Intelligent data analysis method for LNG facilities integrating multidimensional data and POI dynamic data

Abstract

This invention discloses an intelligent data analysis method for LNG facilities that integrates multi-dimensional data and dynamic POI data. The method includes: constructing an LNG facility thesaurus; obtaining and organizing a list of core LNG facilities to obtain a preliminary LNG facility database; using a big data model to obtain LNG facility data and LNG electronic waybill data to supplement and optimize the preliminary LNG facility database, resulting in a final LNG facility database; a dynamic attribute update module using a dynamic dataset to dynamically adjust the operational status and activity level attribute information in the final LNG facility database; and a multi-scenario in-depth analysis module for LNG facility statistical display, supply and demand forecasting, and scheduling matching planning. This invention can obtain a comprehensive, accurate, timely, and dynamically attributed final LNG facility database, providing comprehensive and accurate data support for subsequent data analysis in the multi-scenario in-depth analysis module, and realizing in-depth analysis and processing of LNG facilities such as monitoring, supply and demand forecasting, and scheduling.

Description

Technical Field

[0001] This invention relates to the fields of hazardous goods facility management, supply and demand transportation data analysis, and LNG hazardous goods transportation management analysis through multi-source data fusion and utilization, and particularly to an intelligent data analysis method for LNG facilities that integrates multi-dimensional data and POI dynamic data. Background Technology

[0002] Liquefied natural gas (LNG) is a type of dangerous goods. LNG is transported through two main types of facilities: loading facilities (the LNG supply side, such as LNG receiving terminals and LNG plants) and unloading facilities (the LNG usage side, such as transportation, industrial use, city gas, and gas-fired power plants). As a clean and efficient energy source, LNG transportation must strictly adhere to the "Regulations on the Safety Management of Road Transportation of Dangerous Goods." Electronic waybills are the core carrier for recording information throughout the entire "loading-transportation-unloading" process (including loading / unloading addresses and their administrative division codes, shipper, consignee, vehicle information, transportation company, transport volume, and transport time). With the expansion of the LNG industry and the advancement of digital transformation, the industry's demand for dynamic, correlated, and intelligent applications of LNG-specific data is becoming increasingly urgent. However, existing technologies still have many unresolved pain points, specifically: First, low address resolution accuracy; waybills contain vague descriptions of "loading address" and "unloading address" (e.g., "Tianjin Binhai New Area") and "unloading address" (e.g., "Beijing Tongzhou District"), and there are frequent typos ("receiving station" is mistakenly written as "receiving station", "Diefu Road" is mistakenly written as "Diefu Road"); there is a lack of accurate identification capabilities for LNG-specific facility types, including loading facilities (receiving stations, LNG plants) and unloading facilities (transportation gas, industrial gas, city gas, gas-fired power plants), making it impossible to form a set of LNG-specific POIs. Second, insufficient data correlation; existing technologies only utilize address text or vehicle trajectory data without considering the characteristics of LNG transportation (e.g., the industry attributes of transportation companies' exclusive cooperative receiving stations and industrial gas terminals), resulting in the inability to complete vague addresses. Third, weak support for trade analysis; traditional POI point sets only contain geographical coordinates, which cannot realize trade-related data such as "LNG loading type - unloading type - regional flow - transportation volume", and cannot meet the needs of LNG trade supply and demand matching, price linkage analysis, etc. Fourth, lack of POI dynamism; traditional POIs are static data, only recording initial construction information, and cannot reflect changes in facility operation status (such as the closure of small LNG refueling stations, temporary shutdown of receiving terminals, fluctuations in facility activity, etc.), resulting in lags in trade planning and capacity scheduling, and even resource waste or supply gaps. Fifth, insufficient data support for vertical domain AI models; existing general AI models lack high-quality labeled data specific to the LNG field, and cannot accurately adapt to industry scenarios (such as supply and demand gap prediction, risk warning, intelligent scheduling), while the industry has not yet formed an integrated AI training dataset of "facilities - waybills - dynamic status - trade attributes". Sixth, supply guarantee analysis is highly lagging; traditional analysis of heating supply and peak summer demand is mostly based on post-event statistics (such as monthly transportation volume summaries), lacking real-time monitoring and pre-event forecasting capabilities, making it impossible to predict regional supply and demand gaps in advance, and difficult to support emergency dispatch decisions.

[0003] Currently, in the area of ​​dangerous goods transportation supervision, Chinese patent CN112200525A discloses a dangerous goods transportation supervision system based on electronic waybills and real-time trajectory monitoring. It primarily focuses on automatically retrieving qualification information for review by obtaining dangerous goods waybill filing requests, and calibrating vehicle movement trajectories and issuing alarms based on multi-party data to achieve transportation qualification review and vehicle operation compliance supervision. This patent does not involve in-depth analysis of electronic waybill address information, cannot accurately locate loading and unloading points and classify their types, and lacks dynamic updates and multi-scenario empowerment functions. In the field of waybill data enhancement, Chinese patent CN119090385B discloses a waybill enhancement method based on dangerous goods transport vehicle trajectory data. It emphasizes using vehicle trajectory data for waybill enhancement, generating enhanced waybills through trajectory data interpolation, identification of valid stopping points, and analysis of stopping behavior, thereby improving the accuracy of transportation trajectory, stopping point, and other information in the waybill. However, its processing of address information is limited to location matching, without delving into the meaning of the address text, making it difficult to achieve accurate classification of POI types based on address semantics, and it does not involve dynamic status annotation and AI model application.

[0004] Existing dynamic POI technologies (mostly focused on map navigation) fail to consider the specific characteristics of the LNG industry (such as transportation company partnerships, supply scheduling needs, and industry-specific facility types). They only focus on dynamic location updates and lack a correlation between "status, activity level, and trade attributes," making them unsuitable for the diverse needs of the LNG industry. Furthermore, current technologies have not yet formed a complete technical solution encompassing "dynamic POI dataset construction, vertical AI model support, and multi-scenario intelligent empowerment," thus failing to meet the digital, intelligent, and emergency response development needs of the LNG industry. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent data analysis method for LNG facilities that integrates multi-dimensional data and dynamic POI data. A preliminary LNG facility database is obtained by organizing the acquired core LNG facility list according to an LNG facility thesaurus. Then, a big data model is used to acquire LNG facility data with LNG as the Point of Interest (POI) to supplement and optimize the preliminary LNG facility database. Next, LNG electronic waybill data is acquired to further supplement and optimize the preliminary LNG facility database, resulting in a comprehensive, accurate, timely, standardized, and dynamic final LNG facility database. This provides data technology support for subsequent data analysis in multi-scenario in-depth analysis modules.

[0006] The objective of this invention is achieved through the following technical solution: A data intelligence analysis method for LNG facilities that integrates multidimensional data and dynamic POI data includes the following steps: S1. Construct an LNG facility thesaurus. The LNG facility thesaurus has a hierarchical structure. The first level of the LNG facility thesaurus includes facility keyword attributes, geographic location mapping attributes, operating entity association attributes, and dynamic attributes. The second level of facility keyword attributes includes the full name of the facility, the abbreviation of the facility, the facility type, and the facility subtype. The facility type includes loading and unloading. The facility subtypes are further subdivided according to loading and unloading types. The second level of geographic location mapping attributes includes administrative division code, administrative region, and geographic identifier. The geographic identifier includes geographic location information and geographic coordinate information. The second level of operating entity association attributes includes the operating entity, the investment entity, and related attributes. The second level of dynamic attributes includes the operating status and activity level.

[0007] S2. Obtain and screen the list of core LNG facilities and organize it according to the LNG facility thesaurus to obtain a preliminary LNG facility database. Use big data models to obtain LNG facility data to supplement and optimize the preliminary LNG facility database.

[0008] S3. Collect and acquire LNG electronic waybill data to supplement and optimize the preliminary LNG facility database to obtain the final LNG facility database.

[0009] S4. Construct a dynamic attribute update module to obtain a dynamic dataset associated with the final LNG facility database. The dynamic dataset includes LNG facility change data and LNG electronic waybill data. The dynamic attribute update module dynamically adjusts the attribute information of the operating status and activity level in the final LNG facility database.

[0010] S5. Construct a multi-scenario deep analysis module, which includes a 3D analysis model unit. The 3D analysis model unit is used to monitor and display the distribution, operating status, activity level, and LNG transportation volume of LNG facilities in real time on a geographic map, with LNG facilities as the point of interest. The 3D analysis model unit performs historical time-series data statistics on supply and demand according to the facility type of LNG facilities and uses a bidirectional long short-term memory network to predict supply and demand data and supply and demand gaps for future time periods. The 3D analysis model unit performs optimal scheduling route matching and planning according to the loading and unloading types of LNG facilities.

[0011] To better realize this invention, the multi-scenario deep analysis module also includes a trade matching module unit. The trade matching module unit extracts the LNG facility type as loading and evaluates loading capacity, activity level, and regional distribution, and regards the loading LNG facility as the trade supplier. The trade matching module unit extracts the LNG facility type as unloading and evaluates unloading demand, activity level, and regional distribution, and regards the unloading LNG facility as the trade demander. The trade matching module unit constructs a supply and demand matching matrix between trade suppliers and trade demanders and sets the shortest supply and demand transportation distance as the optimal supply and demand matching matrix.

[0012] Preferably, the multi-scenario deep analysis module further includes an operation and maintenance supervision module unit, which is used to dynamically add or remove data on the trading suppliers and trading demanders in the supply and demand matching matrix and rematch the optimal supply and demand matching matrix; the operation and maintenance supervision module unit extracts LNG electronic waybill data from the dynamic dataset to identify and warn of transportation non-compliance.

[0013] Preferably, in method S2, the big data model obtains LNG facility data from the Internet platform and performs attribute matching with the LNG facilities included in the preliminary LNG facility database and completes the attribute information. If the LNG facility data contains LNG facilities not included in the preliminary LNG facility database, a list of LNG facilities to be supplemented is formed, and the preliminary LNG facility database is supplemented with LNG facilities from the list of LNG facilities to be supplemented.

[0014] Preferably, the facility keyword attributes in the preliminary LNG facility database are processed using high-frequency related words obtained from LNG facility data through a big data model. The method includes: extracting high-frequency related words from the LNG facility data using the TextTank algorithm, setting a word frequency threshold ≥ 5 and a mutual information threshold ≥ 0.8; calculating the semantic cosine similarity between the high-frequency related words and the facility keyword attributes in the preliminary LNG facility database using a pre-trained BERT-Large model, and using high-frequency related words with a semantic cosine similarity ≥ 0.85 as related word tag information for the facility keyword attributes; and adding related word tag information for the geographic location mapping attribute and / or the operating entity association attribute according to the above method.

[0015] Preferably, in method S3, the geographic location data obtained from the LNG electronic waybill data is labeled and completed for the geographic location mapping attributes in the preliminary LNG facility database. The method includes: performing word segmentation processing on the geographic location data in the LNG electronic waybill data, and performing data completion and standardization processing according to three levels: administrative region, road or park, and location feature words. The administrative region includes the three levels of province, city, and county or district.

[0016] Preferably, in method S3, data optimization includes keyword clustering processing. The K-Means clustering algorithm is used in combination with TF-IDF feature weighting to cluster the semantically split keywords. The facility type clustering processing includes loading and unloading classes. The loading and unloading classes are respectively set with subtypes according to the hierarchy. The subtypes of the loading class include LNG receiving station and LNG plant. The subtypes of the unloading class include gas power plant, transportation gas, industrial gas and city gas. The subtypes of the loading class correspond to the facility subtypes of the loading subclass, and the subtypes of the unloading class correspond to the facility subtypes of the unloading subclass.

[0017] Preferably, data standardization is performed before obtaining the final LNG facility database. Data standardization includes coordinate unification and address normalization. Geographic coordinate information in the geographic identifier is uniformly converted to coordinate information in the WGS-84 coordinate system. Address normalization is performed according to three levels: administrative region, road or park, and location feature words.

[0018] Preferably, in method S4, the final LNG facility database includes the following operational status types: normal operation, temporary shutdown, renovation and upgrade, seasonal shutdown, pending attention, relocation pending update, shutdown, and deregistration. The dynamic attribute update module sets the logical combination judgment rules corresponding to each status type. The dynamic attribute update module judges the operational status of the LNG facility in the dynamic dataset according to the logical combination judgment rules and records the attribute labels of the operational status in real time.

[0019] Preferably, in method S4, the activity level in the final LNG facility database includes activity types such as high activity, medium activity, low activity, and very low activity. The dynamic attribute update module sets the activity state variable combination judgment rules corresponding to each activity type. The activity state variable combination judgment rules are constructed according to the upper and lower levels of activity type and facility subtype. The dynamic attribute update module judges the activity level of LNG facilities in the dynamic dataset according to the activity state variable combination judgment rules and records the attribute labels of the activity level in real time.

[0020] Compared with the prior art, the present invention has the following advantages and beneficial effects: (1) This invention constructs an LNG facility thesaurus. By obtaining the list of core LNG facilities, a preliminary LNG facility database is obtained by organizing the LNG facility thesaurus. Then, a big data model is used to obtain LNG facility data with LNG as the point of interest (POI) to supplement and optimize the preliminary LNG facility database. Next, LNG electronic waybill data is obtained to further supplement and optimize the preliminary LNG facility database, resulting in a comprehensive, accurate, timely, standardized, and dynamic final LNG facility database, which provides data technology support for various data analyses in the subsequent multi-scenario in-depth analysis module.

[0021] (2) This invention obtains a dynamic dataset associated with the final LNG facility database through a dynamic attribute update module, enabling dynamic adjustment of the attribute information of the operating status and activity level in the LNG facility database. This allows for timely updates and adjustments of dynamic attributes, achieving real-time adjustment and updating of dynamic attributes. The three-dimensional analysis model unit of the multi-scenario deep analysis module can monitor and display the distribution, operating status, activity level, and LNG facility transportation volume in real time on a geographic map according to geographical location. It also predicts future supply and demand data and supply and demand gaps through a bidirectional long short-term memory network, and performs optimal scheduling route matching and planning according to loading and unloading types.

[0022] (3) The trade matching module of the multi-scenario deep analysis module of the present invention constructs a supply and demand matching matrix between trade suppliers and trade demanders and sets the shortest supply and demand transportation distance as the optimal supply and demand matching matrix to realize the supply and demand matching optimization of LNG loading and unloading facilities; the operation and maintenance supervision module can dynamically add or remove data of trade suppliers and trade demanders in the supply and demand matching matrix and rematch the optimal supply and demand matching matrix, while extracting LNG electronic waybill data from the dynamic dataset for identification and early warning of non-compliant transportation. Attached Figure Description

[0023] Figure 1 This is a flowchart of the intelligent data analysis method for LNG facilities according to the present invention; Figure 2 This is a schematic diagram illustrating the principles of data collection, supplementation, and optimization in the LNG facility database of this embodiment. Figure 3 This is a schematic diagram illustrating the principle of LNG facility operation status determination in the embodiment; Figure 4 is a schematic diagram of the principle for determining the active status of LNG facilities in the embodiment; Figure 5 is a 3D map of LNG facilities in a certain province in the embodiment; Figure 6 is a comparison chart of monthly LNG loading at an imported LNG receiving terminal in a certain study area, as exemplified in the embodiments. Figure 7 is a comparison chart of LNG loading and transportation volume in a certain study area in the example for two consecutive years; Figure 8 is a diagram showing the LNG loading and unloading transportation flow of the top 10 loading and unloading areas in a certain study region, as illustrated in the example. Figure 9 is a comparison chart of the top 15 regions in terms of LNG loading volume in a certain study area, as exemplified in the embodiments. Detailed Implementation

[0024] The present invention will be further described in detail below with reference to embodiments: Example

[0025] like Figure 1 As shown, an intelligent data analysis method for LNG facilities that integrates multidimensional data and POI dynamic data includes the following steps: S1. Construct an LNG facility thesaurus. The LNG facility thesaurus has a hierarchical structure. The first level of the LNG facility thesaurus includes facility keyword attributes, geographic location mapping attributes, operating entity association attributes, and dynamic attributes. The facility keyword attributes include the full name of the facility, the abbreviation of the facility (including the facility's short name and alternative name), facility type, and facility subtypes at the second level. Facility types include loading and unloading, and facility subtypes are further subdivided according to loading and unloading types. There are associations between facility types and facility subtypes. If the facility type is unloading, the facility subtypes include LNG receiving terminals and LNG plants according to the unloading type. If the facility type is loading, the facility subtypes include gas-fired power plants, transportation gas, industrial gas, and city gas according to the loading type. The geographic location mapping attributes include administrative division codes, administrative regions, and geographic identifiers at the second level. Geographic identifiers include geographic location information and geographic coordinate information. The operating entity association attributes include the operating entity, investment entity, and related attributes at the second level. The dynamic attributes include the operating status and activity level at the second level. An example of the LNG facility thesaurus is shown in Table 1 below.

[0026] S2. Obtain and screen the list of core LNG facilities and organize them according to the LNG facility thesaurus to obtain a preliminary LNG facility database. See Table 2. A small number of LNG facilities are organized and recorded in the preliminary LNG facility database according to the LNG facility thesaurus.

[0027] Table 2. Examples of some LNG facilities in the preliminary LNG facility database.

[0028] like Figure 1 , Figure 2As shown, a big data model is used to acquire LNG facility data to supplement and optimize the preliminary LNG facility database (including completing the data for included LNG facilities and supplementing the data for unincluded LNG facilities). In some embodiments, the big data model acquires LNG facility data from internet platforms (including industry data platforms, local energy bureau LNG facility registration platforms, and industry association public directory platforms) (preferably, data matching of LNG facility data is performed according to an LNG facility thesaurus using regular expressions) and performs attribute matching and attribute information supplementation on the LNG facilities included in the preliminary LNG facility database. The LNG facilities originate from facility keyword attributes (including full facility name and abbreviation) in the preliminary LNG facility database. An LNG facility is an independent LNG facility in the preliminary LNG facility database (for facilities that are essentially the same LNG facility, the abbreviation and full name are classified as the same entity and assigned a unique LNG facility ID). The preliminary LNG facility database first includes LNG facilities one by one based on the LNG core facility list (the full name of the same LNG facility is recorded in the full facility name field, and the abbreviation or alias of the same LNG facility is recorded in the abbreviation / alias field). If the LNG facility data contains LNG facilities not included in the preliminary LNG facility database, a list of LNG facilities to be supplemented will be generated, and the preliminary LNG facility database will be supplemented with LNG facilities from this list.

[0029] The LNG facilities included in the preliminary LNG facility database are supplemented using LNG facility data obtained through a big data model. Preferably, the facility keyword attributes in the preliminary LNG facility database are supplemented with high-frequency related words using the LNG facility data obtained through the big data model (when supplementing the full name and abbreviation attributes of facility keyword attributes, the big data model obtains high-frequency related words related to LNG facility data for information supplementation and completion). The method includes: extracting high-frequency related words from the LNG facility data using the TextTank algorithm, setting a word frequency threshold ≥5 and a mutual information threshold ≥0.8; calculating the semantic cosine similarity between the high-frequency related words and the facility keyword attributes in the preliminary LNG facility database using a pre-trained BERT-Large model, and using high-frequency related words with a semantic cosine similarity ≥0.85 as related word tag information for the facility keyword attributes. The above method demonstrates the information completion of facility keyword attributes. Other attributes in the preliminary LNG facility database (including geographic location mapping attributes, operator entity association attributes, etc.) can also be processed using the same method. Specifically, this involves adding relevant keyword tags to geographic location mapping attributes and / or operator entity association attributes or other attributes. For example, in the operator entity association attribute, tags for the operator entity and investor entity are added. The preliminary LNG facility database of this invention includes "status tags" (normal operation / pending attention / closed / temporarily shut down, etc.) and "activity level reference thresholds" (high / medium / low activity corresponding to transportation volume / berthing frequency standards), providing field support for subsequent dynamic updates.

[0030] S3. Collect and acquire LNG electronic waybill data to supplement and optimize the preliminary LNG facility database to obtain the final LNG facility database. In some embodiments, the geographic location data in the LNG electronic waybill data is acquired and the geographic location mapping attributes in the preliminary LNG facility database are labeled and completed. The method includes: performing word segmentation on the geographic location nodes in the geographic location data of the LNG electronic waybill data, and realizing the identification of points of interest (POIs) in the geographic location data through six steps: word segmentation, clustering, error correction, labeling, completion, and standardization. Data completion and standardization are performed according to three levels: administrative region, road or park, and location feature words. The administrative region includes the three levels of province, city, and county or district.

[0031] For example, the study area is selected within the geographical scope of China, using national electronic waybill data from 2023 to 2024. Core fields are extracted: loading address, unloading address, loading administrative division code, unloading administrative division code, transportation company, shipper, consignee, license plate number, and loading time.

[0032] Address segmentation is performed based on the loading and unloading addresses in the waybill, using a word segmentation algorithm. Preliminary segmentation: Jieba (a natural language processing) is used, with a custom LNG facility module (method 1) loaded to achieve preliminary address segmentation (e.g., address "Guangdong Province, Zhuhai City, Jinwan District, Gaolan Port, Huandao West Road, Zhuhai LNG Receiving Station" → preliminary segmentation: "Guangdong Province / Zhuhai City / Jinwan District / Gaolan Port / Huandao West Road / Zhuhai / LNG / Receiving Station"). Segmentation optimization: The segmentation results are optimized using a BiLSTM-CRF model (bidirectional long short-term memory network + conditional random field), correcting semantic boundaries and outputting a structured address: "Administrative region (Guangdong Province / Zhuhai City / Jinwan District) + road / park (Gaolan Port, Huandao West Road) + location feature word (Zhuhai LNG Receiving Station)". The word segmentation process includes preliminary segmentation and segmentation optimization; the segmentation results are shown in Table 3.

[0033] This example uses the POI of Zhuhai Jinwan LNG receiving terminal as an example. The loading address of a certain electronic waybill is "Gaolan Port LNG receiving terminal, Jinwan District, Zhuhai City, Guangdong Province". The address is incomplete and has no coordinates. Its operation status needs to be tracked in the future.

[0034] 1. Implementation Process (1) POI construction: Jieba+BiLSTM-CRF word segmentation → matching LNG facility thesaurus (matching degree 98%) → calling Gaode API to obtain coordinates → standardized address is "Zhuhai LNG receiving station in the petrochemical storage area of ​​Gaolan Port Huandao West Road, Jinwan District, Zhuhai City, Guangdong Province, initial status marked "normal operation", facility category: loading, facility subcategory: LNG receiving station".

[0035] (2) Dynamic update: In Q3 2024, the receiving station had 20,000 stops and an average monthly transport volume of 140,000 tons in 3 months → the status was updated to "normal operation - high activity"; Assuming that it was suspended from operation in November 2024 due to equipment maintenance → the system can capture the notification through public information, automatically update its status from "normal operation" to "temporary shutdown", and record the notification number and validity period" → resume operation → update to "normal operation - high activity".

[0036] Implementation effect

[0037] The POI information is complete and its status is synchronized in real time, supporting trading companies in continuously allocating resources and preventing any waste of transportation capacity. The POI information type is accurately identified and can be directly used for transportation trajectory monitoring and trade flow analysis.

[0038] This embodiment takes the supplementation of information from Hongyuan Natural Gas Co., Ltd. in Jiaxian County, Yulin City, Shaanxi Province as an example. In the electronic waybill, there are 8,081 loading addresses marked as "Jiaxian County, Yulin City, Shaanxi Province". Only the administrative region to the county level is specified, but the specific facility name, geographical coordinates and type details are missing. It is necessary to take advantage of the industry characteristics of LNG transportation, which are "strong demand stability and high route fixity", to complete the accurate supplementation.

[0039] 1. Implementation Process (1) Analysis of Transportation Behavior Patterns Shaanxi Yulin Jiaxian Zhonghong Industry and Trade Co., Ltd. has a total of 1293 waybills. Shaanxi Yulin Jiaxian Hongyuan Natural Gas Co., Ltd. has a total of 503 waybills. The transportation company with the most trips for Shaanxi Yulin Jiaxian Zhonghong Industry and Trade Co., Ltd. is Yulin Yinzhou Automobile Transportation Co., Ltd., with 292 trips, accounting for 22.6%. 152 waybills were for unloading at Shenmu Shunda LNG refueling station, all transported by vehicle Shaanxi KC5661. The transportation company with the most trips for Shaanxi Yulin Jiaxian Hongyuan Co., Ltd. is Shanxi Xin'anjie Automobile Transportation Co., Ltd., with 117 trips, accounting for 22.1%, unloading at Dongguan Town, transported by 7 vehicles.

[0040] (2) Targeted analysis of fuzzy addresses Based on the above-mentioned transportation behavior patterns, a collaborative annotation model of "BERT classification algorithm + semantic vector + transportation characteristics" is used to analyze the fuzzy address.

[0041] Candidate facility matching: Match "Jiaxian County, Yulin City, Shaanxi Province" with LNG facility thesaurus such as Jiaxian Hongyuan Natural Gas Co., Ltd. and Jiaxian Zhonghong Industry and Trade Co., Ltd.

[0042] Transportation characteristics verification: Verification is conducted by combining the cooperative relationship with the transportation company "Shanxi Xin'anjie", the unloading address, and the carrier vehicles.

[0043] LNG facility identification: Jiaxian County, Yulin City, Shaanxi Province in the waybill refers to Shaanxi Yulin City Jiaxian Hongyuan Natural Gas Co., Ltd.

[0044] (3) POI information completion and standardization. The loading address is standardized as "Hongyuan Natural Gas Co., Ltd., Jiaxian County, Yulin City, Shaanxi Province, Facility Category: Loading, Facility Subcategory: LNG Plant". Implementation effect

[0045] Once identified, it can directly support regional industrial gas supply and demand statistics and compliance supervision of transportation routes.

[0046] The present invention performs data optimization processing on the preliminary LNG facility database. The data optimization includes keyword clustering processing. The K-Means clustering algorithm is used and combined with TF-IDF feature weighting to perform clustering processing on the keywords after semantic splitting. The clustering processing of facility types includes loading types and unloading types. The sub-types of the loading type and the unloading type are set hierarchically. The sub-types of the loading type include LNG receiving stations (including coastal import storage hubs, such as Zhuhai LNG Receiving Station) and LNG plants (including domestic LNG production and manufacturing bases, such as Shaanxi Yulin LNG Plant and Xinjiang Guanghui LNG Plant). The sub-types of the unloading type include gas power plants (for power production), transportation gas (for LNG heavy trucks, ship refueling, etc.), industrial gas (for industrial fields such as ceramic firing and chemical fiber production), and city gas (including gas supply stations for urban residents and commercial users). The sub-types of the loading type correspond to the facility sub-types in the loading breakdown, and the sub-types of the unloading type correspond to the facility sub-types in the unloading breakdown. During the clustering processing, a keyword vectorization processing method is adopted, and the Word2Vec model is used to perform keyword vectorization processing. The optimal number of clusters K is determined according to the required clustering situation (for example, if the sub-types of the loading type only include LNG receiving stations and LNG plants, then the number of clusters K is 2), and the optimal number of clusters K is determined by the Elbow Method; clusters with similar semantics are merged (such as "LNG receiving station" and "liquefied natural gas receiving station" are merged into the same cluster), and abnormal clusters (such as irrelevant address keywords) are removed; category labels are marked: category labels are marked for each cluster (for example, "receiving station cluster" → label "loading type - receiving station", "industrial gas terminal cluster" → label "terminal consumption type - industrial gas"), forming an exclusive keyword clustering library. In some embodiments, the data optimization processing further includes misspelling recognition and correction processing. The method includes: corpus support: adding the address data of LNG electronic waybills in the past year to the pre-constructed dangerous goods transportation address corpus, covering address expressions in different regions and different enterprises across the country to improve the adaptability of the error correction model. Dual-algorithm collaborative error correction: 1) Basic error correction: An error correction model based on the edit distance is adopted to calculate the character difference between the misspelling and the correct word, identify and correct the misspelling in the address. For example, "接受战" is corrected to "接收站", and the edit distance is 1. 2) Semantic error correction: On the basis of the error correction model based on the edit distance, the BERT pre-trained model is introduced to improve the error correction accuracy through context semantic analysis, and "跌幅" is modified to "迭福".

[0047] In some embodiments, data standardization is performed before obtaining the final LNG facility database. Data standardization includes coordinate unification and address normalization. Geographic coordinate information in the geographic identifier is uniformly converted to coordinate information in the WGS-84 coordinate system (or other internationally used geographic coordinate system). Address normalization is performed according to three levels: administrative region, road or park, and location feature words. Example: Zhuhai LNG receiving station in the petrochemical storage area of ​​Gaolan Port Huandao West Road, Jinwan District, Zhuhai City, Guangdong Province.

[0048] S4. Construct a dynamic attribute update module to obtain a dynamic dataset associated with the final LNG facility database. The dynamic dataset includes LNG facility change data and LNG electronic waybill data. The dynamic attribute update module dynamically adjusts the attribute information of the operating status and activity level in the final LNG facility database.

[0049] In some embodiments, the final operational status in the LNG facility database includes the following types: normal operation, temporary shutdown, renovation and upgrade, seasonal shutdown, pending monitoring, relocation pending update, closure, and deregistration. Normal operation is defined as continuous facility operation with loading / unloading capabilities and stable data feedback; examples include the daily operation of receiving terminals and industrial gas terminals. Temporary shutdown is defined as a short-term suspension of operation due to policy requirements (environmental protection restrictions), equipment maintenance, extreme weather, etc., which can be resumed later; examples include a gas-fired power plant suspending operations for one month due to maintenance, and a gas station being shut down for two weeks due to environmental inspections. Renovation and upgrade is defined as facility undergoing capacity enhancement, equipment updates, etc., during which operations are suspended; an example is the renovation of an LNG plant's new storage tank, lasting three months. Seasonal shutdown is defined as operation only during specific seasons (such as the heating season), with no data in non-operating seasons but valid registration; an example is small gas stations in some northern rural towns (operating only in winter). Pending monitoring is defined as missing operational data, but the registration has not been deregistered, with the possibility of resuming operations or complete closure; an example is a small industrial gas terminal with no delivery orders for three months, but the company has not been deregistered. "Relocation pending update" is defined as the actual relocation of the facility, with no operational data at the original address, but the company's registration has been changed (not cancelled). An example is an LNG refueling station relocated from town A to town B; the original address has no GPS data, but the registration has been updated. "Cessation of operation and cancellation" is defined as the permanent cessation of facility operation, with the company's registration cancelled or the regulatory authority issuing a public notice of cessation of operation. An example is a small, non-compliant LNG regasification station whose license has been revoked by the regulatory authority.

[0050] The dynamic attribute update module sets logical combination judgment rules for each status type. The "normal operation" status type uses operational data extracted from the dynamic dataset over the past 3 months, enterprise registration information, and whether the business is closed or deregistered to set corresponding logical combination judgment rules. For example: (Waybill GPS data: ≥1 valid record within 3 months) AND (Enterprise registration: valid, not deregistered) AND (Policy data: no closure / deregistration notice). The "temporary closure" status type uses operational data extracted from the dynamic dataset over the past 3 months, enterprise registration information, and whether the business is closed or deregistered to set corresponding logical combination judgment rules. For example: (Policy data: received short-term closure notice from the Energy Bureau / Environmental Protection Department) OR (Enterprise registration: submitted maintenance registration, specifying resumption time) AND Enterprise registration: not deregistered). Similarly, logical combination judgment rules are set for seasonal shutdown, pending monitoring, relocation pending updates, business closure, and deregistration status types, and the operational status is judged and determined through these logical combination judgment rules. Examples of logical combination judgment rules for seasonal shutdown, pending monitoring, relocation pending updates, business closure, and deregistration status types are shown in Table 4 below. Table 4. Examples of Logical Combination Judgment Rules for Various Operational Status Types

[0051] like Figure 3 As shown, the dynamic attribute update module determines the operational status of LNG facilities in the dynamic dataset according to logical combination judgment rules and records the attribute labels of the operational status in real time. The specific method is as follows: acquire data such as waybill data, enterprise registration information, and policy data from the dynamic dataset; extract the feature data (including valid waybill records, registration status, policy notices, resumption time, and historical patterns) of the dynamic dataset according to logical combination judgment rules; then determine the operational status according to priority matching logical combination judgment rules; and update the attribute labels in the operational status attribute labels (finally, the LNG facility database records the updated operational status attribute labels in real time).

[0052] In some embodiments, the activity level in the final LNG facility database includes four activity types: high activity, medium activity, low activity, and extremely low activity. High activity is defined as high operating frequency and stable transportation volume, indicating it is a core facility for regional LNG supply and demand. Medium activity is defined as moderate operating frequency and small fluctuations in transportation volume, capable of stably supporting local supply and demand. Low activity is defined as low operating frequency and relatively low transportation volume, but with sustained demand. Extremely low activity is defined as extremely low operating frequency and scattered transportation volume, with poor demand stability. The dynamic attribute update module sets the activity state variable combination judgment rules corresponding to each activity type. These rules are constructed hierarchically according to activity type and facility subtype. Variables in the activity state variable combination judgment rules include minimum docking frequency, average monthly minimum transportation volume, minimum transportation volume for low activity, and transportation volume variation coefficient. The expression for the transportation volume variation coefficient (CV) is as follows: , , Let be the transportation volume for month i, and n be the total number of months for which statistics are being compiled. The average transport volume is given, and SD represents the standard deviation of monthly transport volume (reflecting the absolute fluctuation range). Different LNG facilities or facility subtypes have corresponding variable combination judgment rules. These rules are combinations of the thresholds corresponding to the aforementioned variables. Examples of variable combination judgment rules are shown in Table 6.

[0053]

[0054] like Figure 4 As shown, the dynamic attribute update module determines the activity level of LNG facilities in the dynamic dataset according to the active state variable combination judgment rules and records the attribute labels of the activity level in real time. The specific method is as follows: acquire the dynamic dataset (including LNG facility change data and electronic waybill data), extract core variables (including 3-month docking frequency, average monthly transport volume, CV value, and facility subtype), identify loading class (including LNG receiving terminal and LNG plant) and unloading class (including gas power plant, transportation gas, industrial gas and city gas) from the facility subtype, and classify and determine the activity level. Update the attribute labels in the activity level attribute labels (finally, the LNG facility database records the updated activity level attribute labels in real time).

[0055] S5. Construct a multi-scenario in-depth analysis module. This module includes a 3D analysis model unit, which is used to monitor and display the distribution, operational status, activity level, and LNG facility transportation volume in real time on a geographic map, with LNG facilities as points of interest (POIs). For example, it can display the distribution of LNG facility status and LNG facility transportation volume within a research area (e.g., China), and can further obtain statistical data on the transportation volume of highly active LNG facilities. Figure 5 As shown; LNG facilities can be categorized, for example, LNG receiving terminals located in coastal port cities supply imported LNG, while LNG plants located in non-port cities supply domestic LNG. Figure 6 The study area (e.g., China) shows the annual LNG receiving terminal import volume, or Figure 7 The chart shows a comparison of LNG loading and transportation volumes for adjacent years within a study area (e.g., China). By statistically analyzing LNG facilities in loading and unloading areas within the study area (e.g., China) and combining this with LNG flow data from centralized electronic waybills in dynamic datasets, the following can be obtained: Figure 8 The diagram shows the LNG loading and unloading transport flow of the top 10 loading and unloading areas in the study region, and the results obtained... Figure 9 The image shows a comparison of LNG facilities in the top 15 regions by LNG loading volume within the study area. The 3D analysis model unit statistically analyzes historical supply and demand data according to the facility type and uses a bidirectional long short-term memory network to predict future supply and demand data and supply and demand gaps (e.g., predicting the supply and demand gap for the study area for 1-3 months). Example: A predicted industrial gas shortage of 50,000 tons in a province in December 2024 → Prioritizing resource allocation to a nearby high-activity receiving terminal. This embodiment uses the need to predict the industrial gas demand of a province (e.g., Guangdong Province) in 2025 for procurement planning as an example; the implementation process is as follows: (1) Data input: Dynamic data of high / medium active industrial gas terminal POI in Guangdong Province, transportation volume in the same period of 2023-2024, and meteorological forecast data for Q1 2025; (2) Model operation: Call the supply and demand gap prediction model (learning rate 2e-5, batch size 16). (3) Results Output: The forecast for industrial gas demand in Guangdong Province in 2025 is 2.35 million tons, with a shortage of 80,000 tons. → It is recommended to prioritize the procurement of resources from Zhuhai Jinwan and Eastern Guangdong LNG receiving terminals. Implementation Results: Enterprise procurement plans are accurately matched with demand, avoiding inventory backlog or supply shortages, and reducing procurement costs by 12%.

[0056] The three-dimensional analysis model unit performs optimal scheduling route matching and planning according to the loading and unloading types of LNG facilities (i.e., the optimal solution among "loading point - transportation route - unloading point"). This embodiment can also re-match the optimal backup assembly LNG facility resources for shut-down assembly LNG facilities (suppliers) according to the corresponding needs of unloading LNG facilities. It can also filter "normal operation + high / medium activity" POIs and combine them with GPS trajectory to plan the optimal route; in this way, transportation companies can effectively avoid points of concern / shutdown, reduce empty running rate by 20%, and shorten the single-trip transportation time by 15%.

[0057] In some embodiments, the multi-scenario deep analysis module further includes a trade matching module unit. The trade matching module unit extracts the LNG facility type for loading and evaluates its loading capacity, activity level, and regional distribution, designating the loaded LNG facilities as trade suppliers. The trade matching module unit also extracts the LNG facility type for unloading and evaluates its unloading demand, activity level, and regional distribution, designating the unloading LNG facilities as trade demanders. The trade matching module unit constructs a supply-demand matching matrix between trade suppliers and trade demanders and sets the shortest supply-demand transportation distance as the optimal supply-demand matching matrix. This allows the module to recommend supply and sales paths with "highly stable supply + low transportation costs" to trading companies, improving supply-demand matching efficiency by 30%. This embodiment uses a cold wave in a certain month of 2024 as an example, predicting a gas shortage of 30,000 tons for gas-fired power plants in a province such as Shandong, requiring emergency resource allocation. The implementation process is as follows: (1) Real-time monitoring: Screening LNG receiving terminals in Shandong Province and surrounding areas that are "normally operating - highly active" (Tianjin receiving terminal, Qingdao receiving terminal); (2) Route planning: The AI ​​model generates the optimal routes for "Tianjin receiving station → Jinan gas power plant" and "Qingdao receiving station → Yantai gas power plant". (3) Emergency dispatch: Coordinate with transportation companies to allocate 200 LNG tank trucks for transportation according to the planned route.

[0058] (4) Implementation results: The gap was filled on time, and there was no interruption of gas power plant supply, ensuring the stability of regional heating.

[0059] In some embodiments, the multi-scenario deep analysis module further includes an operation and maintenance supervision module unit. This module unit is used to dynamically add, remove, and adjust data on trade suppliers and demanders in the supply and demand matching matrix, and re-match the optimal supply and demand matching matrix. The module unit also extracts LNG electronic waybill data from the dynamic dataset for identification and early warning of transportation non-compliance; monitors the POI's operational status in real time; and verifies transportation compliance (such as whether transportation exceeds the permitted scope) by linking waybill data, improving the enforcement efficiency of regulatory authorities by 40% and focusing on safety hazards of low-activity / problem-prone facilities. This embodiment can also construct a corporate credit assessment model based on POI activity levels, trade flows, and price data; increasing the credit limit for enterprises associated with highly active facilities by 30% and reducing the risk for financial institutions.

[0060] This embodiment's multi-scenario deep analysis module utilizes a dynamic dataset (including dynamic POI data) to achieve accurate model adaptation. The method includes: (1) Dataset preprocessing: 1. Feature Engineering: Integrate dynamic datasets (status, activity level) with waybill, geographic, and market data to generate feature vectors (POI type, administrative region, activity level, transportation volume in the past 3 months, average transportation distance, unloading point type, supply and demand gap prediction, etc.). 2. Data standardization: Unify coding rules (administrative division code, facility type code) to generate a sample dataset with 480,000+ labeled samples (including 380,000+ training set, 48,000+ validation set, and 48,000+ test set).

[0061] (2) Layered application scheme for multi-scenario deep analysis module. Some examples of the layered application scheme are shown in Table 7 below:

[0062] (3) Closed-loop iteration mechanism: POI dynamically updates data and feeds it back to the model regularly. The model is fine-tuned once a quarter, realizing a self-loop of "data update → model iteration → application optimization".

[0063] This invention achieves six core objectives through hierarchical analysis and fusion processing of multi-source data, combined with the characteristics of the LNG field: First, a classification system for LNG facilities is established to accurately identify LNG loading and unloading facilities and classify site types, addressing the issue of low address resolution accuracy. Second, precise location of LNG facility Point of Interest (POI) sets is achieved by combining electronic waybill data and map reverse engineering interfaces to complete fuzzy addresses and define a unified coordinate system for site names, ensuring the accuracy of LNG facility POI geographical locations. Third, a dynamic update mechanism for LNG facility POIs is established, marking facility operational status and activity levels to form a four-dimensional dynamic dataset of "location-type-status-activity level," addressing the issue of static data lag. Fourth, a high-quality dataset dedicated to large-scale LNG vertical models is constructed to support accurate model adaptation in scenarios such as supply and demand forecasting and route optimization. Fifth, a real-time analysis and pre-prediction system is established for scenarios such as heating and supply assurance and peak summer demand, enabling the prediction and intelligent scheduling of regional supply and demand gaps. Sixth, a closed-loop empowerment system of "dataset-model-application" is formed, providing one-stop solutions for multiple scenarios such as trade matching, transportation optimization, regulatory enforcement, and financial empowerment.

[0064] This embodiment can also provide customized service outputs for various departments. Based on the dynamic dataset, it can also provide customized services as shown in Table 8 below:

[0065] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent analysis of LNG facility data that integrates multidimensional data and POI dynamic data, characterized in that: The methods include: S1. Construct an LNG facility thesaurus. The LNG facility thesaurus has a hierarchical structure. The first level of the LNG facility thesaurus includes facility keyword attributes, geographic location mapping attributes, operating entity association attributes, and dynamic attributes. The second level of facility keyword attributes includes the full name of the facility, the abbreviation of the facility, the facility type, and the facility subtype. The facility type includes loading and unloading. The facility subtypes are further subdivided according to loading and unloading types. The second level of geographic location mapping attributes includes administrative division code, administrative region, and geographic identifier. The geographic identifier includes geographic location information and geographic coordinate information. The second level of operating entity association attributes includes the operating entity, the investment entity, and related attributes. The second level of dynamic attributes includes the operating status and activity level. S2. Obtain and screen the list of core LNG facilities and organize it according to the LNG facility thesaurus to obtain a preliminary LNG facility database. Use big data models to obtain LNG facility data to supplement and optimize the preliminary LNG facility database. S3. Collect and acquire LNG electronic waybill data to supplement and optimize the preliminary LNG facility database to obtain the final LNG facility database; S4. Construct a dynamic attribute update module to obtain a dynamic dataset associated with the final LNG facility database. The dynamic dataset includes LNG facility change data and LNG electronic waybill data. The dynamic attribute update module dynamically adjusts the attribute information of the operating status and activity level in the final LNG facility database. S5. Construct a multi-scenario deep analysis module, which includes a 3D analysis model unit. The 3D analysis model unit is used to monitor and display the distribution, operating status, activity level, and LNG transportation volume of LNG facilities in real time on a geographic map, with LNG facilities as the point of interest. The 3D analysis model unit performs historical time-series data statistics on supply and demand according to the facility type of LNG facilities and uses a bidirectional long short-term memory network to predict supply and demand data and supply and demand gaps for future time periods. The 3D analysis model unit performs optimal scheduling route matching and planning according to the loading and unloading types of LNG facilities.

2. The intelligent data analysis method for LNG facilities that integrates multidimensional data and POI dynamic data according to claim 1, characterized in that: The multi-scenario in-depth analysis module also includes a trade matching module. The trade matching module extracts the LNG facility type for loading and evaluates loading capacity, activity level, and regional distribution, and treats the loaded LNG facilities as trade suppliers. The trade matching module extracts the LNG facility type for unloading and evaluates unloading demand, activity level, and regional distribution, and treats the unloading LNG facilities as trade demanders. The trade matching module constructs a supply and demand matching matrix between trade suppliers and trade demanders and sets the shortest supply and demand transportation distance as the optimal supply and demand matching matrix.

3. The intelligent data analysis method for LNG facilities that integrates multidimensional data and POI dynamic data according to claim 2, characterized in that: The multi-scenario in-depth analysis module also includes an operation and maintenance supervision module unit. The operation and maintenance supervision module unit is used to dynamically add or remove data on the trade suppliers and trade demanders in the supply and demand matching matrix and rematch the optimal supply and demand matching matrix. The operation and maintenance supervision module unit extracts LNG electronic waybill data from the dynamic dataset to identify and warn of transportation non-compliance.

4. The intelligent data analysis method for LNG facilities that integrates multidimensional data and POI dynamic data according to claim 1, characterized in that: In method S2, the big data model obtains LNG facility data from the Internet platform and performs attribute matching with the LNG facilities included in the preliminary LNG facility database and completes the attribute information. If the LNG facility data contains LNG facilities not included in the preliminary LNG facility database, a list of LNG facilities to be supplemented is formed, and the preliminary LNG facility database is supplemented with LNG facilities from the list of LNG facilities to be supplemented.

5. The intelligent data analysis method for LNG facilities that integrates multidimensional data and POI dynamic data according to claim 1, 2, or 4, characterized in that: The facility keyword attributes in the preliminary LNG facility database are processed using high-frequency related words obtained from LNG facility data through a big data model. The method includes: extracting high-frequency related words from the LNG facility data using the TextTank algorithm, setting a word frequency threshold ≥5 and a mutual information threshold ≥0.8; calculating the semantic cosine similarity between the high-frequency related words and the facility keyword attributes in the preliminary LNG facility database using a pre-trained BERT-Large model, and using high-frequency related words with a semantic cosine similarity ≥0.85 as related word tag information for the facility keyword attributes; and adding related word tag information for the geographic location mapping attribute and / or the operating entity association attribute according to the above method.

6. The intelligent data analysis method for LNG facilities that integrates multidimensional data and POI dynamic data according to claim 1, characterized in that: In method S3, the geographic location data obtained from the LNG electronic waybill data is labeled and completed for the geographic location mapping attributes in the preliminary LNG facility database. The method includes: performing word segmentation processing on the geographic location nodes in the geographic location data of the LNG electronic waybill data, and performing data completion and standardization processing according to three levels: administrative region, road or park, and location feature words. The administrative region includes the three levels of province, city, and county or district.

7. The intelligent data analysis method for LNG facilities that integrates multidimensional data and POI dynamic data according to claim 1, characterized in that: In method S3, data optimization includes keyword clustering. The K-Means clustering algorithm is used in conjunction with TF-IDF feature weighting to cluster the semantically split keywords. The facility type clustering includes loading and unloading classes. The loading and unloading classes are set with subtypes at different levels. The subtypes of the loading class include LNG receiving terminals and LNG plants. The subtypes of the unloading class include gas-fired power plants, transportation gas, industrial gas, and city gas. The subtypes of the loading class correspond to the facility subtypes of the loading subclass, and the subtypes of the unloading class correspond to the facility subtypes of the unloading subclass.

8. The intelligent data analysis method for LNG facilities that integrates multidimensional data and POI dynamic data according to claim 1, characterized in that: Before obtaining the final LNG facility database, data standardization processing is carried out. Data standardization processing includes coordinate unification processing and address normalization processing. Geographic coordinate information in the geographic identifier is uniformly converted to coordinate information in the WGS-84 coordinate system. Address normalization processing is carried out according to three levels: administrative region, road or park, and location characteristic words.

9. The intelligent data analysis method for LNG facilities that integrates multidimensional data and POI dynamic data according to claim 1, characterized in that: In method S4, the final LNG facility database includes operational status types such as normal operation, temporary shutdown, renovation and upgrade, seasonal shutdown, pending attention, relocation pending update, suspension of business, and deregistration. The dynamic attribute update module sets the logical combination judgment rules corresponding to each status type. The dynamic attribute update module judges the operational status of the LNG facility in the dynamic dataset according to the logical combination judgment rules and records the attribute labels of the operational status in real time.

10. The intelligent data analysis method for LNG facilities that integrates multidimensional data and POI dynamic data according to claim 1, characterized in that: In method S4, the activity level in the final LNG facility database includes high activity, medium activity, low activity, and very low activity. The dynamic attribute update module sets the activity state variable combination judgment rules corresponding to each activity type. The activity state variable combination judgment rules are constructed according to the upper and lower levels of activity type and facility subtype. The dynamic attribute update module judges the activity level of LNG facilities in the dynamic dataset according to the activity state variable combination judgment rules and records the attribute labels of the activity level in real time.

Images