Power grid key technology prediction method based on Fisher-Pry model

Abstract

The invention discloses a Fisher-Pry model-based power grid key technology prediction method. The method comprises the steps of constructing an original corpus; lexical item representations are generated; performing topic mining on the lexical item representation to obtain a plurality of topics and corresponding keyword sets; labeling a keyword set of the theme, determining a theme label and establishing mapping between the theme and a technical name; summarizing according to time granularity to generate time sequence index data of each theme; performing standardization processing on the time sequence index data to obtain a time sequence for fitting; taking the time sequence as input, fitting each theme through a Fisher-Pry model, and obtaining fitting parameters and fitting curves of each theme; dividing life cycle stages for each theme based on the fitted curve and identifying key time nodes; and outputting technical permeability prediction and staged research and development layout suggestions of a plurality of periods in the future based on a fitting result and stage division, thereby providing a quantitative basis for power grid technology strategic decision making.

Description

Technical Field

[0001] This invention belongs to the field of data mining technology, specifically relating to a method for predicting key power grid technologies based on the Fisher-Pry model. Background Technology

[0002] With the rapid development of new energy integration, the energy internet, and power electronics technology, the types and evolution speed of key power grid technologies are becoming increasingly diversified and accelerated. Timely and objective identification and development trend prediction of key power grid technologies have become essential for power companies, research institutions, and decision-making departments in their R&D planning, resource allocation, and risk control. Traditional technology forecasting methods often rely on expert experience, qualitative assessments, or simple statistical extrapolation of explicit indicators. These methods struggle to handle the mining and quantitative lifecycle analysis of large-scale, multi-source scientific and technological text data, exhibiting problems such as strong subjectivity, poor reproducibility, and insufficient early identification capabilities.

[0003] In recent years, research on technology trend analysis based on text mining has gradually increased. Among them, Chinese patent CN118349589A discloses a wind power technology patent analysis system based on LDA. This system includes a patent data acquisition and preprocessing module, an LDA topic modeling module, a hierarchical clustering-based topic clustering module, a time series-based trend analysis module, and a results visualization and interaction module. The system provides a relatively complete engineering implementation path in terms of large-scale patent text acquisition, TF-IDF feature construction, LDA topic extraction, topic clustering, and visualization. It also uses linear regression to fit the topic popularity time series to identify trends, facilitating interactive viewing of topic evolution by users.

[0004] However, existing technologies use linear regression to fit the time series of topic weights and extrapolate short-term trends. While linear regression models are suitable for linear growth / decline scenarios, they are difficult to characterize the life cycle (emergence-growth-maturity-decline) characteristics of typical technology diffusion and cannot identify substitution inflection points and saturation levels. Therefore, their predictive ability for long-term technology evolution and substitution processes is limited.

[0005] Currently, forecasting key power grid technologies mainly relies on expert judgment, qualitative analysis, or statistical trend extrapolation. For example, while forecasting methods based on expert interviews or the Delphi method can reflect domain knowledge, they suffer from strong subjectivity and long update cycles. Trend analysis methods based on changes in the number of patents or papers, while possessing a certain degree of objectivity, struggle to reveal potential connections between technological themes and to identify emerging technologies at an early stage. Furthermore, existing methods generally lack systematic modeling of the technological evolution process, failing to characterize the different stages of the technology lifecycle from a temporal perspective.

[0006] With the development of natural language processing and data mining technologies, text-based topic mining methods have been widely used in scientific and technological literature analysis and technology trend research. By modeling the topics of a large number of documents, technological hotspots, evolutionary directions, and potential connections can be identified from large-scale corpora. However, existing research mostly remains at the level of topic discovery or hotspot analysis, failing to integrate with technology lifecycle models and lacking quantitative analysis of technology development stages and prediction of future trends.

[0007] Therefore, current technologies lack a systematic approach capable of automatically extracting technological themes, identifying their lifecycle stages, and predicting trends from large-scale power grid literature data. This method should be able to achieve objective quantitative analysis and development stage identification of key power grid technologies without relying on prior expert knowledge, providing data support for technology planning, research investment, and policy formulation. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a method for predicting key technologies of power grids based on the Fisher-Pry model.

[0009] The objective of this invention can be achieved through the following technical solutions: This invention provides a method for predicting key power grid technologies based on the Fisher-Pry model, comprising the following steps: Acquire relevant literature data on key power grid technologies to form an original corpus; Preprocess the original corpus to generate term representations; Based on topic modeling, topic mining is performed according to term representations to obtain topic, keyword set, and topic-document distribution of documents; The keyword set is labeled, topic tags are determined, and a mapping between topics and technology names is established; Summarize topic-document distribution and topic tags by time granularity, generate time series index data for each topic, and perform standardization processing to obtain time series for fitting; Using time series data as input, the Fisher-Pry model is used to fit each topic, and the fitting parameters and fitting curves for each topic are obtained. Based on the fitted curve, the technology lifecycle stages are divided and key time nodes are identified; The development prediction results of key power grid technologies are generated based on fitting parameters, life cycle division, and key time nodes.

[0010] Furthermore, the relevant literature data on key power grid technologies includes papers, patent documents, technical reports, and standard documents in the field of power grids.

[0011] Furthermore, the preprocessing of the original corpus to generate term representations specifically includes: Each original document in the original corpus is segmented into words, and a custom dictionary for the power grid field is used to identify technical terms and multi-word phrases. For the identified technical terms and multi-word phrases, stop words, punctuation marks, and meaningless terms are removed, and synonyms and word forms are uniformly merged to obtain a standardized term set.

[0012] Furthermore, the topic-based model performs topic mining based on term representations to obtain topics, keyword sets, and the topic-document distribution of documents, specifically including: Establish an LDA generation model and set hyperparameters. and Its generation process is as follows: For the first d Original documents, topic distribution For the first k Each topic, word distribution ; regarding the first in the original document n Theme assignment for each word And generate terms from the assigned topic. ; in, Indicates the first d The topic distribution vector of the original document; Indicates parameters For the Dirichlet distribution with hyperparameters, Indicates the first k Word distribution vectors for each topic Indicates parameters For the Dirichlet distribution with hyperparameters, k For topic index, the value range is: , For the set number of topics, Indicates the first d The first original document n Thematic assignment of each word Represented by vector The multinomial distribution of the parameter means that, according to The probability from K One topic is selected from the existing topics. n The ordinal number of the word in the original document, with a value range of 100. , For the first d Total word count of the original document; Indicates the first d The first original document n The generated terms, Indication and topic assignment The corresponding keyword distribution vector, Indicates The multinomial distribution with parameters is used to generate terms from the vocabulary according to probability; The LDA model is parameter estimated using Gibbs sampling or variational Bayes methods to obtain the topic-word distribution. Distribution of original documents by topic The point estimate can be expressed as: in, The estimated first number is... k Keywords under each topic w The conditional probability, i.e., the topic-word distribution component; This indicates that it has been assigned as the subject in all the original documents. k terms w The count; Indicates the topic k The sum of the counts of all terms under; Indicates the size of the vocabulary, representing the total number of different terms; The estimated number of digits is 1. d The original document corresponds to the first k The probability value of each topic, i.e., the original document-topic distribution component; Indicates the first d Assigned as subject in the original document k Word count; Indicates the first d Total word count in the original document; Based on the estimated topic k according to Values ​​extracted in descending order M A set of thematic keywords is formed from these keywords, and the topic distribution vector corresponding to each original document is generated. This corresponds to the topic-document distribution.

[0013] Furthermore, the step of tagging the keyword set, determining topic tags, and establishing a mapping between topics and technology names specifically includes: Extract the keyword set for each topic and compare it with a pre-built candidate library of technical names; Two metrics, semantic similarity-based automatic matching and keyword overlap-based matching, are used to match topics with candidate technologies and calculate matching scores. When the matching score of a candidate technology is greater than or equal to the preset automatic assignment threshold, the candidate technology is assigned as the technology name of the topic. When the matching score is lower than the preset automatic assignment threshold or when there are multiple candidate technologies with scores that are too close to be distinguished, a manual review process is triggered, in which several domain experts review the candidate technology names and vote to determine the final label.

[0014] Furthermore, the step of summarizing topic-document distribution and topic tags by time granularity to generate time-series indicator data for each topic specifically includes: The original corpus is divided into non-overlapping time interval sets according to a preset time granularity. The time granularity is annual or quarterly; In the t Within a given time interval, the collection of original documents falling within that time interval is statistically analyzed. This indicates the total number of original documents within that time interval; Based on topic-document distribution Thematic temporal intensity is calculated by summarizing time intervals: in, Indicates the first k The topic in t Time-series indicators for each time interval reflect the activity level of the topic within that time interval; Indicates the first d The first original document k Thematic distribution values ​​for each theme; Indicates the first t The collection of original documents within a time interval; Time series indicators for each topic in each time interval Arranged chronologically, the time-series index data for each topic are obtained. ,in T This indicates the total number of time intervals, used to reflect the development trend of each theme within different time intervals.

[0015] Furthermore, the standardization process specifically includes: For the first k Time series index data sequences of each theme Perform normalization processing and calculate standardized time series. The formula is: in, Indicates the first k The topic in t Standardized index values ​​for each time interval. , These represent the minimum and maximum values ​​of the topic across all time intervals, respectively. Obtain the standardized time series .

[0016] Furthermore, the process of using time series data as input and fitting each topic to a Fisher-Pry model to obtain fitting parameters and curves for each topic specifically includes: For the first k Time series of each theme As input, the Fisher-Pry model is used for fitting, and the fitting formula is: in, Indicates the first k The topic in t Fitted values ​​for the time interval; For the first k The growth rate parameter for each topic For the first k The turning point of each theme, i.e., the time of the midpoint of the fitted curve; Using nonlinear least squares method and Perform parameter estimation to fit the curve. Minimize the input time series The error; After fitting, the Fisher-Pry fitting curve and corresponding fitting parameters for each topic are obtained. and .

[0017] Furthermore, the step of dividing the technology lifecycle stages based on the fitted curve and identifying key time nodes specifically includes: The first k Fisher-Pry fitting curves for each topic It is divided into several life cycle stages, including the budding stage, growth stage, maturity stage, and decline stage; Key nodes are calculated using the first and second derivatives of the fitted curve: in, Indicates the first k The topic in t The cumulative rate of development over time Indicates the first k The topic in t The rate of change over time is used to identify inflection points; Inflection point time Corresponding to acceleration The location where the maximum / minimum value is reached may mark a key turning point in the stage of technological development. Define the key time point: the end of the budding stage. This corresponds to the fitted value reaching the 10% level; the end of the growth period is the inflection point. That is, the fitted value reaches 50%; the end of the maturity period. The corresponding fitted value reached 90%; the decline period started from... Until the curve saturates or declines; Based on the fitted curves and derivative analysis results, the time intervals for each stage are divided as follows: Germination period: from the start time to ; Growth period: From to ; Maturity: From to ; Recession period: From To the end of the time series or at a point in time when the downward trend is obvious; Output the lifecycle stage division results for each topic and the corresponding list of key time nodes.

[0018] Furthermore, the generation of development prediction results for key power grid technologies based on fitting parameters, lifecycle division, and key time nodes specifically includes: Based on the Fisher-Pry fitting curves and fitting parameters, combined with the life cycle stages and key time nodes of each theme, the technological development trends in each future time interval are predicted. Combining the life cycle division results of each theme, the future development trends of the predicted curves are analyzed. The predicted technological development values, life cycle stages and corresponding key time nodes of each theme in each future time interval are output, providing a reference for power grid key technology planning, investment and R&D decisions.

[0019] Compared with the prior art, the present invention has the following advantages: (1) In the prior art, the traditional prediction of key power grid technologies mainly relies on expert experience or statistical extrapolation, which has the problems of strong subjectivity and poor reproducibility, making it difficult to achieve objective technology identification and trend analysis in the context of large-scale text data. This invention uses a text preprocessing and topic mining mechanism based on natural language processing to automatically extract topic distribution and keyword sets from large-scale power grid literature using the LDA topic model, thereby achieving objective identification and quantitative expression of technical topics, reducing reliance on expert experience, and significantly improving the automation and repeatability of the technology identification process.

[0020] (2) Existing methods rely solely on changes in the number of patents or papers for trend analysis, which makes it difficult to reveal the potential connections and structured relationships between technical topics. This invention establishes a correspondence between topics and technical names based on the topic modeling results by constructing a topic-document distribution matrix and a topic tag mapping mechanism. This enables semantic classification at the topic level and a structured description of the technical system, revealing the connections between different technical topics and providing fundamental support for the systematic analysis of key power grid technologies.

[0021] (3) In the prior art, as shown in Chinese patent CN118349589A, linear regression is used to predict the trend of topic popularity time series. It can only depict linear growth or decline trends and cannot reflect the S-shaped evolution law of the technology life cycle, making it difficult to identify the budding, maturity and decline stages of technology. The present invention uses a time series fitting method based on the Fisher-Pry model to perform nonlinear fitting on standardized topic time series indicators, which can accurately describe the S-shaped growth process of technology diffusion, realize the quantitative division of the life cycle of each topic and the identification of key inflection points, thereby effectively solving the problem that linear models are difficult to capture the dynamic changes of the life cycle.

[0022] (4) In existing research, thematic trend analysis mostly focuses on identifying hotspots, lacking quantitative identification of life cycle stages and modeling of evolution paths. This invention extracts growth rate and acceleration characteristics by performing derivative analysis on Fisher-Pry fitting curves, automatically dividing the technology into nascent, growth, maturity, and decline stages, and identifying corresponding key time nodes. This gives the technology evolution stages clear quantitative boundaries, thus providing verifiable data for subsequent technology substitution analysis and strategic planning.

[0023] (5) Existing methods often fail to integrate topic modeling results with quantitative prediction models, making it difficult to apply trend analysis results to practical technology planning. This invention achieves a complete analysis link from text topic discovery to life cycle prediction by organically integrating topic mining results with the Fisher-Pry evolutionary model. This fusion method not only reflects the current level of attention to technology topics, but also predicts future evolutionary trends based on fitting parameters, providing a scientific basis for the forward-looking layout of key power grid technologies. Attached Figure Description

[0024] Figure 1 This is a flowchart of the power grid key technology prediction method according to an embodiment of the present invention; Figure 2 This is a model diagram of a power grid key technology prediction system according to an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] Example 1: This embodiment provides a method for predicting key power grid technologies based on the Fisher-Pry model, such as... Figure 1 As shown, it includes the following steps: Step S1: Obtain literature data related to key technologies of the power grid in the field of power grid to form an original corpus; Literature data related to key technologies in the power grid field includes papers, patent documents, technical reports, and standard documents in the power grid field.

[0027] Step S2: Perform text preprocessing on the original corpus to generate term representations, specifically including: Each original document in the original corpus is segmented into words, and a custom dictionary for the power grid field is used to identify technical terms and multi-word phrases. For the identified technical terms and multi-word phrases, stop words, punctuation marks, and meaningless terms are removed, and synonyms and word forms are uniformly merged to obtain a standardized term set.

[0028] Step S3: Based on the topic model, perform topic mining on the term representations to obtain several topics and their corresponding keyword sets, and obtain the topic-document distribution for each document, specifically including: Establish an LDA generation model and set hyperparameters. and Its generation process is as follows: For the first d Original documents, topic distribution For the first k Each topic, word distribution ; regarding the first in the original document n Theme assignment for each word And generate terms from the assigned topic. ; in, Indicates the first d The topic distribution vector of the original document; Indicates parameters For the Dirichlet distribution with hyperparameters, Indicates the first k Word distribution vectors for each topic Indicates parameters For the Dirichlet distribution with hyperparameters, k For topic index, the value range is: , For the set number of topics, Indicates the first d The first original document n Thematic assignment of each word Represented by vector The multinomial distribution of the parameter means that, according to The probability from K One topic is selected from the existing topics. n The ordinal number of the word in the original document, with a value range of 100. , For the first d Total word count of the original document; Indicates the first d The first original document n The generated terms, Indication and topic assignment The corresponding keyword distribution vector, Indicates The multinomial distribution with parameters is used to generate terms from the vocabulary according to probability; The LDA model is parameter estimated using Gibbs sampling or variational Bayesian methods to obtain the topic-word distribution. Distribution of original documents by topic The point estimate can be expressed as: in, The estimated number of digits is 1. k Keywords under each topic w The conditional probability, i.e., the topic-word distribution component; This indicates that it has been assigned as the subject in all the original documents. k terms w The count; Indicates the topic k The sum of the counts of all terms under; Indicates the size of the vocabulary, representing the total number of different terms; The estimated number of digits is 1. d The original document corresponds to the first k The probability value of each topic, i.e., the original document-topic distribution component; Indicates the first d Assigned as subject in the original document k Word count; Indicates the first d Total word count in the original document; Based on the estimated topick according to Values ​​extracted in descending order M A set of thematic keywords is formed from these keywords, and the topic distribution vector corresponding to each original document is generated. This corresponds to the topic-document distribution.

[0029] Step S3 maps the sparse, noisy text information at the document-word level into two probability distributions: topic-word and document-topic. This transforms unstructured text into a statistically quantifiable, aggregateable, and time-series-sequential structured signal. Specifically, LDA uses a generative probabilistic model that assumes each original document is generated by a probabilistic mixture of several latent topics. Each theme is further divided into one word. Terms are generated from the vocabulary list according to probability; by analyzing each word in the text... Topic assignment Statistical inference (using Gibbs sampling or variational Bayesian methods) involves the model learning the probability structure of which words within a document belong to the same latent topic and which words in the entire corpus collectively constitute a topic. The point estimation in the formula reveals two important implications: firstly, it involves word counting... The first step is to convert empirical frequencies into probability estimates; the second is to use hyperparameters. Weighted smoothing avoids the zero-probability problem and controls the sparsity of topic and word distribution, thereby affecting the fine-grained or coarse-grained division of topics.

[0030] Step S3 automatically discovers the implicit technical topics in the corpus and provides an interpretable set of keywords for each topic, thereby abstracting massive amounts of unstructured text into a small number of meaningful technical units; it outputs a document-topic distribution vector for each original document, which can be directly used to construct a topic time series signal by accumulating at the time granularity, realizing the quantification of technical popularity and penetration rate in the time dimension; it improves noise resistance through probabilistic modeling and smoothing terms (more resistant to the influence of accidental high-frequency words in a single document than simple word frequency or TF-IDF), thereby improving the stability and reliability of subsequent life cycle fitting based on time series.

[0031] Step S4: Annotate the keyword set of the topic, determine the topic tags, and establish a mapping between the topic and the technology name, specifically including: Extract the keyword set for each topic and compare it with a pre-built candidate library of technical names; Two metrics, semantic similarity-based automatic matching and keyword overlap-based matching, are used to match topics with candidate technologies and calculate matching scores. When the matching score of a candidate technology is greater than or equal to the preset automatic assignment threshold, the candidate technology is assigned as the technology name of the topic. When the matching score is lower than the preset automatic assignment threshold or when there are multiple candidate technologies with scores that are too close to be distinguished, a manual review process is triggered, in which several domain experts review the candidate technology names and vote to determine the final label.

[0032] Step S5: Generate time-series indicator data for each topic based on topic-document distribution and topic tag aggregation at the time granularity, specifically including: The original corpus is divided into non-overlapping time interval sets according to a preset time granularity. The time granularity is in the form of a year or a quarter; In the t Within a given time interval, the collection of original documents falling within that time interval is statistically analyzed. This indicates the total number of original documents within that time interval; Based on topic-document distribution Thematic temporal intensity is calculated by summarizing time intervals: in, Indicates the first k The topic in t Time-series indicators for each time interval reflect the activity level of the topic within that time interval; Indicates the first d The first original document k Thematic distribution values ​​for each theme; Indicates the first t The collection of original documents within a time interval; Time series indicators for each topic in each time interval Arranged chronologically, the time-series index data for each topic are obtained. ,in T This indicates the total number of time intervals, used to reflect the development trend of each theme within different time intervals.

[0033] Step S6: Standardize the time series index data to obtain the time series for fitting, specifically including: For the first k Time series index data sequences of each theme Perform normalization processing and calculate standardized time series. The formula is: in, Indicates the first k The topic in t Standardized index values ​​for each time interval. , These represent the minimum and maximum values ​​of the topic across all time intervals, respectively. Obtain the standardized time series .

[0034] Step S7: Using the time series as input, fit each topic to the Fisher-Pry model to obtain the fitting parameters and fitting curves for each topic, specifically including: For the first k Time series of each theme As input, the Fisher-Pry model is used for fitting, and the fitting formula is: in, Indicates the first k The topic in t Fitted values ​​for the time interval; For the first k The growth rate parameter for each topic For the first k The turning point of each theme, i.e., the time of the midpoint of the fitted curve; Using nonlinear least squares method and Perform parameter estimation to fit the curve. Minimize the input time series The error; After fitting, the Fisher-Pry fitting curve and corresponding fitting parameters for each topic are obtained. and .

[0035] Step S8: Based on the fitted curves, divide each topic into lifecycle stages and identify key time nodes, specifically including: The first k Fisher-Pry fitting curves for each topic It is divided into several life cycle stages, including the budding stage, growth stage, maturity stage, and decline stage; Key nodes are calculated using the first and second derivatives of the fitted curve: in, Indicates the first k The topic in t The cumulative rate of development over time Indicates the first k The topic in t The rate of change over time is used to identify inflection points; Inflection point time Corresponding to acceleration The location where the maximum / minimum value is reached may mark a key turning point in the stage of technological development. Define the key time point: the end of the budding stage. This corresponds to the fitted value reaching the 10% level; the end of the growth period is the inflection point. That is, the fitted value reaches 50%; the end of the maturity period. The corresponding fitted value reached 90%; the decline period started from... Until the curve saturates or declines; Based on the fitted curves and derivative analysis results, the time intervals for each stage are divided as follows: Germination period: from the start time to ; Growth period: From to ; Maturity: From to ; Recession period: From To the end of the time series or at a point in time when the downward trend is obvious; Output the lifecycle stage division results for each topic and the corresponding list of key time nodes.

[0036] Step S9: Based on fitting parameters, lifecycle segmentation, and key time node generation technology development prediction, specifically including: Based on the Fisher-Pry fitting curves and fitting parameters, combined with the life cycle stages and key time nodes of each theme, the technological development trends in each future time interval are predicted. Combining the life cycle division results of each theme, the future development trends of the predicted curves are analyzed. The predicted technological development values, life cycle stages and corresponding key time nodes of each theme in each future time interval are output, providing a reference for power grid key technology planning, investment and R&D decisions.

[0037] Example 2: This embodiment provides a power grid key technology prediction system based on the Fisher-Pry model, such as... Figure 2 As shown, it includes: The data acquisition module is used to batch retrieve original documents and metadata from multiple sources such as patent databases, academic paper databases, standards / policy databases, corporate white papers, and conference materials according to preset search strategies; it supports searching by keywords, classification numbers, time windows, and other conditions, and outputs fields such as document identifier, title, abstract, claims, application / publication date, applicant, number of citations, and original full text.

[0038] The metadata verification and storage module is used to deduplicate, verify the format, standardize the timestamps, and persistently store the collected raw data. It adopts a hybrid architecture of relational database and large-scale file / object storage to save structured metadata and raw text, and establishes a retrieval view by time index to provide callable dataset identifiers to downstream modules.

[0039] The text preprocessing module is used to clean and normalize the original document text, including removing noisy text, sentence and word segmentation, identifying professional terms and multi-word phrases based on a custom dictionary for the power grid field, removing stop words, merging synonyms / lexical forms and constructing n-grams. This module outputs a normalized term set and term frequency records for each original document for subsequent feature construction.

[0040] The feature building module is used to generate text feature matrices based on word frequency and corpus statistics. It supports common text featureization methods and outputs the results as a document-term sparse matrix for use by topic models.

[0041] The topic mining module (LDA module) is used to train and infer the document-term matrix based on LDA or other topic models; it supports multiple estimation methods, hyperparameter search, and automated topic number selection strategies; the module outputs topic-term distribution and original document-topic distribution for subsequent analysis.

[0042] The topic keyword and automatic tagging module is used to extract several representative keywords for each topic according to the topic word distribution and perform automatic matching based on a pre-built candidate technology name library; when the automatic matching result is uncertain, a manual review process is triggered; this module outputs a topic-technology mapping table and its confidence metadata.

[0043] The time series index generation module is used to calculate the time series index for each topic based on the original literature-topic distribution and topic tag summary at a preset time granularity (such as year / quarter). The module supports assigning configurable weights to literature from different sources to reflect quality or influence, and writes the time series results into the time series database.

[0044] The time series preprocessing and standardization module is used to smooth and normalize the time series indicators by topic to meet the requirements of subsequent model input, and provides strategies for anomaly detection and pruning of contributions from single subjects to improve the robustness of time series data.

[0045] The Fisher-Pry fitting module is used to perform nonlinear fitting on standardized time series for each topic, estimate fitting parameters, and evaluate goodness of fit; it supports combining resampling or other methods to evaluate the stability and uncertainty of parameters; the output is the fitting curve and parameter estimation results for each topic.

[0046] The life cycle segmentation and key node identification module is used to automatically segment the technology life cycle stages (such as the budding stage, growth stage, maturity stage, and decline stage) and identify key time nodes based on the fitted curve and derivative features; this module also outputs uncertainty information of key nodes for decision-making reference.

[0047] The prediction and strategy generation module is used to generate penetration rate prediction curves for several future periods based on Fisher-Pry fitting parameters and life cycle division, and to generate R&D layout, resource allocation and risk response suggestions based on phased rules. It also supports the configuration and export of strategy templates.

[0048] If the aforementioned functions are implemented as 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 this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0049] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting key technologies of power grids based on the Fisher-Pry model, characterized in that, Includes the following steps: Acquire relevant literature data on key power grid technologies to form an original corpus; Preprocess the original corpus to generate term representations; Based on topic modeling, topic mining is performed according to term representations to obtain topic, keyword set, and topic-document distribution of documents; The keyword set is labeled, topic tags are determined, and a mapping between topics and technology names is established; Summarize topic-document distribution and topic tags by time granularity, generate time series index data for each topic, and perform standardization processing to obtain time series for fitting; Using time series data as input, the Fisher-Pry model is used to fit each topic, and the fitting parameters and fitting curves for each topic are obtained. Based on the fitted curve, the technology lifecycle stages are divided and key time nodes are identified; The development prediction results of key power grid technologies are generated based on fitting parameters, life cycle division, and key time nodes.

2. The method for predicting key power grid technologies based on the Fisher-Pry model according to claim 1, characterized in that, The relevant literature and data on key power grid technologies include papers, patent documents, technical reports, and standard documents in the field of power grids.

3. The method for predicting key power grid technologies based on the Fisher-Pry model according to claim 1, characterized in that, The preprocessing of the original corpus to generate term representations specifically includes: Each original document in the original corpus is segmented into words, and a custom dictionary for the power grid field is used to identify technical terms and multi-word phrases. For the identified technical terms and multi-word phrases, stop words, punctuation marks, and meaningless terms are removed, and synonyms and word forms are uniformly merged to obtain a standardized term set.

4. The method for predicting key power grid technologies based on the Fisher-Pry model according to claim 1, characterized in that, The topic-based model performs topic mining based on term representations to obtain topics, keyword sets, and the topic-document distribution of documents, specifically including: Establish an LDA generation model and set hyperparameters. and Its generation process is as follows: For the first d Original documents, topic distribution For the first k Each topic, word distribution ; regarding the first in the original document n Theme assignment for each word And generate terms from the assigned topic. ; in, Indicates the first d The topic distribution vector of the original document; Indicates parameters For the Dirichlet distribution with hyperparameters, Indicates the first k Word distribution vectors for each topic Indicates parameters For the Dirichlet distribution with hyperparameters, k For topic index, the value range is: , For the set number of topics, Indicates the first d The first original document n Thematic assignment of each word Represented by vector The multinomial distribution of the parameter means that, according to The probability from K One topic is selected from the existing topics. n The ordinal number of the word in the original document, with a value range of 100. , For the first d Total word count of the original document; Indicates the first d The first original document n The generated terms, Indication and topic assignment The corresponding keyword distribution vector, Indicates The multinomial distribution with parameters is used to generate terms from the vocabulary according to probability; The LDA model is parameter estimated using Gibbs sampling or variational Bayes methods to obtain the topic-word distribution. Distribution of original documents by topic The point estimate can be expressed as: in, The estimated number of digits is 1. k Keywords under each topic w The conditional probability, i.e., the topic-word distribution component; This indicates that it has been assigned as the subject in all the original documents. k terms w The count; Indicate topic k The sum of the counts of all terms under; Indicates the size of the vocabulary, representing the total number of different terms; The estimated number of digits is 1. d The original document corresponds to the first k The probability value of each topic, i.e., the original document-topic distribution component; Indicates the first d Assigned as the subject in the original document k Word count; Indicates the first d Total word count in the original document; Based on the estimated topic k according to Values ​​extracted in descending order M A set of thematic keywords is formed from these keywords, and the topic distribution vector corresponding to each original document is generated. As the corresponding topic-document distribution.

5. The method for predicting key power grid technologies based on the Fisher-Pry model according to claim 1, characterized in that, The process of tagging the keyword set, determining topic tags, and establishing a mapping between topics and technology names specifically includes: Extract the keyword set for each topic and compare it with a pre-built candidate library of technical names; Two metrics, semantic similarity-based automatic matching and keyword overlap-based matching, are used to match topics with candidate technologies and calculate matching scores. When the matching score of a candidate technology is greater than or equal to the preset automatic assignment threshold, the candidate technology is assigned as the technology name of the topic. When the matching score is lower than the preset automatic assignment threshold or when there are multiple candidate technologies with scores that are too close to be distinguished, a manual review process is triggered, in which several domain experts review the candidate technology names and vote to determine the final label.

6. The method for predicting key power grid technologies based on the Fisher-Pry model according to claim 1, characterized in that, The process of summarizing topic-document distribution and topic tags by time granularity to generate time-series indicator data for each topic specifically includes: The original corpus is divided into non-overlapping time interval sets according to a preset time granularity. The time granularity is annual or quarterly; In the t Within a given time interval, the collection of original documents falling within that time interval is statistically analyzed. This indicates the total number of original documents within that time interval; Based on topic-document distribution Thematic temporal intensity is calculated by summarizing time intervals: in, Indicates the first k The topic in t Time-series indicators for each time interval reflect the activity level of the topic within that time interval; Indicates the first d The first original document k Thematic distribution values ​​for each theme; Indicates the first t The collection of original documents within a time interval; Time series indicators for each topic in each time interval Arranged chronologically, the time-series index data for each topic are obtained. ,in T This indicates the total number of time intervals, used to reflect the development trend of each theme within different time intervals.

7. The method for predicting key power grid technologies based on the Fisher-Pry model according to claim 1, characterized in that, The standardization process specifically includes: For the first k Time series index data sequences of each theme Perform normalization processing and calculate standardized time series. The formula is: in, Indicates the first k The topic in t Standardized index values ​​for each time interval. , These represent the minimum and maximum values ​​of the topic across all time intervals, respectively. Obtain the standardized time series .

8. The method for predicting key power grid technologies based on the Fisher-Pry model according to claim 1, characterized in that, The process of using time series data as input and fitting each topic with a Fisher-Pry model to obtain fitting parameters and curves for each topic specifically includes: For the first k Time series of each theme As input, the Fisher-Pry model is used for fitting, and the fitting formula is: in, Indicates the first k The topic in t Fitted values ​​for the time interval; For the first k The growth rate parameter for each topic For the first k The turning point of each theme, i.e., the time of the midpoint of the fitted curve; Using nonlinear least squares method and Perform parameter estimation to fit the curve. Minimize the input time series The error; After fitting, the Fisher-Pry fitting curve and corresponding fitting parameters for each topic are obtained. and .

9. The method for predicting key power grid technologies based on the Fisher-Pry model according to claim 1, characterized in that, The method of dividing the lifecycle stages based on fitted curves and identifying key time nodes specifically includes: The first k Fisher-Pry fitting curves for each topic It is divided into several life cycle stages, including the budding stage, growth stage, maturity stage, and decline stage; Key nodes are calculated using the first and second derivatives of the fitted curve: in, Indicates the first k The topic in t The cumulative rate of development over time Indicates the first k The topic in t The rate of change over time is used to identify inflection points; Inflection point time Corresponding to acceleration The location where the maximum / minimum value is reached may mark a key turning point in the stage of technological development. Define the key time point: the end of the budding stage. This corresponds to the fitted value reaching the 10% level; the end of the growth period is the inflection point. That is, the fitted value reaches 50%; the end of the maturity period. The corresponding fitted value reached 90%; the decline period started from... Until the curve saturates or declines; Based on the fitted curves and derivative analysis results, the time intervals for each stage are divided as follows: Germination period: from the start time to ; Growth period: From to ; Maturity: From to ; Recession period: From To the end of the time series or at a point in time when the downward trend is obvious; Output the lifecycle stage division results for each topic and the corresponding list of key time nodes.

10. A method for predicting key power grid technologies based on the Fisher-Pry model according to claim 1, characterized in that, The development prediction results for key power grid technologies generated based on fitting parameters, lifecycle division, and key time nodes specifically include: Based on the Fisher-Pry fitting curves and fitting parameters, combined with the life cycle stages and key time nodes of each theme, the technological development trends in each future time interval are predicted. Combining the life cycle division results of each theme, the future development trends of the predicted curves are analyzed. The predicted technological development values, life cycle stages and corresponding key time nodes of each theme in each future time interval are output, providing a reference for power grid key technology planning, investment and R&D decisions.

Images