 the auxiliary office application method for power grid regulation business based on AI technology are implemented.

[0056] 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 auxiliary office application method for power grid regulation business based on AI technology are implemented.

[0057] Beneficial effects of the present invention: The present invention introduces generative AI technology into the field of electric power dispatching. Previously, similar grid control and operation interaction projects only deployed traditional analytical artificial intelligence (Analytical AI), which analyzed and learned historical static data to achieve prediction and judgment. However, traditional analytical AI does not have strong understanding and processing capabilities for natural language, and there are many inconveniences and limitations when using analysis; in addition, its ability to generate content and make decisions is relatively limited, and it cannot adapt to the complex and changeable grid operation needs. Generative AI technology has powerful natural language understanding, content generation and innovation capabilities, which can meet the complex needs of various business processes in the dispatching domain, greatly improve the work efficiency of the entire grid control and operation business, and further promote the digital transformation of enterprises and the construction of digital power grids.

[0058] Linkage and coordination of multiple existing production management systems in the dispatching domain. In previous similar projects, only the focus was on improving or optimizing a certain business process or existing production management system for power grid regulation and operation. Each production management system was still relatively independent, and there was a large asymmetry in information, which affected work efficiency. This project introduced a generative AI large model interface to coordinate multiple existing production management systems in the dispatching domain, and realized the intelligent integration and processing of various dispatching operation information, which greatly reduced the text work burden of dispatchers and enabled dispatchers to focus more on professional work such as power grid operation and accident handling.

[0059] A supplement to the construction of integrated control and dispatching intelligent error prevention technology. Previous projects related to technologies such as integrated control and dispatching intelligent error prevention focused on the technical error prevention and system construction of the core business of power grid control and operation, such as dispatching orders and equipment operation; while less attention was paid to the risks of misoperation and mis-dispatch that may exist in other auxiliary business processes of control and operation. For example, errors in the instructions of the dispatching operation ticket may cause misoperation in the case of omissions in manual review; or mis-dispatch may occur due to the omission of important information due to incomplete dispatching operation logs. This project deploys a generative AI system to intelligently generate various types of dispatching operation information and records, effectively avoiding hidden dangers and risks of power grid operation such as mis-dispatch and mis-operation caused by errors in dispatching operation information. At the same time, it can also further reduce the on-duty burden of controllers, allowing controllers to focus more on power grid control and operation, which is a further supplement to the construction of integrated control and dispatching intelligent error prevention technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0061] Figure 1 A schematic diagram of the modules of an auxiliary office application system for power grid control business based on AI technology provided by one embodiment of the present invention.

[0062] Figure 2 The BERT_DPCNN overall model of the power grid control business auxiliary office application method based on AI technology is provided in one embodiment of the present invention.

[0063] Figure 3 A BERT_DPCNN model training and optimization flowchart of an auxiliary office application method for power grid control business based on AI technology provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0064] 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.

[0065] 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.

[0066] 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 an embodiment that is mutually exclusive with other embodiments, either individually or selectively.

[0067] 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.

[0068] 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, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying 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.

[0069] 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.

[0070] Example 1, reference Figure 1-Figure 2, which is the first embodiment of the present invention, and provides an auxiliary office application system for power grid control business based on AI technology, including a data preparation and preprocessing module 10, a model building module 20, a model combination module 30, and a model training and optimization module 40;

[0071] The data preparation and preprocessing module 10 obtains data from the system, organizes the collected data, cleans, standardizes and segments the data, and improves the model's ability to understand the text and the effectiveness of subsequent processing;

[0072] The model building module 20 loads the pre-trained BERT model to obtain the context information of the text and generate word vectors, designs the DPCNN convolutional block to extract deep text features, and expands the feature extraction capability of the BERT model;

[0073] The model combines module 30 to connect the last hidden state of BERT output with the input of DPCNN to form a coherent data flow. The convolutional blocks and pooling layers of DPCNN process the data in sequence, extract deep features and map the features to classification labels through the fully connected layer for text classification;

[0074] The model training and optimization module 40 uses the word vectors obtained through BERT training to perform forward propagation and back propagation, dynamically adjusts the learning rate and optimizer parameters, improves the training strategy by monitoring the loss function and accuracy, and uses the softmax function to output the probability of each classification to provide a basis for decision making.

[0075] The overall framework of the BERT-DPCNN model is as follows Figure 2 As shown in the figure, the short-circuit connection plays a vital role. It is responsible for adding the tensor before the pooling operation to the tensor of the current layer, and then continuously importing the fused result into the repeated structure of the model for downsampling. This process not only helps the model to mine text features at a deeper level, but also enhances the model's ability to process complex text information through feature fusion, providing strong technical support for intelligent text processing in the field of power grid dispatching and operation. Its implementation principle is as follows:

[0076] Furthermore, the data preparation and preprocessing module 10 obtains three types of data including maintenance application form data, business rule data and annotation data through the system;

[0077] Obtain maintenance application data from the OMS system, including key information such as operation approval opinions, power outage scope, and affected equipment. This data will be used for model training and testing.

[0078] Collect and organize historical typical operation ticket data, new equipment commissioning plans and risk warning notices. These data provide rich text samples, which help the model learn the business logic and rules of power grid regulation and operation.

[0079] Collect and organize relevant dispatching procedures and grid-related annotations. These annotation data will be used for supervised learning of the model to improve the prediction accuracy of the model.

[0080] The three types of data obtained are preprocessed, including text cleaning, standardization and word segmentation;

[0081] The text cleaning includes removing irrelevant characters and punctuation marks.

[0082] The standardization includes unified text formats and units.

[0083] The word segmentation includes splitting the text into words or phrases so that the model can understand the semantics of the text.

[0084] It should be noted that the model building module 20 first prepares the BERT model and loads the BERT model from the pre-trained model library. The model defines 12 Transformer layers, 768 hidden units, 12 self-attention Headers and 110 million parameters.

[0085] BERT uses a self-attention mechanism to obtain contextual semantic information of sentences. Before entering BERT, the original text is segmented by Tokenizer, and these segmented words are mapped to a predefined embedding matrix to obtain corresponding embedding vectors. These embedding vectors are then input into BERT's Transformer encoder and further processed by the self-attention mechanism to generate word representations containing contextual information. The self-attention calculation formula is:

[0086]

[0087] Where Q, K and V are matrices consisting of input word vectors, d k is the dimension of the key vector, used to scale the dot product results to prevent the gradients from vanishing or exploding.

[0088] Next, define the DPCNN architecture:

[0089] Define the convolution blocks of the DPCNN model. Each convolution block contains 250 convolution kernels of fixed size to extract text features. The size and number of convolution kernels can be adjusted according to actual needs.

[0090] The DPCNN architecture is applied on the output of the BERT model, and the last hidden state of the BERT model is used as the input of the DPCNN. In this way, the DPCNN can further extract deep features from the contextual word representations generated by BERT.

[0091] It should also be noted that the model combination module 30 includes text classification and classification prediction;

[0092] Connect the output of the BERT model to the input of the DPCNN and pass the last hidden state of BERT to the first convolutional block of the DPCNN.

[0093] The convolution blocks and pooling layers in the DPCNN architecture process the input data in sequence to extract deeper text features. Each convolution block contains two convolution layers and one pooling layer. The convolution layer uses equal-length convolution and the pooling layer uses 1 / 2 maximum pooling.

[0094] The output of DPCNN is mapped to the classification label through the fully connected layer to complete the text classification task.

[0095] The fully connected layer converts the feature vector extracted by DPCNN into classification probability, thereby realizing the classification prediction of text.

[0096] Furthermore, the model training and optimization module 40 uses BERT to convert each word in the text into a word vector at the input layer and concatenates them into a vector matrix. Then, the vector matrix is ​​input into the BERT model to perform forward propagation to calculate the loss.

[0097] The gradient is calculated through the back-propagation algorithm and the parameters of the BERT model are updated. This step will optimize the performance of the BERT model and enable it to better extract text semantic features.

[0098] The output of the BERT model is used as the input of the DPCNN network and is fed into the DPCNN network. During the training process, the BertAdam optimizer and learning rate strategy are used to monitor the model's loss function and accuracy and adjust the training strategy.

[0099] The DPCNN network is defined to consist of 7 convolution blocks, each of which contains two convolution layers and one pooling layer. In each convolution block, the two convolution layers use equal-length convolutions with a convolution kernel size of 250. The pooling layer uses size=3 and stride=2 for 1 / 2 maximum pooling.

[0100] When the convolution process generates features, it uses convolution kernels of equal length for matrix multiplication and uses the Mish activation function for nonlinear transformation. Compared with the traditional ReLU activation function, the Mish activation function can more keenly capture the key information in the text and effectively alleviate the problem of gradient explosion.

[0101] The transformed vector is passed into the Transformer encoding layer to extract vector features, and the BertAdam algorithm is used to dynamically adjust the learning rate. In the early stage of training, the learning rate is high so that the model converges quickly; in the later stage of training, the learning rate gradually decreases so that the model can fine-tune the weights and avoid overfitting.

[0102] The output layer uses the softmax function to assign a probability value to each classification result, indicating the possibility of each classification. In this way, the model can output the probability distribution of the prediction results for users to refer to and make decisions.

[0103] Use the test dataset to test the optimal model to evaluate its performance in real application scenarios. The test dataset should contain samples that are similar but different from the training dataset to ensure the generalization ability of the model.

[0104] The BERT_DPCNN model combines the advantages of BERT and DPCNN to improve the accuracy and efficiency of text classification. BERT, based on the Transformer architecture, captures the dependencies between words through bidirectional context, enhancing the text representation ability. DPCNN, a deep convolutional network designed for text classification, can effectively extract long-distance text dependencies. In BERT_DPCNN, the text is first encoded by BERT, making full use of its bidirectional modeling ability to obtain rich text embeddings. Subsequently, DPCNN performs deep feature extraction on these embeddings to further capture the complex relationships in the text. Finally, the processed features are classified through the classification layer, and the probability of the category to which the text belongs is output. This combination enables BERT_DPCNN to utilize both BERT's contextual understanding ability and DPCNN's advantages in deep feature extraction, thereby significantly improving the understanding of text. This project uses the BERT_DPCNN model, adds professional word meanings and corpus in the field of power grid regulation, realizes the auxiliary question-answering function in the field of power grid dispatching, and improves office efficiency.

[0105] Embodiment 2 is the second embodiment of the present invention, which provides an auxiliary office application system for power grid control business based on AI technology. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0106] In the experiment, the training data and environment set parameters are set according to Tables 1, 2, and 3. From epoch 1 to 10,000, the training loss (Train Loss) and verification loss (Val Loss) of the model gradually decrease, and the training accuracy (Train Acc) and verification accuracy (Val Acc) gradually increase, which shows the optimization effect of the model during the training process.

[0107] Table 1: Experimental data

[0108] Total data Number of training sets Validation set number Number of test sets 22,800 27,000 0.06 million 0.06 million

[0109] Table 2: Experimental environment

[0110] Related Systems Configuration operating system Windows 11 processor 12thGenIntel(R)Core(TM)i7-12700H2.30GHz Memory 40.0GB language Python 3.8.10 frame 2.4.0+cpu

[0111] Table 3: Experimental parameter settings

[0112] parameter Configuration 10000 Sentence length (short fill long cut) 32 Attention Head 128 Learning Rate 5e-5 Number of convolution kernels Python 3.8.10 Hidden layer dimensions 768

[0113] Table 4: Optimization training results

[0114] epoch Training loss Training accuracy Validation loss Verification accuracy 1 2.31 10.12％ 2.4 8.01％ 2 2.12 19.89％ 2.2 13.1％ 3 1.28 25.3％ 2.05 22.36％ ... ... ... ... ... 9999 0.5 95.99％ 1.12 90.32％ 10000 0.49 96.01％ 1.1 91.03％

[0115] Embodiment 3, the third embodiment of the present invention, is different from the first two embodiments in that:

[0116] 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, and other media that can store program codes.

[0117] 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.

[0118] 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.

[0119] 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.

[0120] Example 4, reference Figure 3 , which is the fourth embodiment of the present invention, and provides an auxiliary office application method for power grid control business based on AI technology, including:

[0121] like Figure 3 As shown in the figure, the BERT model, DPCNN model, the connection relationship between them, and the model training and optimization process are shown.

[0122] Acquire data from the OMS system, collect and organize historical data and annotated data, and perform preprocessing operations.

[0123] Load the BERT model, define the DPCNN architecture, and prepare the parameters and configurations required for model training.

[0124] Connect the output of the BERT model with the input of the DPCNN and build a complete BERT-DPCNN model.

[0125] Model training, parameter updating, performance monitoring, and test evaluation.

[0126] Further, model training and optimization module:

[0127] First, BERT is used to convert each word in the text into a word vector in the input layer. This process is processed by the Transformer encoder in the semantic extraction layer. Then, each word vector is concatenated to obtain the vector matrix X. The formula is as follows:

[0128]

[0129] Among them, n is the number of words in the sequence, and the vector obtained by each word after passing through the encoder is x1, x2, ..., x n , represents a concatenation operation (in matrices, this usually means stacking these vectors row by row).

[0130] The BERT model forward propagates to calculate the loss, calculates the gradient through the back-propagation algorithm, and updates the model parameters.

[0131] The output of the BERT model is used as the input of the DPCNN network. In addition to the [CLS] vector output by BERT, vector representations of each token in the question and answer sentences are extracted. These vectors are fed into the DPCNN network together with the BERT model for fine-tuning using the BertAdam optimizer and learning rate strategy, monitoring the model's loss function and accuracy, and adjusting the training strategy.

[0132] The DPCNN network is defined to consist of 7 convolution blocks, each of which contains two convolution layers and one pooling layer. In each convolution block, the two convolution layers use equal-length convolutions with a convolution kernel size of 250. The pooling layer uses size=3 and stride=2 for 1 / 2 maximum pooling. This design allows the length of the sequence to be compressed to half of its original length after each convolution block, thereby doubling the range of text segments that can be perceived.

[0133] The convolution process generates features using equal length convolution. h is the size of the convolution kernel W, and the feature S i can be expressed by the following formula:

[0134]

[0135] Among them, S i is the value at position j in the output feature map, W i is the weight of position i in the convolution kernel, X j+i-1 is the element in the input data corresponding to the current position of the convolution kernel, k is the size of the convolution kernel, b is the bias term, and f is the nonlinear transformation function.

[0136] Its process is to perform matrix multiplication first, and then use the activation function after getting the result:

[0137] f(x)=xtanh(ln(1+ex ))

[0138] Perform nonlinear transformation and finally perform maximum pooling operation.

[0139] After the above steps, the model obtains high-level features of the data.

[0140] Then the transformed vector is passed to the Transformer encoding layer to extract vector features. The BertAdam algorithm is used to dynamically adjust the learning rate so that the model can converge quickly in the early stage of training and finely adjust the weights in the later stage of training to avoid overfitting. Finally, the output layer uses the softmax function to assign a probability value to each classification result to indicate the possibility of each classification. The function is as follows:

[0141]

[0142] Among them, y i is the output of each node, and n is the number of output nodes, that is, the number of classification categories.

[0143] Finally, the optimal model is tested using the test dataset to evaluate its performance in real application scenarios.

[0144] 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. The auxiliary office application system for power grid control business based on AI technology is characterized by: It includes data preparation and preprocessing module, model building module, model combination module, model training and optimization module; The data preparation and preprocessing module obtains data from the system, organizes the collected data, cleans, standardizes and segments the data, and improves the model's ability to understand the text and the effectiveness of subsequent processing; The model building module loads the pre-trained BERT model to obtain the context information of the text and generate word vectors, designs the DPCNN convolutional block to extract deep text features, and expands the feature extraction capability of the BERT model; The model combines the modules to connect the last hidden state of BERT output with the input of DPCNN to form a coherent data flow. The convolutional blocks and pooling layers of DPCNN process the data in sequence, extract deep features and map the features to classification labels through the fully connected layers for text classification. The model training and optimization module uses the word vectors obtained through BERT training to perform forward propagation and back propagation, dynamically adjusts the learning rate and optimizer parameters, improves the training strategy by monitoring the loss function and accuracy, and uses the softmax function to output the probability of each classification to provide a basis for decision-making.

2. The power grid control business auxiliary office application system based on AI technology as claimed in claim 1, characterized in that: The data preparation and preprocessing module includes obtaining three types of data including maintenance application form data, business rule data and annotation data through the system; The maintenance application form data includes obtaining maintenance application form data from the OMS system, including operation policy approval opinions, power outage scope, and key information affecting the equipment, and inputting the maintenance application form data into the pre-trained BERT model of the model building module for training and testing; The business rule data includes collecting and organizing historical typical operation ticket data, new equipment commissioning and startup plans, and risk warning notices, providing text samples, and inputting the business logic and rules of power grid regulation and operation into the pre-trained BERT model of the model building module for learning; The labeled data includes collecting and organizing dispatching procedures and grid-related annotations, and inputting the data into the pre-trained BERT model of the model building module for supervised learning; The three types of data obtained are preprocessed, including text cleaning, standardization and word segmentation; The text cleaning includes removing irrelevant characters and punctuation marks, the standardization includes unifying text formats and units, and the word segmentation includes splitting the text into words or phrases so that the model can understand the text semantics.

3. The power grid control business auxiliary office application system based on AI technology as claimed in claim 2, characterized in that: The model building module includes loading the BERT model from the pre-trained model library, where the model defines 12 Transformer layers, 768 hidden units, 12 self-attention Headers, and 110 million parameters; Use Tokenizer to segment the original text, map the segmentation results to the predefined embedding matrix to obtain the corresponding embedding vector, and input it into the Transformer encoder of BERT for further processing through the self-attention mechanism; Define the convolution blocks of the DPCNN model, each of which contains 250 fixed-size convolution kernels, apply the DPCNN architecture to the output of the BERT model, and use the last hidden state of the BERT model as the input of the DPCNN.

4. The power grid control business auxiliary office application system based on AI technology as claimed in claim 3 is characterized by: The self-attention mechanism is further processed by generating word representations containing contextual information through the self-attention mechanism: Among them, Q, K and V are matrices composed of input word vectors, d k is the dimension of the key vector, used to scale the dot product results to prevent the gradients from vanishing or exploding.

5. The power grid control business auxiliary office application system based on AI technology as claimed in claim 4, characterized in that: The model combination module includes text classification and classification prediction; The text classification is to connect the output of the BERT model with the input of the DPCNN, and pass the last hidden state of BERT to the first convolutional block of the DPCNN; The convolution blocks and pooling layers in the DPCNN architecture process the input data in sequence to extract deep text features. Each convolution block contains two convolution layers and one pooling layer. The convolution layer uses equal-length convolution, and the pooling layer uses 1 / 2 maximum pooling. The output of DPCNN is mapped to the classification label through the fully connected layer to complete the text classification task; The classification prediction is that the fully connected layer converts the feature vector extracted by DPCNN into classification probability to perform text classification prediction.

6. The power grid control business auxiliary office application system based on AI technology as claimed in claim 5, characterized in that: The model training and optimization module includes using BERT to convert each word in the text into a word vector at the input layer, and concatenate them into a vector matrix, input the vector matrix into the BERT model, and perform forward propagation to calculate the loss; The gradient is calculated through the back-propagation algorithm, and the parameters of the BERT model are updated. The output of the BERT model is used as the input of the DPCNN network and is jointly sent to the DPCNN network. During the training process, the BertAdam optimizer and learning rate strategy are used to adjust the training strategy. The definition of DPCNN network consists of 7 convolutional blocks, each of which contains two convolutional layers and one pooling layer; When the convolution process generates features, it uses convolution kernels of equal length for matrix multiplication and uses the Mish activation function for nonlinear transformation; The transformed vector is passed into the Transformer encoding layer to extract vector features, and dynamically adjusted using the BertAdam algorithm; The output layer uses the softmax function to assign a probability value to each classification result, indicating the possibility of each classification; The optimal model is tested using the test dataset to evaluate its performance in real application scenarios.

7. A method for using the AI-based power grid control business auxiliary office application system as claimed in any one of claims 1 to 6, characterized in that: Obtain data from the OMS system, collect and organize historical data and annotated data, and perform pre-processing operations; Load the BERT model, define the DPCNN architecture, and prepare the parameters and configurations required for model training; Connect the output of the BERT model with the input of the DPCNN and build a complete BERT-DPCNN model; Perform training and parameter update, performance monitoring, and test evaluation of the BERT-DPCNN model.

8. The power grid control business auxiliary office application system based on AI technology as claimed in claim 7, characterized in that: The training of the BERT-DPCNN model includes using BERT to convert each word in the text into a word vector at the input layer through the Transformer encoder in the semantic extraction layer, and concatenating each word vector to obtain a vector matrix X, as shown in the following formula: Among them, n is the number of words in the sequence, and the vector obtained by each word after passing through the encoder is x1, x2, ..., x n , Represents a splicing operation; The BERT model forward propagates to calculate the loss, calculates the gradient through the back-propagation algorithm, and updates the model parameters; The CLS vector output by the BERT model and the vectors of each token in the extracted question and answer sentences are used as the input of the DPCNN network to fine-tune the BERT model. The BertAdam optimizer and learning rate strategy are used to monitor the model's loss function and accuracy to adjust the training strategy. The DPCNN network is defined to consist of 7 convolution blocks, each of which contains two convolution layers and one pooling layer. The convolution process generates features using equal-length convolutions, where h is the size of the convolution kernel W and the feature S i It is expressed as: Among them, S i is the value at position j in the output feature map, W i is the weight of position i in the convolution kernel, X j+i-1 is the element in the input data corresponding to the current position of the convolution kernel, k is the size of the convolution kernel, b is the bias term, and f is the nonlinear transformation function; After performing matrix multiplication and obtaining the result, use the activation function: f(x)=xtanh(ln(1+e x ) The model obtains high-level features of the data, passes the transformed vector into the Transformer encoding layer to extract vector features, uses the BertAdam algorithm to dynamically adjust the learning rate to make the model converge quickly in the early stage of training, and adjusts the weights in the later stage of training. The output layer uses the softmax function to assign a probability value to each classification result, indicating the possibility of each classification: Among them, y i is the output of each node, n is the number of output nodes, that is, the number of classification categories; The optimal model is tested using the test dataset to evaluate the performance of the model in real application scenarios.

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 auxiliary office application method for power grid control business based on AI technology described in 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 auxiliary office application method for power grid control business based on AI technology described in any one of claims 1 to 7 are implemented.

Images