Novel power system knowledge tracking method based on co-occurrence network and deep learning

Abstract

The invention discloses a novel electric power system knowledge tracking method based on a co-occurrence network and deep learning, and belongs to the technical field of electric power system knowledge tracking. Comprising the following steps: step 1, collecting literature data of the novel power system, and carrying out feature analysis and subject term selection; 2, drawing a word cloud picture by using a wordcloud word cloud picture display library of Python, and analyzing a research hotspot and a development trend of each year; drawing a keyword co-occurrence network by using a network workx library of Python, and analyzing the relevance between keywords; 3, analyzing the network structure of the GRU, setting GRU model parameters, and determining evaluation indexes of the GRU model; and 4, predicting the hot words of the novel power system by using the GRU network. According to the method, research hotspots and development trends in the field of novel power systems are tracked, and complexity and dynamic changes of knowledge in the field of the novel power systems can be dealt with more efficiently.

Description

Technical Field

[0001] The present invention relates to the technical field of power system knowledge tracing, and in particular to a novel power system knowledge tracing method based on co-occurrence network and deep learning. Background Art

[0002] Knowledge tracking technology, as a technology that predicts future behavior by analyzing historical behavior data, has been widely studied and applied in the field of intelligent education. Knowledge tracking technology tracks students' knowledge mastery through their historical learning records, and then predicts their future grades in order to provide personalized guidance services. By introducing its concept into the new power system, we can track the future output changes of keywords by analyzing and processing a large amount of relevant literature in the field of new power systems, find hot spots and trends in the field of new power systems, and predict the development direction of future technologies and markets.

[0003] Faced with an increasing amount of literature and a complex and ever-changing technical environment, traditional knowledge tracking methods are unable to cope with the complexity of new power systems. Applying knowledge tracking technology to new power systems still faces many challenges. How to build accurate and reliable prediction models, how to effectively collect and process data, and how to integrate it into existing power system management are all issues that need in-depth research.

[0004] Therefore, a new power system knowledge tracing method based on co-occurrence networks and deep learning is needed to efficiently and intelligently track and manage knowledge in the power system field. Summary of the invention

[0005] The purpose of this invention is to propose a new power system knowledge tracing method based on co-occurrence network and deep learning, which includes the following steps:

[0006] Step 1: Collect literature data on new power systems and conduct feature analysis and keyword selection;

[0007] Step 2: Use Python's wordcloud library to draw a word cloud diagram and analyze the research hotspots and development trends of each year; use Python's networkx library to draw a keyword co-occurrence network and analyze the correlation between keywords;

[0008] Step 3: Analyze the network structure of GRU, set GRU model parameters and determine the evaluation indicators of GRU model;

[0009] Step 4: Use the GRU network to predict the hot words of the new power system.

[0010] The new power system literature data in step 1 includes literature title, link, author, publication date, year, keywords, abstract, literature source, literature type, downloads and citations.

[0011] The evaluation indicators of the co-occurrence network in step 2 include node degree centrality, node closeness centrality and node betweenness centrality.

[0012] The evaluation indicator of the GRU model in step 3 is the mean absolute error.

[0013] The GRU model parameters in step 3 are as follows:

[0014] The hidden layer dimension is 32, the maximum number of iterations is 200, the learning rate is 0.005, the drop rate of neuron weight connections is 0.1, and the future prediction data length and observation window length are 4.

[0015] The beneficial effects of the present invention are:

[0016] The present invention tracks the research hotspots and development trends in the field of new power systems. It can not only point out the hot words of each year and the words with a high correlation with other keywords, but also predict the ten hot words that may appear in 2025, which can more efficiently cope with the complexity and dynamic changes of knowledge in the field of new power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a novel power system knowledge tracking method based on co-occurrence network and deep learning of the present invention;

[0018] Figure 2 (a)(b)(c)(d) are word cloud diagrams for 2021, 2022, 2023, and 2024 respectively;

[0019] Figure 3 The generated keyword co-occurrence network graph;

[0020] Figure 4 It is the basic structure diagram of the GRU model of the present invention;

[0021] Figure 5 This is the keyword prediction result chart for 2025. DETAILED DESCRIPTION

[0022] The present invention proposes a novel power system knowledge tracking method based on co-occurrence network and deep learning. The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0023] Figure 1 The flowchart of the novel power system knowledge tracking method based on co-occurrence network and deep learning of the present invention is as follows:

[0024] 1. Collection and processing of literature data on new power systems

[0025] First, conduct feature analysis and keyword selection, analyzing the word frequency, citation and download volume, and ranking features.

[0026] Secondly, we use web crawlers to crawl data through the sub-website of HowNet - HowNet Space. The data set we crawled is the literature data of new power systems from 2021 to 2024. We searched with the above ten keywords to capture the title, link, author, publication date, year, keywords, abstract, document source, document type, download volume and citation volume of the document, and exported it in Excel format to facilitate subsequent data cleaning.

[0027] Finally, data processing is performed. The data format is sorted, all keywords are counted and irrelevant words are removed, and the data is clustered and similar words are merged. The word frequency is divided into high frequency, medium frequency, and low frequency, and the processed data is saved.

[0028] 2. Word Cloud and Keyword Co-occurrence Network

[0029] First, read the preprocessed data and sort all the keywords by frequency. Then filter by year and divide the keywords into 2021, 2022, 2023, 2024, etc.

[0030] The parameters for generating word cloud images in wordcloud are set, where the maximum number of words displayed in the word cloud is 100, and the word cloud shape mask is the default rectangle. Finally, a 1000x7000 word cloud image is generated, and the word cloud images of the four years are spliced ​​together to form a large word cloud for easy viewing, such as Figure 2 As shown in (a)(b)(c)(d).

[0031] The word cloud chart intuitively shows the research hotspots and overall development trends of each year. In the field of new power systems in 2021, researchers focused on carbon neutrality, renewable energy, virtual power plants, new energy, etc. In 2022, in addition to carbon neutrality, new energy, and renewable energy, which are still hot, the popularity of new energy vehicles, low carbon, demand response, and power markets has gradually increased. In 2023, hot keywords will further increase, such as pumped storage power stations, low-carbon economy, carbon emissions, low-carbon transformation, etc. By 2024, the number of newly emerging hot words has decreased, but carbon neutrality, new energy, renewable energy, green and low-carbon, new energy vehicles, demand response, power markets, etc. that have appeared before still maintain a high level of popularity.

[0032] By using the networkx library in Python, which is dedicated to building complex networks, we draw a keyword co-occurrence network based on the edge list and adjust the thickness of the lines between nodes based on the size of the node degree. Figure 3 shown.

[0033] The evaluation indicators used in the co-occurrence network of this embodiment are node degree centrality, node closeness centrality and node betweenness centrality. The following will introduce the three evaluation indicators respectively and display the results in combination with data.

[0034] (1) Node degree centrality

[0035] Node degree centrality is used to measure the centrality of a node in a network. The higher the centrality of a node, the more important the node is in the network. The calculation formula is as follows:

[0036]

[0037] In the formula, k i represents the number of existing edges connected to node i, and N-1 represents the number of edges connecting node i to other nodes.

[0038] (2) Node proximity centrality

[0039] Node proximity centrality refers to the reciprocal of the average shortest path length from a node to all other nodes in the network. The higher the proximity centrality of a node, the closer its connection with other nodes is. The calculation formula is as follows:

[0040]

[0041] Where d(i,j) represents the distance from node i to node j.

[0042] (3) Node betweenness centrality

[0043] The node betweenness centrality coefficient is the number of shortest paths through a node in a network. Nodes with high node betweenness centrality are usually key bridge nodes and play an important role in the network. The calculation formula is:

[0044]

[0045] Where n st represents the number of paths that pass through node i and are the shortest paths, g st Represents the number of shortest paths connecting s and t. Through calculation, the degree centrality, closeness centrality and betweenness centrality of each keyword node in the field of new power system are obtained, and they are sorted and the top 10 nodes are output, as shown in Table 1.

[0046] Table 1 Degree centrality, closeness centrality, and betweenness centrality of the top 10 keywords

[0047]

[0048] Combining the keyword co-occurrence network and degree centrality, closeness centrality and betweenness centrality, it can be found that the ten keywords of carbon neutrality, new energy, renewable energy, carbon peak, carbon emissions, "dual carbon" goal, green and low-carbon, new energy vehicles, low-carbon economy, and low-carbon occupy an important position in the network.

[0049] 3. Analyze the network structure of the Gate Recurrent Unit (GRU)

[0050] The GRU model consists of an input layer, an output layer, and a hidden layer. Its basic structure is as follows: Figure 4 When the model is trained, the input data is first loaded into the input layer, and then the input data features are extracted through the hidden layer. Finally, the output layer outputs the data that has passed the fully connected layer according to the state of the hidden layer. When processing time series data, the GRU model can efficiently capture and utilize the dependencies between different time steps. The new power system literature data used in this embodiment is arranged year by year, which is very suitable for predictive analysis using the GRU model.

[0051] GRU model parameter settings and evaluation indicators:

[0052] For the GRU model used in this embodiment, the default hidden layer dimension is set to 32, the maximum number of iterations is 200, the learning rate is 0.005, the discard rate of the neuron weight connection is 0.1, and the future prediction data length and observation window length are 4. In addition, the data set is divided into two parts: a training set and a test set. The training set is used for model training and parameter adjustment, and the test set is used to evaluate the performance of the model. In this embodiment, the model will sequentially divide all the data into four and six, with 60% as a training set and 40% as a test set.

[0053] The value of loss is used to determine whether the model has converged. When the value of loss gradually decreases and shows a downward trend, it indicates that the model has converged. The error between the predicted value and the true value can be evaluated using the loss function. In this embodiment, the mean absolute error (MAE) is used. The mean absolute error (MAE) is a commonly used loss function for evaluating the prediction accuracy of a regression model. It can measure the average error between the predicted value and the true value. The specific formula is as follows:

[0054]

[0055] Where y i is the actual value, is the predicted value, and n is the number of samples. In order to make the program converge faster when running, it is necessary to limit the processed data to a certain range, that is, normalize the data in the optimizer. In the model, we use the MinMaxScaler function to normalize the data so that the data value is between 0 and 1. The specific formula for normalization is as follows:

[0056]

[0057] 4. Use the Gated Recurrent Unit (GRU) network in deep learning to predict hot words for new power systems

[0058] Epoch is the number of times the model runs on the training set. When there are too few epochs, the model will not update the weights sufficiently. However, when there are too many epochs, the weights will be updated too many times, causing the model to overfit.

[0059] Another important factor that affects the prediction effect of the model is the size of the data set. When the data set is too small, the network cannot extract enough features, resulting in poor training results and inaccurate predictions. When the data set is too large, it takes a long time to train, and redundancy and noise will appear in the data. Therefore, it is crucial to choose a data set of a reasonable size.

[0060] Through experiments with different epoch values ​​and different data sets, it was finally determined that when the epoch is 200, the model with a lower error rate can be obtained by training the model with the full data set. In addition, the default hidden layer dimension is set to 32, the learning rate is 0.005, and the drop rate of the neuron weight connection is 0.1. For parameters such as batch size, future prediction data length, and observation window length, fine-tuning is required according to the size of the input data set.

[0061] In order to predict the hot words of the new power system in 2025, this embodiment inputs all the literature data from 2021 to 2024 into the model to train the model, and then predicts the keywords of 2025 based on the hot words in 2024. The prediction results are as follows: Figure 5 shown.

[0062] By comparing with the serial number table, we can find that the hot words that may appear in 2025 are: carbon neutrality, renewable energy, carbon peak, new energy, new energy vehicles, green and low-carbon, electricity market, low carbon, low-carbon economy, and demand response.

[0063] The present invention can not only point out the hot words of each year and the words with high correlation with other keywords, but also predict the ten hot words that may appear in 2025, which can more efficiently cope with the complexity and dynamic changes of knowledge in the field of new power systems.

Claims

1. A new power system knowledge tracing method based on co-occurrence network and deep learning, characterized by: The steps include: Step 1: Collect literature data on new power systems and conduct feature analysis and keyword selection; Step 2: Use Python's wordcloud library to draw a word cloud diagram and analyze the research hotspots and development trends of each year; use Python's networkx library to draw a keyword co-occurrence network and analyze the correlation between keywords; Step 3: Analyze the network structure of GRU, set GRU model parameters and determine the evaluation indicators of GRU model; Step 4: Use the GRU network to predict the hot words of the new power system.

2. According to claim 1, the novel power system knowledge tracking method based on co-occurrence network and deep learning is characterized in that: The new power system literature data in step 1 includes the literature title, link, author, publication date, year, keywords, abstract, literature source, literature type, download volume and citation volume.

3. According to claim 1, the novel power system knowledge tracking method based on co-occurrence network and deep learning is characterized in that: The evaluation indicators of the co-occurrence network in step 2 include node degree centrality, node closeness centrality and node betweenness centrality.

4. According to claim 1, the novel power system knowledge tracking method based on co-occurrence network and deep learning is characterized in that: The evaluation index of the GRU model in step 3 is the mean absolute error.

5. The novel power system knowledge tracing method based on co-occurrence network and deep learning according to claim 1 or 4 is characterized in that: The GRU model parameters in step 3 are as follows: The hidden layer dimension is 32, the maximum number of iterations is 200, the learning rate is 0.005, the drop rate of neuron weight connections is 0.1, and the future prediction data length and observation window length are 4.

Images