A Spatial Data Intelligent Analysis and Decision Support Method and System Based on Large Language Model

Abstract

This invention relates to the field of machine learning technology, specifically to a method and system for intelligent spatial data analysis and decision support based on a large language model. The method includes the following steps: acquiring spatial entities, location phrases, and task descriptions; analyzing word order and modification relationships; extracting task elements and action sequences; constructing continuous semantic paths; identifying geographic terms and range changes; extracting statements whose angle between action instruction direction vectors is less than a preset threshold; and obtaining a set of spatial action structures. In this invention, a semantic task list is constructed by extracting spatial entities, location phrases, and task content. Actions are ordered by combining action words and word order relationships. The connection methods between segments are analyzed, and a coherent task path is organized. The sequential matching and semantic connection between spatial objects and actions are completed. By filtering the information structure of action direction and location changes, task statements are arranged in an orderly manner according to the process, avoiding ambiguity and interruptions in instruction transmission, and maintaining the orderly integration and flow of spatial information.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a method and system for intelligent analysis and decision support of spatial data based on large language models. Background Technology

[0002] Machine learning technology involves learning, classifying, predicting, and reasoning about data through algorithmic models. Its core aspects include neural network architecture design, parameter training methods, feature representation learning, contextual modeling, and multimodal information fusion. This technology covers multiple levels from shallow perceptual computing to deep semantic understanding and is widely applied in natural language processing, image recognition, intelligent recommendation, and automated decision-making. Driven by the development of deep learning, large language models have become an important foundation for handling complex language expression and understanding tasks. They possess large-scale knowledge modeling and contextual reasoning capabilities, and can uniformly model heterogeneous data such as text, speech, and images, outputting structured information. Traditional spatial data intelligent analysis and decision support methods refer to processing vector and raster data using geographic information system tools in spatial scenarios such as urban planning, disaster early warning, and resource allocation to generate auxiliary decision-making information. Traditional methods typically involve users manually setting operation parameters such as buffer analysis and overlay analysis, calling spatial analysis function libraries for layer processing, generating map results with the help of visualization plugins, and then having professionals write analysis texts or make recommendations based on the results.

[0003] Existing technologies rely excessively on manually configured spatial analysis operations, lacking the ability to automatically identify and semantically interpret spatial entities in complex natural language. When dealing with multiple spatial objects, action words, and modifiers, existing methods are prone to spatial relationship analysis biases, especially in scenarios with complex spatial task instruction structures. This can easily lead to inconsistencies between spatial objects and actions, thus affecting the final decision-making outcome. When faced with multi-layered semantic relationships, traditional methods cannot effectively integrate this information, resulting in a lack of smooth integration between spatial data analysis and task execution. For example, in disaster emergency response, the priority of tasks or the order of resource allocation may not be correctly identified, increasing decision-making delays and execution difficulties. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a spatial data intelligent analysis and decision support method based on a large language model, comprising the following steps:

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a spatial data intelligent analysis and decision support method based on a large language model, comprising the following steps:

[0006] S1: Obtain spatial entities, location phrases and task descriptions from the input statement, locate spatial operation sentences, distinguish the first and second occurrences of entities, analyze the relationship between action words and modifiers, split sentences according to task content, and obtain a list of spatial semantic instruction task items.

[0007] S2: Based on the list of spatial semantic instruction task items, filter statements that point to the same spatial entity, analyze the word order relationship of direction, position and range, determine the language connection before and after the action word, and obtain the task action sorting result set;

[0008] S3: Based on the task action sorting result set, identify the language structure of actions and spatial entities, determine the word order relationship between modifiers and action statements, extract continuous semantic structure, and obtain a set of continuous instruction paths for spatial tasks;

[0009] S4: Based on the set of continuous instruction paths for the spatial task, analyze the language connection between space and action, filter out paths with varying ranges and coherence, and obtain a set of spatial task hierarchical jump paths.

[0010] S5: Based on the spatial task hierarchical jump path set, identify the connection order between actions and spatial objects, determine the direction of actions and the order of objects, extract statements whose angle between the direction vectors of action instructions is less than a preset threshold, and obtain a spatial action structure set.

[0011] As a further aspect of the present invention, the spatial semantic instruction task item list includes spatial entity names, location phrases, task description phrases, operation sentence annotations, entity first and repeat markers, action modifiers, action-location relationships, and task statement divisions. The task action sorting result set includes entity pointing markers, direction and range language, structural boundary relationships, and semantic connection methods. The spatial task continuous instruction path set includes associated entity phrases, structural location differences, modification and word order relationships, action connection components, and unified paragraphs. The spatial task hierarchical jump path set includes spatial range terms, semantic direction order, location-action connection methods, range changes, and connecting content. The spatial action structure set includes action objects, directional term positions, connection order, and directional consistent action flows.

[0012] As a further aspect of the present invention, the spatial operation sentence refers to the statement relating spatial entities and action behaviors, and the linguistic expression of position movement, position transformation and spatial task execution;

[0013] The aforementioned linguistic cohesion refers to the logical connection in linguistic order between differentiated action words, modifiers, and spatial entities in a sentence, which is reflected in the naturalness of semantic transition and the rationality of word order.

[0014] As a further aspect of the present invention, the spatial entity language structure refers to the linguistic relationship between the grammatical position of the spatial entity in a sentence and the action word, including subject, object, and positional components;

[0015] The connecting word order refers to the order and connection between actions, directional words and spatial objects in language.

[0016] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0017] S101: Obtain the spatial entity name, location phrase and task description phrase in the input statement, locate the sentences associated with spatial operations, and obtain the set of sentences with spatial operation semantics;

[0018] S102: Based on the set of sentences with the aforementioned spatial operation semantics, distinguish the first and second occurrences of repeated entities, locate the position of each entity in the sentence, and obtain spatial entity position information;

[0019] S103: Based on the spatial entity location information, identify the relationship between action words and modifiers, associate action words with target objects, and obtain a list of spatial semantic instruction task items.

[0020] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0021] S201: Based on the spatial semantic instruction task item list, identify the content in the statement that points to the same spatial entity, extract the phrases in each instruction that are associated with spatial direction, position and range, analyze the position of the phrases in the sentence, and determine their role in the task to obtain the direction, position and range phrase information.

[0022] S202: Based on the directional position and range phrase information, analyze the phrase order in the sentence, identify the order and position of their appearance in the sentence, decompose the connection relationship between action words and phrases, and obtain the spatial task action sequence structure;

[0023] S203: Based on the spatial task action sequence structure, analyze the connection relationship between action words and modifiers, calculate the semantic similarity of adjacent sentences and compare it with a preset threshold to obtain a set of task action ranking results.

[0024] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0025] S301: Based on the task action sorting result set, identify spatial entity phrases associated with action content, obtain the position of each action and corresponding entity in the sentence, determine the grammatical position of the entity in the differentiated sentence structure, and obtain the linguistic positional relationship between the action and the spatial entity.

[0026] S302: Based on the linguistic positional relationship between the action and the spatial entity, analyze the word order relationship between the modifiers and entity phrases before and after the action, and analyze the logical sequence relationship between adjacent actions according to the preset spatiotemporal knowledge graph, determine whether the action words before and after are connected to each other, and obtain the connection relationship between the action and the modifier.

[0027] S303: Based on the connection relationship between the action and the modifier, identify the connecting components in the adjacent action language structure, filter the segments where the subject and object have a coreference relationship, and advance the semantic extension according to the action sequence to obtain the set of continuous instruction paths for the spatial task.

[0028] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0029] S401: Based on the continuous instruction path set of the spatial task, identify words in the path that include descriptions of geographic spatial range, locate the position of the words in the sentence, and connect them with the associated actions in grammatical order to obtain the spatial range description and action connection relationship;

[0030] S402: Based on the spatial range description and action connection relationship, analyze the order of spatial position words before and after the statement, determine the connection mode between differentiated spatial ranges, and filter the language transition from local to whole in the path to obtain spatial range jump information;

[0031] S403: Based on the spatial range jump information, identify the continuous semantics generated by the change of spatial range in the path, filter the paths that form continuous jumps at the spatial level, and obtain the set of spatial task level jump paths.

[0032] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0033] S501: Based on the set of spatial task hierarchical jump paths, locate the spatial object pointed to by each action, and determine the order of actions and the spatial objects pointed to by actions according to the positional relationship between actions and spatial objects in the statement, so as to obtain the correspondence between actions and spatial objects.

[0034] S502: Based on the correspondence between the action and the spatial object, identify the word in the action that indicates directional movement, determine the position of the word in the sentence, compare the connection between the word and the spatial object, and obtain the connection relationship between the action direction and the object;

[0035] S503: Based on the connection relationship between the action direction and the object, extract statements whose angle between the action instruction direction vectors is less than a preset threshold, and process them according to the language connection order between the spatial object and the action direction to obtain a spatial action structure set.

[0036] This invention also provides a spatial data intelligent analysis and decision support system based on a large language model, comprising:

[0037] The spatial sentence recognition module is configured to acquire spatial entity names, location phrases and task description phrases in the input sentence, locate sentences expressing spatial operations, distinguish the first and second occurrences of entities, analyze the relationship between action words and modifiers, split sentences according to task content, and obtain a list of spatial semantic instruction task items.

[0038] The action element parsing module is configured to filter statements pointing to the same spatial entity based on the list of spatial semantic instruction task items, identify the word order relationship between directional words, position words and range words in the statements and entity phrases, determine the language connection before and after the action words, and obtain a set of task action ranking results.

[0039] The semantic path connection module is configured to identify the connection content between action language segments based on the task action sorting result set, determine the word order relationship between modifiers and action statements, extract continuous semantic structure, and obtain a set of continuous instruction paths for spatial tasks.

[0040] The spatial scope jump module is configured to identify phrases representing geographical scope in statements based on the continuous instruction path set of the spatial task, analyze the linguistic connection between space and action, identify whether the action crosses spatial levels in the language, filter coherent paths where the scope changes at different levels, and obtain a set of spatial task level jump paths.

[0041] The task flow output module is configured to locate the spatial object phrases that perform actions based on the set of spatial task hierarchical jump paths, identify the position of directional words in the sentence structure, analyze the word order connection between the action structure and the spatial object, extract sentences with an angle of less than a preset threshold, and obtain a set of spatial action structures.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0043] In this invention, a semantic task list is constructed by extracting spatial entities, location phrases and task content. The action order is completed by combining action words and word order relationships. The connection between segments is sorted out, and a coherent task path is formed. The sequential matching and semantic connection between spatial objects and actions are completed. By filtering the information structure of action direction and position change, the task statements are arranged in an orderly manner according to the process, avoiding ambiguity and interruption in instruction transmission, and maintaining the orderly integration and flow of spatial information. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the steps of the present invention.

[0046] Figure 2 This is a detailed schematic diagram of S1 of the present invention.

[0047] Figure 3 This is a detailed schematic diagram of S2 in this invention.

[0048] Figure 4 This is a detailed schematic diagram of S3 of the present invention.

[0049] Figure 5 This is a detailed schematic diagram of S4 of the present invention.

[0050] Figure 6 This is a detailed schematic diagram of S5 of the present invention.

[0051] Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0052] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0053] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0054] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0055] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0056] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0057] Please see Figure 1 This invention provides a spatial data intelligent analysis and decision support method based on a large language model, comprising the following steps:

[0058] S1: Obtain the spatial entity names, position phrases and task description phrases in the input statement, locate the sentences of spatial operations, distinguish repeated entities according to the position of their first and second occurrences, identify the modifiers and positional relationships corresponding to the action words, divide the task target statements according to the language collocation between the action and the task content, and obtain a list of spatial semantic instruction task items.

[0059] S2: Based on the list of spatial semantic instruction task items, identify the content in the sentences that points to the same spatial entity, locate the language content in each sentence that describes direction, position and range, analyze the boundary relationship between language structures according to the order and position features of phrases in the sentence, process the semantic order according to the language connection method before and after the action word, and obtain the task action sorting result set.

[0060] S3: Based on the task action sorting result set, extract the spatial entity phrases associated with the action content, identify the position of the entity in the differentiated language structure, make a forward language transition judgment on the word order relationship between the modifiers and entity phrases before and after the action, and identify the connecting components in the adjacent action language structure. Merge the segments that can form continuous semantics into a unified structure to obtain the set of continuous instruction paths for spatial tasks.

[0061] S4: Based on the continuous instruction path set of spatial tasks, identify the geospatial range terms that appear in the path, judge the sentence connection between terms and action words according to the semantic direction from local to global, compare and analyze the spatial location words and action sentences that appear before and after the sentences, identify the content in the path where the spatial range changes and the language connection is continuous, and obtain the spatial task level jump path set.

[0062] S5: Based on the spatial task hierarchical jump path set, locate the spatial object that each action acts upon, identify the position of the word in the action that represents directional movement, determine the connection order between the word and the spatial object in the sentence, process the semantic connection between the action direction and the object position in the language, extract sentences where the angle between the action instruction direction vectors is less than a preset threshold, and obtain the spatial action structure set.

[0063] The list of spatial semantic instruction tasks includes spatial entity names, location phrases, task description phrases, operation sentence annotations, entity first and repeat markers, action modifiers, action-location relationships, and task statement divisions. The task action sorting result set includes entity pointing markers, directional and range language, structural boundary relationships, and semantic connection methods. The spatial task continuous instruction path set includes associated entity phrases, structural and positional differences, modifier and word order relationships, action connection components, and unified paragraphs. The spatial task hierarchical jump path set includes spatial range terms, semantic direction order, location-action connection methods, range changes, and connecting content. The spatial action structure set includes action objects, directional term positions, connection order, and directional consistent action flow.

[0064] Please see Figure 2 The specific steps of S1 are as follows:

[0065] S101: Obtain the spatial entity name, location phrase and task description phrase in the input statement, locate the sentences associated with spatial operations, and obtain the set of sentences with spatial operation semantics;

[0066] After obtaining the spatial entity names, location phrases, and task description phrases from the input text, the text needs to be segmented according to natural pauses such as periods and commas. Each sentence should be examined to see if it contains nouns with spatial properties, such as "under the bridge area," "south entrance," and "outside the fence." By comparing the repetition rate of these phrases in different sentences, representative spatial entity names can be identified. Further analysis is needed to subdivide the subject and object positions of verbs in each sentence to determine if the action is related to the identified spatial entities. If the action expresses a change in location or contact with an area, such as "moving to," "approaching," or "crossing," the sentence is considered to have a spatial operation relationship. Subsequently, the task description phrases extracted should focus on the object following the action word and its modifiers, such as "complete the monitoring action" or "perform cleaning." Phrases such as "management work" and "deploying sensing equipment" can be directly used as task descriptions. By combining the location of spatial entities with the semantic orientation relationship formed by the action words, the task intent content corresponding to the spatial action sentences can be listed. If the same action word and spatial entity combination appears, it is necessary to check whether the task associated with the combination in different sentences is consistent. For example, although the action words are the same in "cleaning the inside of the passage" and "cleaning the area under the bridge", the spatial location is significantly different. They need to be divided into independent spatial task items. During the execution process, the comparison method can be used to distinguish the degree of proximity between spatial phrases. If two spatial phrases have the same root word but significantly different modifiers, a similarity threshold such as 0.6 can be set to distinguish phrases with directional tendencies such as "inside", "east", and "edge". Finally, a set of sentences with spatial operation semantics is obtained.

[0067] S102: A collection of sentences based on spatial operation semantics, distinguishing the first and second occurrences of repeated entities, locating the position of each entity in the sentence, and obtaining spatial entity position information;

[0068] First, extract the spatial entity phrases appearing in each sentence sequentially, count the sentence number of their first appearance, and record their character index position in the sentence. For example, in the sentence "move to the south side of the warehouse and then conduct the inspection," the first appearance of the entity "south side of the warehouse" starts at the 5th character position. If this spatial entity appears again in the subsequent sentence "transport the equipment out of the south side of the warehouse," it is necessary to determine whether the phrase is completely identical to the previous one. If it is identical, it is recorded as a duplicate entity, and its period and position index are marked. Then, according to the order of verbs in the sentence, locate the semantic relationship between verbs and entities, and determine whether there is any usage of word order fronting, backing, or nesting. For example, "enter from the south side of the warehouse" and "enter the south side of the warehouse" have the same semantics but different structures. In the processing, such verbs should be classified as verbs. The word position and entity position are combined and used as a reference for entity localization. Then, all marked spatial entity phrases in the sentences are traversed and compared. For identical phrases, they are merged and numbered. If there are cases where only the direction words are different but the location names are the same, such as "east side of the warehouse" and "west side of the warehouse", a similarity judgment interval needs to be set. If the names are the same, the threshold is set to 0.8. If the direction difference is obvious, they are classified as different entities. Furthermore, the directional verb terms can be combined to determine whether the action has spatial movement behavior. For example, the semantic combination between "driving in", "approaching", "exiting" and other spatial entities can be used to confirm the semantic position of the entity. Finally, the position, role and combination relationship of each entity in the sentence with the action words are organized into position data items to obtain the spatial entity position information.

[0069] S103: Based on spatial entity location information, identify the relationship between action words and modifiers, associate action words with target objects, and obtain a list of spatial semantic instruction task items;

[0070] First, each located spatial entity phrase in the sentence is read, and its index range in the sentence is extracted. Using this index as a boundary, semantically related action words are retrieved forward or backward. If the action word precedes the entity phrase and the interval is less than five words, the action is considered an actively triggered behavior. For example, in "cleaning the equipment on the west side of the operating area," "cleaning" precedes "equipment on the west side of the operating area," so it should be identified as the controlling action word of this spatial entity. If the action word appears after the entity phrase, it needs to be determined whether it is a passive modifier. For example, in "equipment on the west side of the operating area was cleaned," "was cleaned" constitutes a passive combination. During execution, when there are multiple action words in the sentence, it is necessary to determine whether adjacent action words share the same spatial entity. For this purpose, a similarity reference threshold can be set. When the frequency of multiple action words combined with the same spatial entity is higher than 0.7, the entity can be considered as the common target object of multiple action words. Then, the semantic type of the action word is judged. If a word contains directional modifiers, such as "advance," "move in," or "lift out," it needs to be compared with possible directional words in the spatial entity. If the directions are consistent, it is recorded as a consistent action pair. Based on this, the attribution of modifiers needs to be addressed, i.e., whether the modifier directly affects the meaning of the action word or the boundary judgment of the spatial entity. For example, in "slowly move into the east side of the warehouse," "slowly" and "move in" form a modifier relationship. "Slowly" and "move in" should be bound together, and then the corresponding spatial entity "east side of the warehouse" should be determined. To verify the consistency of the operation, the statement "quickly move the equipment to the inspection room" can be constructed. In this statement, "quickly" modifies "move to," and "inspection room" is the target entity. This establishes a three-item combination relationship. Finally, the combination of the action word, modifier, and target spatial entity in each statement is extracted, and the pairing data between the action and the entity is output to obtain a list of spatial semantic instruction task items.

[0071] Please see Figure 3 The specific steps of S2 are as follows:

[0072] S201: Based on the list of spatial semantic instruction task items, identify the content in the statement that points to the same spatial entity, extract the phrases in each instruction that are associated with spatial direction, position and range, analyze the position of the phrases in the sentence, and determine their role in the task to obtain the information of direction, position and range phrases.

[0073] First, each statement is read line by line, and the spatial entity names mentioned are extracted. Spatial entities appearing twice or more are uniformly named and marked with the same number. Then, spatial entities with the same number are compared and analyzed in the context of different statements to determine whether the statements they occupy express spatial attributes such as direction, location, or range. If so, phrases related to these spatial attributes are extracted, and the starting position and length of the words in the sentences are recorded. For spatial direction phrases, such as "left side," "north," or "facing outwards," the distance parameter between them and the action verb in the main clause is determined. If the distance is within five words, the phrase is considered to be associated with the action and possess directional attributes. For spatial location phrases, such as "next to the entrance," "near the window," or "located in the central area," semantic boundary judgment is required based on the sentence segment containing the verb. If the phrase is located before the verb in the sentence and adjacent to the verb... If no connecting words are present, the phrase is presumed to be the spatial starting position. For spatial range phrases, such as "the entire northern area," "part of the operating area," or "outside the equipment area," their modifying structures are decomposed, the regional range words are extracted, and mapped with the entity number to determine whether they are used to limit different regional representations of the same spatial entity. Taking "the entire equipment area" and "the western part of the equipment area" as examples, the two have related entity numbers but different spatial range words. Their roles in the sentence need to be recorded separately and compared with the action words. If the range word is close to the verb, the range phrase plays the role of action constraint range in the task. During the execution process, multiple fields such as the position of the phrase in the sentence, entity number, and spacing between action words need to be used for cross-judgment to ensure that the extraction results have spatial semantic positioning function. Finally, the directional phrases, position phrases, and range phrases in each sentence are filtered and processed to obtain directional position and range phrase information.

[0074] S202: Based on directional and range phrase information, analyze the phrase order in the sentence, identify the order and position of their appearance in the sentence, decompose the connection relationship between action words and phrases, and obtain the spatial task action sequence structure;

[0075] First, extract phrases from each statement. Sequence the starting positions of all directional, positional, and range phrases within the statement, and record their positional relationships within the sentence. For example, when phrases like "go north," "to the edge of the operating area," and "cross the middle passage" appear in sequence, they should be ordered as the first, second, and third phrases according to their actual order of appearance. Then, compare the phrase's position index with the action word index to determine if the phrase precedes or follows the action word. If a directional phrase like "east" precedes the action word "move," it is marked as a preceding guiding phrase. If a range phrase like "the entire working area" follows the action word "scan," it is marked as a following range limiting phrase. This method is used to determine the positional relationships of phrases in each statement, prioritizing those closest to the action word and without interference from other phrases. The phrases are labeled and identified, and the average word distance between them and the action words is calculated. Then, the semantic attributes of the action words are cross-validated with the semantic categories of the phrases before and after them. For example, if the action word is "enter" and the preceding phrase is "pass through the main passage", it can be regarded as a semantic pair with a logical order. If the action word is "view" but the preceding phrase is "approach the machine area", it is necessary to further determine whether it has logical continuity. Then, the position order of all phrases is compared vertically to confirm whether there is a misorder between action words in the same spatial task. If the actions "advance", "turn around", and "stay" appear in the phrases with the sequence numbers 1, 3, and 2 respectively, the action order should be marked as abnormal. Finally, the overall semantic flow chain is identified based on the order of the phrases and the connection relationship of the action words, and the spatial task action order structure is obtained.

[0076] S203: Based on the spatial task action sequence structure, analyze the connection relationship between action words and modifiers, calculate the semantic similarity of adjacent sentences and compare it with a preset threshold to obtain a set of task action ranking results;

[0077] First, the sentence content corresponding to each pair of adjacent action words is extracted. Semantic encoding is performed on the semantic connections between adjacent action words. After obtaining the action words, a word vector model is called to convert each action word into a high-dimensional vector form. For example, the action words "grab" and "move" are mapped to vector A and vector B, respectively. Cosine similarity is calculated for these two vectors using the cosine of the angle between vector A and vector B. The similarity S = cosθ. If S is close to 1, it indicates that the two actions are related in the task flow. This process is then repeated for all action word pairs to form a similarity set between action pairs. Simultaneously, when obtaining the modifiers involved in each pair of adjacent actions, modifier phrases related to the action words are extracted from these sentences, such as "grab slowly" and "move precisely." Adverbs of degree are separated from the modifiers using rule extraction methods and classified into numerical orders according to their level, such as assigning "slowly" a value of 0.3 and "quickly" a value of 0.8. Then, these are compared with... Action word vectors are combined to form expanded action feature vectors. A new similarity calculation is performed between the expanded vectors. In the example, if the similarity S2 between the vectors formed by "slowly grab" and "quickly move" is 0.45, which is lower than the set semantic association threshold T=0.6, it is determined that this action pair does not have a close logical connection in the actual task ranking. Conversely, if S2≥0.6, it is included in the candidate ranking set. The threshold T is set with reference to the average semantic similarity μ and standard deviation σ in the task action combination, and T=μ+0.2σ. For example, if μ=0.5 and σ=0.3, then T=0.56. Then, all action pairs that satisfy S≥T are arranged in order as the task action ranking candidate sequence, and their ranking positions are re-sorted according to their occurrence frequency and average similarity. For example, if action pair A→B appears 3 times with an average similarity of 0.78, while action pair B→C appears 2 times with an average similarity of 0.65, then the preceding order of A→B is retained first, and the task action ranking result set is obtained.

[0078] Please see Figure 4 The specific steps of S3 are as follows:

[0079] S301: Based on the task action ranking result set, identify spatial entity phrases associated with action content, obtain the position of each action and corresponding entity in the sentence, determine the grammatical position of the entity in the differentiated sentence structure, and obtain the linguistic position relationship between the action and the spatial entity.

[0080] First, spatial entity phrases semantically related to the action words are extracted. During extraction, syntactic analysis is used to scan the action statements character by character, identifying phrases with noun and location attributes, such as "main entrance," "left-side building," and "central area." Then, the accurate character position indexes of the action words and their corresponding spatial entities within the sentence are obtained, and their relative order is recorded. Next, the word spacing between the action and the entity is quantified to determine whether it is at the beginning, middle, or end of the sentence, reflecting its positional relationship within the grammatical structure. For example, in "place the device near the main entrance," the spatial entity "main entrance" is located near the action "near." The right side of the word is spaced 1 word, and its semantic structure belongs to the object category. Next, the grammatical positions of the same entity in different sentence structures are classified to determine whether it is always the object of the action. If a spatial entity is found to have changed its subject or object in different sentences, such as "main entrance controller starts" and "controller approaches the main entrance", then the linguistic attachment direction between the spatial entity and the action word needs to be marked and classified. Finally, the semantic direction, positional relationship and linguistic positional attribute between all actions and spatial entities are integrated to establish a linguistic index matching table for each action sentence to obtain the linguistic positional relationship between actions and spatial entities.

[0081] S302: Based on the linguistic positional relationship between actions and spatial entities, analyze the word order relationship between modifiers and entity phrases before and after an action, analyze the logical sequence of adjacent actions according to a pre-set spatiotemporal knowledge graph, determine whether the action words before and after are connected, and obtain the connection relationship between actions and modifiers.

[0082] First, the positions of the action words and their corresponding spatial entity phrases are extracted from each sentence. The sentences are then divided into three parts: "modifiers before the action word," "the action word itself," and "the entity phrase after the action word." For example, in "quickly move the robotic arm to the target point," "quickly" is the modifier, "move" is the action word, and "to the target point" is the entity phrase. The word order of these three parts is extracted from all sample sentences, and the absolute position index of each component in the sentence is recorded. The frequency distribution of the components before and after the action word is statistically analyzed. In all samples, the sentence with the modifier before the action word and the entity phrase after the action word is extracted. The proportion of words followed by the action word is considered significant if it is greater than 0.85. In a sample dataset, if the proportion of words preceded by the action word "move" with modifiers such as "quickly," "slowly," or "gently" is 93%, and followed by spatial entity phrases such as "to the workbench" or "to the designated location" is 90%, then the condition of a significant word order relationship before and after the action word is met. Next, adjacent action word pairs in each sentence are processed, and each pair of action words is extracted and its position in the sentence is marked. For example, "grab" appears in sentence 3, and "place" appears in sentence 4. By comparing the word order relationships and consulting the pre-defined spatiotemporal knowledge graph, it is determined whether there is a semantic correlation. If "grab → place" is a common temporal relationship in the knowledge graph, and this order is consistent with the sentence order in the current task flow, it is recorded as a temporally consistent instance. If there is a conflict between the sentence order and the spatiotemporal knowledge graph, such as "move → grab" being an unrecommended temporal order but arranged in this way in the actual sentence, it is judged as a word order conflict and the conflict point is marked. The proportion of all action word pairs that conform to the knowledge graph order is counted. If the proportion is greater than 0.75, it means that the action arrangement in the sentence basically conforms to common tasks. In one instance, 50 action word pairs were extracted, of which 37 pairs were consistent with the knowledge graph's order, a ratio of 0.74. This does not meet the judgment threshold of 0.75, and further checks are needed to see if there are any modifiers or spatial entity errors in the erroneous action pairs that cause temporal confusion. Further analysis of the semantic and word order coupling relationship of each action word pair is needed to determine whether they are logically coherent actions. For example, "identify → locate", "locate → grasp", "grab → transport" are highly coupled pairs. Their ordering relationship is prioritized and recorded as valid action chains, ultimately obtaining the connection relationship between actions and modifiers.

[0083] S303: Based on the connection relationship between actions and modifiers, identify the connecting components in adjacent action language structures, filter out segments where the subject and object have a common reference relationship, and advance semantic extension according to the action sequence to obtain a set of continuous instruction paths for spatial tasks;

[0084] First, extract the position and type of connecting elements in sentences containing adjacent action words, and identify connecting words or phrases such as "then," "next," "and," "following," and "furthermore." Record these connecting elements sequentially as semantic progression nodes according to the order in which the action words appear. By marking the preceding and following action words connected by the connecting words, construct an action semantic relationship chain. In actual task text, such as "grab the parts, then move them to the assembly area," "then" is a connecting element, corresponding to the action words "grab" and "move" respectively. Mark "then" as a connecting identifier and record its sentence number in the original text. Word order is used to analyze the referential relationship between subject and object in the sentences following the connecting elements. In the sentence "put it into the trough," "it" refers to the "parts" mentioned in the previous sentence. By calling the subject-object phrase position corresponding to the verb, it is determined whether the subject "robotic arm" appears continuously in multiple sentences. If an action is "robotic arm grabs parts," and the subsequent sentence is "put into the designated position," it is determined whether there is a subject omission that semantically points to the aforementioned subject. Words with such subjects or objects that recur in multiple sentences are extracted, and coreference chains are established. If a noun has a recurrence rate of more than 80% in three or more sentences, it is judged as... The semantic continuation process continues according to the order of action words, marking the existence and identifiable semantic progression phrases between each pair of adjacent action words. The density of the connecting elements is calculated as D = number of connecting phrases / number of action pairs. When D ≥ 0.6, the connecting element information is considered sufficient. For example, if 25 pairs of action words are extracted from a description and 15 connecting phrases are identified, then D = 0.6, satisfying the recognition condition. For action pairs that meet the condition, a continuous path is constructed, where each node is an action word, and its corresponding modifiers and spatial entities are recorded. If the action "locates..." If there is a connecting phrase "then" between "grab" and "grab", and the subjects are consistent, then the two actions are grouped into the same continuous path. In the example, in the paragraph "start the robotic arm, identify the target, then locate and grab the object, and then move to the assembly area", the four action words "start → identify → locate → grab → move" are identified in sequence, the connecting components "then" and "next" are extracted, the subject "robotic arm" is marked to appear continuously, and when "object" is used as the object, it is consistent with the target object in "grab" and "move", and finally a complete continuous task action path chain is constructed, and finally the continuous instruction path set of the space task is obtained.

[0085] Please see Figure 5 The specific steps of S4 are as follows:

[0086] S401: Based on the continuous instruction path set of spatial tasks, identify words in the path that include descriptions of geographic spatial range, locate the position of the words in the sentence, and connect them with related actions in grammatical order to obtain the relationship between spatial range description and action connection;

[0087] First, the linguistic components in each path are classified by word class, extracting spatial range words with regional characteristics, such as "outer side of the building," "near the entrance," and "middle section area." Their word order is then marked within the sentence. For each word, its position in the sentence is determined—whether it precedes, is between, or follows the action word. For example, in "walk along the road to the outer side of the building," "outer side of the building" follows "walk," indicating it represents the target area of ​​the action. Then, based on the word order of the action words in the sentence, the spatial range words are linguistically concatenated and compared with their corresponding actions to analyze whether a semantically coherent spatial action combination is formed. During this process, non-geographically oriented range phrases are eliminated, such as "partial area," which, if lacking spatial orientation modification... The embellishments are excluded and not processed. Then, a mapping table based on action relationships is established for all located words. The position of the action word is used as the index field, and the spatial range descriptor is used as the matching value. The mapping combination is arranged in an ordered manner through the position sorting mechanism. Furthermore, the semantic comparison of the connection logic between the action word and the spatial word is performed according to the original sentence order. For example, in the sentence "crossing through the atrium and heading towards the edge of the square", "atrium" and "edge of the square" are respectively connected with "crossing" and "heading towards". Finally, after all semantic matching is completed, the semantic connection relationship group of the action word and the spatial range descriptor is uniformly output as the basis for parsing the subsequent action path instruction sequence, so as to obtain the spatial range description and action connection relationship.

[0088] S402: Based on the spatial scope description and action connection relationship, analyze the order of spatial position words before and after the sentence, determine the connection mode between differentiated spatial scopes, and filter the language transition from local to whole in the path to obtain spatial scope jump information;

[0089] First, extract the spatial location phrases involved in each language path, such as "entrance," "center of the plaza," and "outer perimeter of the building." Arrange the phrases in word order and match them with their corresponding action words according to the order in which the verbs appear. Then, compare the linguistic expression levels between adjacent spatial phrases to determine whether they represent a spatial expansion from the local to the global. For example, in "entering the hall from the entrance," "entrance" corresponds to the local location, while "hall" represents the overall area. By analyzing the word order, determine whether the action "entering" conforms to the logical sequence of moving from a smaller area to a larger one. If the word order is "from A to B," and A is a spatial term indicating a boundary or starting point, while B is a relatively inclusive core location, then this is judged as an expression of ascending spatial hierarchy. This applies to the positional expression in the sentence structure. For fuzzy intervals, such as "walking along the road to its end," it is necessary to determine the semantic level of phrases like "road" and "end." By comparing with a pre-defined spatial lexicon, the hierarchical labels of each word in the spatial division are determined sequentially. Semantic trends that point from the boundary to the core and from the periphery to the center are organized. Then, by analyzing sentence connecting components such as "then," "next," and "next," it is determined whether there is a sequential action logic at the semantic level. Non-sequential expressions or semantically broken sentences are eliminated, and only sentence fragments with spatial continuity and sequential logic are retained. Finally, all sentence connection points that conform to the spatial transition relationship from the local to the whole are summarized, and the changes in spatial position in the corresponding path of the segment are output according to the action sequence to obtain spatial range jump information.

[0090] S403: Based on spatial range jump information, identify continuous semantics in the path caused by changes in spatial range, filter paths that form continuous jumps at the spatial level, and obtain a set of spatial task level jump paths;

[0091] First, identify the spatial location phrase corresponding to each action in the task statement and extract its spatial scope hierarchy. Divide the spatial scope hierarchy into multiple levels, such as "workshop" as level 1, "assembly area" as level 2, and "assembly table" as level 3. Assign a hierarchy label L_i to each spatial entity and record the mapping relationship between actions and spatial entities in each statement. In the example, "assembly area" in "move to assembly area" is labeled L2, and "assembly table" in "place on assembly table" is labeled L3. Then, extract the relationships between all actions. The spatial jump path is identified, and the direction of change of spatial hierarchy in adjacent statements is identified and its jump type is recorded. If the current action involves L2 and the next action involves L3, it is recorded as a lower-level jump; if it jumps from L3 to L1, it is recorded as an upper-level jump. Then, semantic content extraction is performed on all jump paths, extracting the connecting modifiers, conjunctions, and subject-object content in the statements before and after the jump action, and judging whether their semantics are continuous. In the two statements "grab materials in the storage area" and "transfer to the assembly area", "materials" as the object appears in both statements. If a path appears, it is determined to have continuous semantics. Further filtering is performed on path segments that jump at the spatial level but also possess semantic continuity, constructing a set of spatial jump paths. Then, it is determined whether a continuous spatial level jump is formed within the jump path; that is, multiple jump actions exhibit a monotonic change trend at the spatial level. For example, in the path "Warehouse L1 → Workshop L2 → Assembly Area L3 → Assembly Table L4", the level markers increase sequentially, thus it is determined to be a continuous jump path. Conversely, if the path "Warehouse L1 → Workshop L2 → Warehouse L1" appears, it is determined to be non-continuous. Jumping and eliminating paths: In practical applications, the proportion of consecutive jump paths in all paths is counted, and a jump continuity threshold T_c=0.7 is set. If the proportion of consecutive jump paths is not lower than this threshold, the spatial jump structure is considered stable. For example, if there are 30 consecutive jump paths in all 40 spatial jump paths, the proportion is 0.75, which meets the screening criteria. Finally, these 30 paths are retained as candidate paths. The actions in each path and their corresponding spatial levels are arranged in the order of action to obtain the set of spatial task level jump paths.

[0092] Please see Figure 6 The specific steps of S5 are as follows:

[0093] S501: Based on the spatial task hierarchy jump path set, locate the spatial object pointed to by each action, and determine the order of actions and the spatial objects pointed to by actions according to the positional relationship between actions and spatial objects in the statement, so as to obtain the correspondence between actions and spatial objects;

[0094] First, the action words in each semantic path are extracted, and the linguistic structures before and after the action are located to find the corresponding spatial object phrases. By using fixed collocations of the action word's syntactic position, such as "walking towards the hall" or "entering the room," the target object of the action is determined. Next, the combination order of each action and its corresponding spatial object in the path is categorized, and the position of the action is numbered and marked. Then, the subsequent spatial phrases are matched one by one. If there is an adjacent relationship between the action and the spatial object, or if they are separated by a prepositional phrase by no more than two word positions, it is considered a direct pointing relationship. For example, in the sentence "After passing through the woods along the path, we arrived at the lakeside," the action "passed through" and the spatial object "woods" are separated only by the preposition "after," and can be directly considered as the spatial object of the action. The phrases "arrived at" and "lakeside" are consecutive and do not require secondary matching. Based on this, for each path... All actions within the path are sequentially ordered. Syntactic position identification determines whether an action word is a core component of the subject-verb-object structure in the main clause or subordinate clause. If multiple actions appear in the same main clause, they are arranged according to semantic priority based on their order of appearance. For example, in "passing through the square and crossing the gate," "passing through" and "crossing" are parallel structures, and their matching order with the corresponding spatial objects "square" and "gate" needs to be recorded according to their order of appearance. At the same time, for expression structures containing indicator words, such as "going to the platform behind it" and "moving to the area," pronoun reference is analyzed. The spatial objects referred to by "behind" and "the" are back located through the preceding list of spatial entities to complete the restoration mapping of spatial objects. After that, all action words and their corresponding spatial objects are archived together, and the word order, sentence position, and modifier interference between actions and spatial objects are recorded. Finally, the correspondence between actions and spatial objects is obtained.

[0095] S502: Based on the correspondence between actions and spatial objects, identify the terms that indicate directional movement in the action, determine the position of the terms in the sentence, compare the connection between the terms and spatial objects, and obtain the connection relationship between the direction of the action and the object;

[0096] First, extract all directional expressions from all action words, such as "upward," "forward," and "eastward." Pair these directional expressions with action words in the sentence according to their frequency of occurrence. Precisely locate the position of each directional expression in the sentence, determining whether it precedes or follows the verb, is sandwiched between the subject and predicate, or is used as an adverb at the beginning of the sentence. Next, trace whether the action modified by the directional expression has a clear connection with a spatial object through syntactic dependency. For example, in the expression "slowly move northward to the building," the directional word "northward" modifies the verb "move," and "move" acts on the spatial object "building," thus forming an indirect connection between "northward" and "building." Therefore, a positional index needs to be established between each directional expression, the action word, and the spatial object. Then, classify the diverse connection structures in the corpus. Cases where directional words connect verbs and objects through prepositions such as "to," "toward," and "enter" are categorized into continuous connection structures. Situations where the action is separated from the other by syntactic inserts by more than three words are considered weak connection structures. Simultaneously, the presence of connection-strengthening words, such as "direct" or "continuous," is flagged as a basis for subsequent sorting. In a practical example, if the input statement is "quickly cross the market in the due east direction and arrive at the destination," then "in the due east direction" is extracted as the direction term, "cross" as the action, and "market" as the object. The direction term and action are adjacent and directly act on the object, forming a strong connection. Conversely, in the sentence "move first at the destination, then turn north into the street," "turn north" precedes the action "turn in," and there is an intermediary action term "turn in" between it and "street," forming an indirect connection. All sample statements are labeled according to connection type. Then, for cases where multiple direction terms point to the same object in the connection pattern, the dominant direction term is determined by the order of the action terms, and redundant segments are filtered out, ultimately yielding the connection relationship between the action direction and the object.

[0097] S503: Based on the connection relationship between the action direction and the object, extract statements whose angle between the action instruction direction vectors is less than a preset threshold, and process them according to the language connection order between the spatial object and the action direction to obtain a spatial action structure set;

[0098] First, the directional descriptive information implied by the action words in each statement is extracted and a direction vector is constructed. For example, in "push the part to the right area", the direction corresponding to the action "push" is identified as "right". Combined with the directional phrase "right area" in the statement, a corresponding direction vector V1 is established. At the same time, the positional relationship of the target object corresponding to this action is extracted, and an object position vector V2 is constructed. The angle between the action direction vector and the object position vector is calculated. If the angle is less than the preset threshold θ0, the statement is retained and included as a statement set that meets the direction consistency. The threshold θ0 is set to 45 degrees, which is derived from the conventional angle error tolerance range in the reference standard spatial action model. In the example, if the direction vector in "push the part to the right side of the assembly table" is V1=(1,0) and the object vector is V2=(0.7,0.7), the angle calculation result is 38.7 degrees, which is less than θ0=45 degrees. Then the statement is valid, and its direction and object matching relationship is recorded. Subsequently, all valid statements are semantically parsed according to the order of spatial objects and directional language in the action instructions. The spatial noun phrases before the action words and the combination order of the direction words are extracted to confirm the statement. The sentence structure is either "spatial object + action direction" or "action direction + spatial object". Its word order structure is coded. For example, "move the material forward along the passage" is in the "direction → object" order, while "move the part to the left passage" is in the "object → direction" order. The proportion of the two word orders in all valid sentences is counted. If the proportion of a certain word order is greater than 0.6, it is determined to be the dominant word order. For example, in 70 valid sentences, 45 are of the "object → direction" type, with a proportion of 0.64, which meets the dominant word order condition. All sentences with dominant word order structures are arranged in the order of action appearance. Each action, its corresponding direction vector, and the target object vector are extracted to construct a structural triplet, with the structure being (action, direction vector, object vector), for example (move, (0, -1), target platform). Each structure is then further judged for connection. If the change in direction vector between adjacent action triplets is less than 30 degrees and the spatial region to which the object belongs has a continuous pointing relationship, then the action structure is combined into a continuous spatial action structure fragment. Finally, all valid structural fragments are integrated into an ordered set, resulting in a spatial action structure set.

[0099] Please see Figure 7 This embodiment also provides a spatial data intelligent analysis and decision support system based on a large language model, including:

[0100] The spatial sentence recognition module is configured to acquire spatial entity names, location phrases and task description phrases in the input sentence, locate sentences expressing spatial operations, distinguish the first and second occurrences of entities, analyze the relationship between action words and modifiers, split sentences according to task content, and obtain a list of spatial semantic instruction task items.

[0101] The action element parsing module is configured to filter statements pointing to the same spatial entity based on the list of spatial semantic instruction tasks, identify the word order relationship between directional words, position words, and scope words in the statements and entity phrases, determine the language connection before and after the action words, and obtain a set of task action ranking results.

[0102] The semantic path connection module is configured to identify the connection content between action language fragments based on the task action sorting result set, determine the word order relationship between modifiers and action statements, extract continuous semantic structure, and obtain a set of continuous instruction paths for spatial tasks.

[0103] The spatial scope jump module is configured to identify phrases representing geographical scope in sentences based on the continuous instruction path set of spatial tasks, analyze the linguistic connection between space and action, identify whether the action crosses spatial levels in the language, filter coherent paths where the scope changes at different levels, and obtain the spatial task level jump path set.

[0104] The task flow output module is configured to locate the spatial object phrases that perform actions based on the spatial task hierarchy jump path set, identify the occurrence position of directional words in the sentence structure, analyze the word order continuation between the action structure and the spatial object, extract sentences with an angle of less than a preset threshold for the action instruction direction vector, and obtain a spatial action structure set.

[0105] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A spatial data intelligent analysis and decision support method based on a large language model, characterized in that, Includes the following steps: S1: Obtain spatial entities, location phrases and task descriptions from the input statement, locate spatial operation sentences, distinguish the first and second occurrences of entities, analyze the relationship between action words and modifiers, split sentences according to task content, and obtain a list of spatial semantic instruction task items. S2: Based on the list of spatial semantic instruction task items, filter statements that point to the same spatial entity, analyze the word order relationship of direction, position and range, determine the language connection before and after the action word, and obtain the task action sorting result set; S3: Based on the task action sorting result set, identify the language structure of actions and spatial entities, determine the word order relationship between modifiers and action statements, extract continuous semantic structure, and obtain a set of continuous instruction paths for spatial tasks; S4: Based on the set of continuous instruction paths for the spatial task, analyze the language connection between space and action, filter out paths with varying ranges and coherence, and obtain a set of spatial task hierarchical jump paths. S5: Based on the spatial task hierarchical jump path set, identify the connection order between actions and spatial entities, determine the action direction and object order, extract statements whose angle between action instruction direction vectors is less than a preset threshold, and obtain a spatial action structure set. The spatial semantic instruction task item list includes spatial entity names, location phrases, task description phrases, operation sentence annotations, entity first and repeat markers, action modifiers, action-location relationships, and task sentence divisions. The task action sorting result set includes entity pointing markers, direction and range language, structural boundary relationships, and semantic connection methods. The spatial task continuous instruction path set includes associated entity phrases, structural location differences, modification and word order relationships, action connection components, and unified paragraphs. The spatial task hierarchical jump path set includes spatial range terms, semantic direction order, location-action connection methods, and range changes and connection content. The spatial action structure set includes action objects, directional term positions, connection order, and directional consistent action flows. The spatial operation sentences refer to statements relating spatial entities to actions, including linguistic expressions of position movement, position transformation, and spatial task execution. The aforementioned linguistic cohesion refers to the logical connection between differentiated action words, modifiers, and spatial entities in a sentence in terms of linguistic order, which is reflected in the naturalness of semantic transition and the rationality of word order; The spatial entity language structure refers to the linguistic relationship between the grammatical position of a spatial entity in a sentence and the action word, including the linguistic relationship between the subject, object, and positional components; The connecting word order refers to the sequence and connection between action words, directional words, and spatial entities in language.

2. The spatial data intelligent analysis and decision support method based on a large language model according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Obtain the spatial entity name, location phrase and task description phrase in the input statement, locate the sentences associated with spatial operations, and obtain the set of sentences with spatial operation semantics; S102: Based on the set of sentences with the aforementioned spatial operation semantics, distinguish the first and second occurrences of repeated entities, locate the position of each entity in the sentence, and obtain spatial entity position information; S103: Based on the spatial entity location information, identify the relationship between action words and modifiers, associate action words with target objects, and obtain a list of spatial semantic instruction task items.

3. The spatial data intelligent analysis and decision support method based on a large language model according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the spatial semantic instruction task item list, identify the content in the statement that points to the same spatial entity, extract the phrases in each instruction that are associated with spatial direction, position and range, analyze the position of the phrases in the sentence, and determine their role in the task to obtain the direction, position and range phrase information. S202: Based on the directional position and range phrase information, analyze the phrase order in the sentence, identify the order and position of their appearance in the sentence, decompose the connection relationship between action words and phrases, and obtain the spatial task action sequence structure; S203: Based on the spatial task action sequence structure, analyze the connection relationship between action words and modifiers, calculate the semantic similarity of adjacent sentences and compare it with a preset threshold to obtain a set of task action ranking results.

4. The spatial data intelligent analysis and decision support method based on a large language model according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the task action sorting result set, identify spatial entity phrases associated with action content, obtain the position of each action and corresponding entity in the sentence, determine the grammatical position of the entity in the differentiated sentence structure, and obtain the linguistic positional relationship between the action and the spatial entity. S302: Based on the linguistic positional relationship between the action and the spatial entity, analyze the word order relationship between the modifiers and entity phrases before and after the action, and analyze the logical sequence relationship between adjacent actions according to the preset spatiotemporal knowledge graph, determine whether the action words before and after are connected to each other, and obtain the connection relationship between the action and the modifier. S303: Based on the connection relationship between the action and the modifier, identify the connecting components in the adjacent action language structure, filter the segments where the subject and object have a coreference relationship, and advance the semantic extension according to the action sequence to obtain the set of continuous instruction paths for the spatial task.

5. The spatial data intelligent analysis and decision support method based on a large language model according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the continuous instruction path set of the spatial task, identify words in the path that include descriptions of geographic spatial range, locate the position of the words in the sentence, and connect them with the associated actions in grammatical order to obtain the spatial range description and action connection relationship; S402: Based on the spatial range description and action connection relationship, analyze the order of spatial position words before and after the statement, determine the connection mode between differentiated spatial ranges, and filter the language transition from local to whole in the path to obtain spatial range jump information; S403: Based on the spatial range jump information, identify the continuous semantics generated by the change of spatial range in the path, filter the paths that form continuous jumps at the spatial level, and obtain the set of spatial task level jump paths.

6. The spatial data intelligent analysis and decision support method based on a large language model according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Based on the set of spatial task hierarchical jump paths, locate the spatial entity pointed to by each action, and determine the order of actions and the spatial entities pointed to by actions according to the positional relationship between actions and spatial entities in the statement, so as to obtain the correspondence between actions and spatial entities. S502: Based on the correspondence between the action and the spatial entity, identify the word in the action that indicates directional movement, determine the position of the word in the sentence, compare the connection between the word and the spatial entity, and obtain the connection relationship between the action direction and the object; S503: Based on the connection relationship between the action direction and the object, extract statements whose angle between the action instruction direction vectors is less than a preset threshold, and process them according to the language connection order between the spatial entity and the action direction to obtain a spatial action structure set.

7. A spatial data intelligent analysis and decision support system based on a large language model, characterized in that, The system is used to implement the spatial data intelligent analysis and decision support method based on a large language model as described in any one of claims 1-6, and the system includes: The spatial sentence recognition module is configured to acquire spatial entity names, location phrases and task description phrases in the input sentence, locate sentences expressing spatial operations, distinguish the first and second occurrences of entities, analyze the relationship between action words and modifiers, split sentences according to task content, and obtain a list of spatial semantic instruction task items. The action element parsing module is configured to filter statements pointing to the same spatial entity based on the list of spatial semantic instruction task items, identify the word order relationship between directional words, position words and range words in the statements and entity phrases, determine the language connection before and after the action words, and obtain a set of task action ranking results. The semantic path connection module is configured to identify the connection content between action language segments based on the task action sorting result set, determine the word order relationship between modifiers and action statements, extract continuous semantic structure, and obtain a set of continuous instruction paths for spatial tasks. The spatial scope jump module is configured to identify phrases representing geographical scope in statements based on the continuous instruction path set of the spatial task, analyze the linguistic connection between space and action, identify whether the action crosses spatial levels in the language, filter coherent paths where the scope changes at different levels, and obtain a set of spatial task level jump paths. The task flow output module is configured to locate spatial entity phrases that perform actions based on the set of spatial task hierarchical jump paths, identify the position of directional words in the sentence structure, analyze the word order connection between the action structure and the spatial entity, extract sentences with an angle of less than a preset threshold, and obtain a set of spatial action structures.

Images