Method for constructing natural disaster risk knowledge graph of photovoltaic project

Abstract

The invention discloses a photovoltaic project natural disaster risk knowledge graph construction method, and relates to the field of key infrastructure risk assessment, and the method comprises the steps: S1, constructing a photovoltaic project natural disaster geographic information database based on text big data; s2, extracting a natural disaster risk causal relationship of the photovoltaic project based on machine learning; and S3, drawing a photovoltaic project natural disaster risk knowledge graph based on the photovoltaic project natural disaster geographic information base and the photovoltaic project natural disaster risk causal relationship. According to the method, the limitation that an existing method depends on traditional statistical data and is based on single disaster case analysis is solved, the data bottleneck is broken through from the source, the advantages of disaster text big data are fully played, and a big sample database for photovoltaic project natural disasters is built through natural language processing and a geographic information system platform; therefore, the natural disaster knowledge graph constructed by the method has higher universality and integrity.

Description

Technical Field

[0001] The present invention relates to the field of critical infrastructure risk assessment, and in particular to a method for constructing a natural disaster risk knowledge graph for a photovoltaic project. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.

[0003] Under the dual carbon background of global response to climate change and my country's "carbon peak and carbon neutrality", photovoltaic power station projects have become one of the important clean energy solutions and have developed rapidly in recent years. However, with the frequent occurrence of extreme disaster events caused by climate change, photovoltaic projects are increasingly affected by extreme disasters such as floods, typhoons, blizzards and freezing weather. Not only does it cause serious losses to the photovoltaic projects themselves, but it also causes serious systemic cascade risks to the wider social and economic system through the network connection between the power sector and other parts and other regions. Therefore, accurately constructing a natural disaster risk knowledge graph for photovoltaic projects is of great significance for risk identification, assessment and early warning, and then ensuring the planning, construction and operation of photovoltaic projects. However, the current construction of the natural disaster knowledge graph for photovoltaic projects focuses on the impact of disaster factors on the physical system of photovoltaic projects based on traditional historical statistical data and a single typical disaster case analysis. In addition, the construction of the causal relationship of disaster risk is mostly based on the subjective literature induction or expert decision-making. This method is not only limited by the information bottleneck of statistical data, but also lacks a systematic perspective. It has limitations in constructing a more universal and objective natural disaster risk complex network, so it is not enough to draw a generalized knowledge graph of natural disaster risks for photovoltaic projects. Summary of the invention

[0004] The purpose of the present invention is to provide a method for constructing a knowledge graph of natural disaster risks for photovoltaic projects in response to the problems existing in the prior art. First, the international emergency disaster data set and network text disaster big data are used to construct a large sample database of natural disaster geographic information for photovoltaic projects through natural language processing and geographic information system modeling. Then, natural language processing and machine learning and other technologies are used to deeply mine the disaster attribute information contained in the text big data and the logical relationship between the causes and consequences of disasters at all levels. Through network analysis, the complex disaster risk system is abstracted into a visual network topology, which provides an innovative method for objectively drawing a knowledge graph of natural disaster risks for photovoltaic projects.

[0005] The technical solution of the present invention is as follows:

[0006] A method for constructing a knowledge graph of natural disaster risks for photovoltaic projects, comprising:

[0007] Step S1: constructing a natural disaster geographic information database for photovoltaic projects based on text big data;

[0008] Step S2: Extracting the causal relationship of natural disaster risks of photovoltaic projects based on machine learning;

[0009] Step S3: Based on the PV project natural disaster geographic information database and the causal relationship of PV project natural disaster risks, a knowledge map of PV project natural disaster risks is drawn.

[0010] Furthermore, the step S1 comprises:

[0011] Step S11: Big data crawling and corpus construction of natural disaster texts of photovoltaic projects;

[0012] Step S12: Disaster text big data cleaning and preprocessing;

[0013] Step S13: mining disaster attribute features based on natural language processing;

[0014] Step S14: Construction of a natural disaster geographic information database for photovoltaic projects.

[0015] Furthermore, the step S11 includes:

[0016] Step S111: Screening and obtaining a list of disaster events for photovoltaic projects based on an internationally authoritative emergency disaster dataset;

[0017] Step S112: using photovoltaic disasters as keywords, using a web crawler to automatically crawl text big data about photovoltaic disasters around the world from mainstream portals at home and abroad, social media, and official reports based on keywords;

[0018] Step S113: Integrate the photovoltaic project disaster event list and disaster text big data to build a photovoltaic project natural disaster text big data corpus.

[0019] Furthermore, the step S12 includes:

[0020] The preprocessing of the photovoltaic project natural disaster text big data corpus is completed through data cleaning, word segmentation, part-of-speech tagging, and removal of stop words to meet the format requirements of natural language processing; wherein, the noise content is removed by data cleaning, and only the title, abstract, and text information are saved; the noise content includes: repetition, advertisement, label, HTML, JS code, and comments; then, the complete sentences in the photovoltaic project natural disaster text big data corpus are decomposed into the most individual words or phrases using statistical or rule-based methods as the smallest unit of text mining processing, and then the part-of-speech tagging is completed according to the division rules; finally, the stop words that do not contribute to the text features are removed.

[0021] Furthermore, the step S13 includes:

[0022] Based on the big data semantic intelligent analysis platform and language technology platform, the spatiotemporal and non-spatiotemporal attributes of each historical disaster event are mined and acquired; the spatiotemporal and non-spatiotemporal attributes include: type, place name, time, disaster situation, disaster-causing factors, and disaster-bearing body.

[0023] Furthermore, the mining obtains the spatiotemporal and non-spatial attributes of each disaster historical event, including:

[0024] Step A: Select a classification model to identify disaster event types and disaster-causing factors;

[0025] Step B: Construct the regular expression of time entity and extract the time information in the disaster text;

[0026] Step C: Based on the global place name database GeoNames, use the language technology platform to extract location entities from the text and accurately locate the location of the disaster to the city-level administrative unit;

[0027] Step D: According to the disaster loss field, the disaster loss information is extracted through Bootstrapping weakly supervised learning;

[0028] Step E: Extract the specific disaster-affected objects according to the system to which the disaster damage belongs.

[0029] Furthermore, the step S14 includes:

[0030] The expressions of the identified event types, place name elements, time elements, disaster damage elements and non-spatiotemporal attributes are standardized to form a complete historical disaster event information table. Based on the ArcGIS geographic information system platform, the place name information in the historical disaster event information table is used to generate a point feature dataset, which is connected to the non-spatiotemporal attribute table to build a photovoltaic project natural disaster geographic information database to achieve data storage, visualization, access and application. The non-spatiotemporal attribute table includes: type, event, disaster damage and affected department.

[0031] Furthermore, the step S2 comprises:

[0032] The photovoltaic project natural disaster geographic information database is divided into a training set and a test set. First, the word2vec and ELMO word embedding models are used to represent the text sequences respectively, and a joint vector is constructed by integrating the pre-trained word vector and the ELMO dynamic word feature. Then the joint vector is input into the neural network model to extract long-distance semantic features in depth, and a weight vector is generated through the attention mechanism. The features are weighted to highlight the contribution of important lexical features. The softmax classifier is then used to classify the weighted features to complete the training of the causal relationship extraction model. Finally, the accuracy and recall rate indicators are used to evaluate the model training effect. After the model accuracy meets the standard, the disaster causal relationship model based on the neural network is used to extract the micro-causal relationship of risk factors at all levels implied in the photovoltaic project natural disaster geographic information database, including the disaster-causing factors, the disaster-affected bodies, and the disaster-causing factors and the disaster-affected bodies.

[0033] Furthermore, the step S3 comprises:

[0034] The semantic similarity between risk factors is measured by calculating the angle cosine distance, and the extracted specific risk factors are abstracted and generalized based on the semantic similarity results;

[0035] Based on the regional natural disaster system theory, from the perspectives of disaster-prone environment, disaster-causing factors, and disaster-bearing bodies, the single causal relationships containing the same risk factors are connected end to end according to the causal logic chain to form a chain of causal relationships, and then interweave and gather a relatively complete disaster knowledge map;

[0036] Based on complex network theory and combined with prior knowledge, a generalized natural disaster risk network topology of photovoltaic projects is objectively drawn to construct a knowledge map of natural disaster risks of photovoltaic projects.

[0037] Furthermore, the angle cosine distance is calculated using the following formula:

[0038]

[0039] Among them: |W|, |K| are the total frequencies of the two co-occurring keywords respectively, W∩K is the co-occurrence frequency of the two keywords, and the value range of Cosine(W,K) is [0,1]. The larger the value, the stronger the keyword correlation.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. A method for constructing a knowledge graph of natural disaster risks for photovoltaic projects, which solves the limitations of existing methods that rely on traditional statistical data and are based on single disaster case analysis: This method breaks through the data bottleneck from the source, gives full play to the advantages of disaster text big data, and builds a large sample database for natural disasters in photovoltaic projects through natural language processing and geographic information system platforms. Therefore, the natural disaster knowledge graph constructed by this method can be more universal and complete.

[0042] 2. A method for constructing a knowledge graph of natural disaster risks for photovoltaic projects solves the subjectivity of existing methods in extracting causal relationships between events (disaster subjects) at all levels in the natural disaster risk system of photovoltaic projects, making the natural disaster knowledge graph constructed by this method more objective: This method constructs an intelligent extraction model of causal relationships of disaster events based on machine learning, and deeply mines the microscopic causal relationship structures at all levels implied in natural disasters in photovoltaic projects, between disaster-stricken bodies, and between disaster-stricken factors and disaster-stricken bodies from the database text information. It has obvious advantages in drawing a generalized and objective knowledge graph of natural disaster risks for photovoltaic projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 The steps of the method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to the present invention;

[0044] Figure 2 It is a schematic diagram of the method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to the present invention. DETAILED DESCRIPTION

[0045] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0046] The features and performance of the present invention are further described in detail below in conjunction with the embodiments.

[0047] Embodiment 1

[0048] See also Figure 1, a method for constructing a knowledge graph of natural disaster risks for photovoltaic projects, specifically comprising the following steps:

[0049] Step S1: constructing a natural disaster geographic information database for photovoltaic projects based on text big data;

[0050] Step S2: Extracting the causal relationship of natural disaster risks of photovoltaic projects based on machine learning;

[0051] Step S3: Based on the PV project natural disaster geographic information database and the causal relationship of PV project natural disaster risks, a knowledge map of PV project natural disaster risks is drawn.

[0052] In this embodiment, specifically, step S1 includes:

[0053] Step S11: Big data crawling and corpus construction of natural disaster texts of photovoltaic projects;

[0054] Step S12: Disaster text big data cleaning and preprocessing;

[0055] Step S13: mining disaster attribute features based on natural language processing;

[0056] Step S14: Construction of a natural disaster geographic information database for photovoltaic projects.

[0057] Furthermore, the step S11 includes:

[0058] In this embodiment, specifically, based on the international authoritative emergency disaster data set (Emergency Events Database, EM-DAT), a list of disaster events for photovoltaic projects is screened and obtained;

[0059] Step S112: using "photovoltaic disaster" as a keyword, using a web crawler to automatically crawl text big data about photovoltaic disasters around the world from mainstream portals at home and abroad, social media, and official reports according to the keyword;

[0060] Step S113: Integrate the photovoltaic project disaster event list and disaster text big data to build a photovoltaic project natural disaster text big data corpus.

[0061] In this embodiment, specifically, step S12 includes:

[0062] The preprocessing of the photovoltaic project natural disaster text big data corpus was completed through data cleaning, word segmentation, part-of-speech tagging, and stop word removal to meet the format requirements of natural language processing;

[0063] Among them, noise content is removed through data cleaning, and only the title, abstract, and text information are saved; the noise content includes: repetition, advertisement, label, HTML, JS code and comments; then the complete sentences in the photovoltaic project natural disaster text big data corpus are decomposed into the most individual words or phrases as the smallest unit of text mining processing by statistical or rule-based methods, and then the part-of-speech tagging is completed according to the division rules of adjectives, verbs, nouns, etc.; finally, stop words that do not contribute to the text features are removed.

[0064] In this embodiment, specifically, step S13 includes:

[0065] Based on the big data semantic intelligent analysis platform NLPIR (Natural Language Processing and Information Retrieval) and the language technology platform LTP (Language Technology Platform), the temporal and non-temporal attributes of each disaster historical event, such as type, place name, time, disaster situation, disaster-causing factors, and disaster-bearing bodies, are mined and acquired; the basic idea is: to take the semantic framework of the event as the guide, and to run through each step of the text information extraction process such as corpus training, text classification and information extraction model, and finally realize the mining of disaster attribute features; the specific steps are as follows:

[0066] Step A: Use classification models such as support vector machines to identify disaster event types and disaster-causing factors;

[0067] Step B: Construct the regular expression of time entity and extract the time information in the disaster text;

[0068] Step C: Based on the global place name database GeoNames, use the language technology platform to extract location entities from the text and accurately locate the location of the disaster to the city-level administrative unit;

[0069] Step D: Extract disaster loss information through Bootstrapping weakly supervised learning according to the disaster loss fields such as the affected population, the death population, and the economic loss;

[0070] Step E: Extract the specific disaster-affected entities according to the system (social, economic, ecological) to which the disaster loss belongs, such as the public health sector and infrastructure sector of the social system, the industrial, agricultural and other production sectors of the economic system, and the forest and other sectors of the ecological system.

[0071] In this embodiment, specifically, step S14 includes:

[0072] The expressions of identified event types, place name elements, time elements, disaster damage elements and non-spatiotemporal attributes are standardized to form a complete historical disaster event information table. Then, based on the ArcGIS geographic information system platform, the place name information in the historical disaster event information table is used to generate a point feature dataset, and the non-spatiotemporal attribute tables such as type, event, disaster damage and affected departments are connected to build a photovoltaic project natural disaster geographic information database to realize data storage, visualization, access and application.

[0073] In this embodiment, specifically, step S2 includes:

[0074] The photovoltaic project natural disaster geographic information database is divided into a training set and a test set. First, the word2vec and ELMO word embedding models are used to represent the text sequences respectively, and a joint vector is constructed by integrating the pre-trained word vector and the ELMO dynamic word feature. Then the joint vector is input into the neural network model to extract long-distance semantic features in depth, and a weight vector is generated through the attention mechanism. The features are weighted to highlight the contribution of important lexical features. The softmax classifier is then used to classify the weighted features to complete the training of the causal relationship extraction model. Finally, the accuracy and recall rate indicators are used to evaluate the model training effect. After the model accuracy meets the standard, the disaster causal relationship model based on the neural network is used to extract the micro-causal relationship of risk factors at all levels implied in the photovoltaic project natural disaster geographic information database, including the disaster-causing factors, the disaster-affected bodies, and the disaster-causing factors and the disaster-affected bodies.

[0075] In this embodiment, specifically, step S3 includes:

[0076] The semantic similarity between risk factors is measured by calculating the angle cosine distance, and the extracted specific risk factors are abstracted and generalized based on the semantic similarity results;

[0077] Based on the regional natural disaster system theory, from the perspectives of disaster-prone environment, disaster-causing factors, and disaster-bearing bodies, the single causal relationships containing the same risk factors are connected end to end according to the causal logic chain to form a chain of causal relationships, and then interweave and gather a relatively complete disaster knowledge map;

[0078] Based on complex network theory and combined with prior knowledge, a generalized natural disaster risk network topology of photovoltaic projects is objectively drawn to construct a knowledge map of natural disaster risks of photovoltaic projects.

[0079] In this embodiment, specifically, the angle cosine distance is calculated using the following formula:

[0080]

[0081] Among them: |W|, |K| are the total frequencies of the two co-occurring keywords respectively, W∩K is the co-occurrence frequency of the two keywords, and the value range of Cosine(W,K) is [0,1]. The larger the value, the stronger the keyword correlation.

[0082] Embodiment 2

[0083] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0084] Download the most authoritative international emergency disaster database (EM-DAT), and screen and obtain disaster events related to PV projects worldwide, including geological and seismic disasters (landslides, mudslides, etc.), meteorological and hydrological disasters (typhoons, tornadoes, floods, blizzards, freezing, etc.), and production accidents. Use web crawler technology (such as the network data collector Octopus software, etc.) to automatically crawl the text big data about PV project natural disaster reports, records, and reports from the most widely covered and influential domestic and foreign news media (Xinhuanet, BBC, CNN, etc.), social media platforms (Weibo, Twitter), and government official websites according to keywords such as "PV project disasters". Integrate the EM-DAT PV project disaster event list and disaster text big data to build a corpus of PV project natural disaster text big data.

[0085] The corpus is preprocessed through four steps: data cleaning, word segmentation, part-of-speech tagging, and stop word removal.

[0086] Corpus cleaning: Remove noise content through manual deduplication, alignment, deletion and annotation, remove invalid information such as advertisements, tags, HTML, JS and other codes and comments, and extract valid information such as title, abstract, and text from the original text.

[0087] Word segmentation: Through string matching, understanding, rule-based and statistical-based word segmentation methods, the complete sentences in the expectation library are broken down into the most individual words or phrases as the smallest unit of text mining processing.

[0088] Part-of-speech tagging: Through rule-based (maximum entropy-based part-of-speech tagging, statistical maximum probability output part-of-speech tagging, HMM-based part-of-speech tagging, etc.) and statistical methods, each word or phrase is labeled with a word class, such as adjective, verb, noun, adverb, preposition, pronoun, conjunction, etc., to incorporate more useful language information into the knowledge reasoning constructed by the knowledge graph.

[0089] Remove stop words: remove words that do not contribute to text features, such as punctuation, tone, person names, and other stop words.

[0090] Using the preprocessed corpus, based on the NLPIR (Natural Language Processing and Information Retrieval) big data semantic intelligent analysis platform and the LTP (Language Technology Platform) platform, the rule-based extraction model and the statistical-based extraction model are used to extract text disaster information to achieve the purpose of mutual verification. Specifically, there are four key steps: type extraction, time extraction, place name extraction, and impact footprint (including system and department) extraction of disaster events:

[0091] Disaster type extraction: Support Vector Machine (SVM) is used as the classification model to extract keywords from all news titles in the corpus. According to the National Standard of the People's Republic of China "Classification and Code of Natural Disasters" (GB / T 28921-2012), the main disaster types and seed trigger words are determined. The 10 words closest to various sub-trigger words are found, and irrelevant words are manually deleted to finally determine the seed word list. Given a certain news text, if a trigger word of a disaster type appears, this type is used as the disaster type of the news text.

[0092] Time extraction: Based on the characteristics of time entities, the corresponding time regular expressions are constructed to match the time in the news text for extraction. Check whether there is a time entity in the text. If not, the news release time is used as the disaster occurrence time. If there is no reporting time, the first time entity identified in the news text is used as the disaster occurrence time.

[0093] Place name extraction: Based on the world place name database, the LTP platform is used to extract location entities from news texts, accurately pinpointing the location of the disaster to the city-level administrative unit.

[0094] Extraction of impact footprint (including systems and sectors): Extract natural disaster disaster information of photovoltaic projects according to fields such as the affected population, death population, and economic losses. Extract the disaster-prone body reflected by the disaster loss information through methods such as Bootstrapping weak supervision learning. Extract the various industries affected by the disaster according to the system (social, economic, and ecological) to which the disaster-prone body belongs, such as the public health department and infrastructure department of the social system, the industrial and agricultural production departments of the economic system, and the ecological service departments of the ecosystem.

[0095] The expression of non-temporal and spatial attributes such as event types, time elements, place name elements, and disaster impact footprints identified in the above steps is standardized to form a complete historical disaster event information table. Based on the structured query language (SQL) database management system and ArcGIS geographic information system platform, the place name information in the disaster information table is used to generate a point feature dataset and connect non-temporal and spatial attribute tables such as type, event, disaster damage, and affected departments. By adding topology, network or subtype to expand the geographic database, maintain data integrity and process important spatial relationship sets, a photovoltaic project natural disaster geographic information database is constructed.

[0096] The PV project natural disaster geographic information database is divided into a training set and a test set in a ratio of 7:3. The word segmentation tool is used to segment the input text data into words to obtain word sequence data. Word2vec is used to complete the vectorization of word text and obtain pre-trained word vectors. Then the ELMO word embedding model is used to generate dynamic word features. The pre-trained word vectors generated by word2vec and the dynamic word features generated by the ELMO model are combined to construct a joint vector.

[0097] The joint vector is input into the neural network model to deeply extract long-distance semantic features, and the weight vector is generated through the attention mechanism to weight the features and highlight the contribution of important vocabulary features. The weighted features are then classified using the softmax classifier to complete the training of the causal relationship extraction model. Finally, the accuracy (Precision), recall rate (Recall) and F value (F-Score) and other indicators are used to evaluate the model training effect. Among them:

[0098]

[0099] After the model accuracy reaches the standard, the disaster causal relationship extraction model based on neural network is used to extract the implicit causal relationship in each individual event in the natural disaster database of photovoltaic projects. The angle cosine distance is used to calculate the semantic similarity of the specific events extracted from the corpus, and the extracted specific events are abstracted and generalized based on the semantic similarity results. The angle cosine distance formula is as follows:

[0100]

[0101] |W|, |K| are the total frequencies of the two co-occurring keywords, W∩K is the co-occurrence frequency of the two keywords, and the value range of Cosine(W,K) is [0,1]. The larger the value, the stronger the keyword correlation.

[0102] Here, by selecting sample data for test analysis multiple times, the similarity threshold is finally set to 0.8. If the similarity between events is higher than the threshold, the event may need to be further generalized.

[0103] Through the abstraction and generalization of events, the causal relationship between primary disasters, secondary disasters and derivative disasters implied in each historical disaster case is obtained. By comprehensively using complex network modeling and analysis software platforms such as NetworkX, igraph, and Gephi, based on the theory of regional natural disaster systems, from different perspectives of disaster-prone environment, disaster-causing factors, and disaster-bearing bodies, the single causal relationship involving the same event (primary disasters, secondary disasters, and derivative disasters) is connected end to end according to the causal logic chain, forming a chain of causal relationships, and then interweaving and gathering a relatively complete knowledge map of natural disasters in photovoltaic projects.

[0104] The above-mentioned embodiments only express the specific implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the protection scope of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the technical solution concept of the present application, and these all belong to the protection scope of the present application.

[0105] This background section is provided to generally present the context of the invention, and the work of the presently named inventors, the work to the extent described in this background section, and aspects of this section that did not constitute prior art at the time of application are neither explicitly nor implicitly admitted to be prior art to the present invention.

Claims

1. A method for constructing a knowledge graph of natural disaster risks for photovoltaic projects, characterized in that: include: Step S1: constructing a natural disaster geographic information database for photovoltaic projects based on text big data; Step S2: Extracting the causal relationship of natural disaster risks of photovoltaic projects based on machine learning; Step S3: Based on the PV project natural disaster geographic information database and the causal relationship of PV project natural disaster risks, a knowledge map of PV project natural disaster risks is drawn.

2. The method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to claim 1, characterized in that: The step S1 comprises: Step S11: Big data crawling and corpus construction of natural disaster texts of photovoltaic projects; Step S12: Disaster text big data cleaning and preprocessing; Step S13: mining disaster attribute features based on natural language processing; Step S14: Construction of a natural disaster geographic information database for photovoltaic projects.

3. The method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to claim 2, characterized in that: The step S11 comprises: Step S111: Screening and obtaining a list of disaster events for photovoltaic projects based on an internationally authoritative emergency disaster dataset; Step S112: using photovoltaic disasters as keywords, using a web crawler to automatically crawl text big data about photovoltaic disasters around the world from mainstream portals at home and abroad, social media, and official reports based on keywords; Step S113: Integrate the photovoltaic project disaster event list and disaster text big data to build a photovoltaic project natural disaster text big data corpus.

4. The method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to claim 3, characterized in that: The step S12 comprises: The preprocessing of the photovoltaic project natural disaster text big data corpus is completed through data cleaning, word segmentation, part-of-speech tagging, and removal of stop words to meet the format requirements of natural language processing; wherein, the noise content is removed by data cleaning, and only the title, abstract, and text information are saved; the noise content includes: repetition, advertisement, label, HTML, JS code, and comments; then, the complete sentences in the photovoltaic project natural disaster text big data corpus are decomposed into the most individual words or phrases using statistical or rule-based methods as the smallest unit of text mining processing, and then the part-of-speech tagging is completed according to the division rules; finally, the stop words that do not contribute to the text features are removed.

5. The method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to claim 4, characterized in that: The step S13 comprises: Based on the big data semantic intelligent analysis platform and language technology platform, the spatiotemporal and non-spatiotemporal attributes of each historical disaster event are mined and acquired; the spatiotemporal and non-spatiotemporal attributes include: type, place name, time, disaster situation, disaster-causing factors, and disaster-bearing body.

6. The method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to claim 5, characterized in that: The mining obtains the spatiotemporal and non-spatial attributes of each disaster historical event, including: Step A: Select a classification model to identify disaster event types and disaster-causing factors; Step B: Construct the regular expression of time entity and extract the time information in the disaster text; Step C: Based on the global place name database GeoNames, use the language technology platform to extract location entities from the text and accurately locate the location of the disaster to the city-level administrative unit; Step D: According to the disaster loss field, the disaster loss information is extracted through Bootstrapping weakly supervised learning; Step E: Extract the specific disaster-affected objects according to the system to which the disaster damage belongs.

7. The method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to claim 6, characterized in that: The step S14 comprises: The expressions of the identified event types, place name elements, time elements, disaster damage elements and non-spatiotemporal attributes are standardized to form a complete historical disaster event information table. Based on the ArcGIS geographic information system platform, the place name information in the historical disaster event information table is used to generate a point feature dataset, which is connected to the non-spatiotemporal attribute table to build a photovoltaic project natural disaster geographic information database to achieve data storage, visualization, access and application. The non-spatiotemporal attribute table includes: type, event, disaster damage and affected department.

8. The method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to claim 7, characterized in that: The step S2 comprises: The photovoltaic project natural disaster geographic information database is divided into a training set and a test set. First, the word2vec and ELMO word embedding models are used to represent the text sequences respectively, and a joint vector is constructed by integrating the pre-trained word vector and the ELMO dynamic word feature. Then the joint vector is input into the neural network model to extract long-distance semantic features in depth, and a weight vector is generated through the attention mechanism. The features are weighted to highlight the contribution of important lexical features. The softmax classifier is then used to classify the weighted features to complete the training of the causal relationship extraction model. Finally, the accuracy and recall rate indicators are used to evaluate the model training effect. After the model accuracy meets the standard, the disaster causal relationship model based on the neural network is used to extract the micro-causal relationship of risk factors at all levels implied in the photovoltaic project natural disaster geographic information database, including the disaster-causing factors, the disaster-affected bodies, and the disaster-causing factors and the disaster-affected bodies.

9. The method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to claim 8, characterized in that: The step S3 comprises: The semantic similarity between risk factors is measured by calculating the angle cosine distance, and the extracted specific risk factors are abstracted and generalized based on the semantic similarity results; Based on the regional natural disaster system theory, from the perspectives of disaster-prone environment, disaster-causing factors, and disaster-bearing bodies, the single causal relationships containing the same risk factors are connected end to end according to the causal logic chain to form a chain of causal relationships, and then interweave and gather a relatively complete disaster knowledge map; Based on complex network theory and combined with prior knowledge, a generalized natural disaster risk network topology of photovoltaic projects is objectively drawn to construct a knowledge map of natural disaster risks of photovoltaic projects.

10. The method for constructing a knowledge graph of natural disaster risks for photovoltaic projects according to claim 9, characterized in that: The angle cosine distance is calculated using the following formula: Among them: |W|, |K| are the total frequencies of the two co-occurring keywords respectively, W∩K is the co-occurrence frequency of the two keywords, and the value range of Cosine(W,K) is [0,1]. The larger the value, the stronger the keyword correlation.

Images