Sea area development suitability evaluation method and system

Abstract

The invention relates to the technical field of sea area development, and particularly discloses a sea area development suitability evaluation method and system, and the method employs a natural language processing technology based on deep learning to carry out the data processing of a sea area development suitability evaluation document. A man-made judgment adaptability grading label and evaluation index related information of sea area development are extracted, semantic coding and semantic feature enhancement processing are carried out on the evaluation index related information, intelligent evaluation of sea area development suitability is achieved, and then an intelligent evaluation result and the man-made judgment adaptability grading label are compared and analyzed. Therefore, the reliability of manual judgment is verified. Through the mode, the subjectivity deviation in the sea area development suitability evaluation process can be effectively reduced, the evaluation objectivity and accuracy are improved, and therefore a more reliable basis is provided for subsequent sea area development activity decision making.

Description

Technical Field

[0001] The present application relates to the technical field of marine area development, and more specifically, to a method and system for evaluating the suitability of marine area development. Background Art

[0002] With the rapid development of the global marine economy, the effective management and rational development of marine resources have become the focus of international attention. Marine development not only involves the utilization of marine resources, but also involves environmental protection, ecological balance, and sustainable development. Therefore, before marine development, a comprehensive and systematic suitability assessment is required to ensure that the development activities can both promote economic development and minimize the impact on the environment.

[0003] Traditional marine development suitability assessment mainly relies on the experience and intuition of experts, and uses qualitative and quantitative methods to evaluate the physical, chemical, biological and other characteristics of the marine area. However, this method has problems such as strong subjectivity, long time consumption, high cost and difficulty in standardization. Especially in the face of large-scale or complex marine environments, due to the subjectivity and uncertainty of human judgment itself, it is often difficult to ensure the accuracy and consistency of the evaluation results.

[0004] Therefore, we look forward to an optimized marine development suitability evaluation system that can assist in verifying the results of human evaluation to improve the objectivity and accuracy of the evaluation. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a method and system for evaluating the suitability of sea area development, which uses natural language processing technology based on deep learning to process data on sea area development suitability evaluation documents, extracts human-judged adaptability grading labels and evaluation index-related information of sea area development, and performs semantic encoding and semantic feature enhancement processing on the evaluation index-related information to achieve intelligent evaluation of the suitability of sea area development, and then compares and analyzes the intelligent evaluation results with the human-judgment adaptability grading labels to verify the reliability of human judgment. In this way, the subjective bias in the process of sea area development suitability evaluation can be effectively reduced, and the objectivity and accuracy of the evaluation can be improved, thereby providing a more reliable basis for subsequent decision-making on sea area development activities.

[0006] Accordingly, according to one aspect of the present application, a marine development suitability evaluation system is provided, which includes: Evaluation document extraction module, used to extract marine development suitability evaluation documents; An information extraction module, used to extract a set of human-judged adaptability classification labels and indicator data from the marine development suitability evaluation document, wherein the indicator data includes evaluation indicators, indicator explanations and indicator assignments; An indicator data semantic embedding coding module, used for performing semantic embedding coding on each indicator data in the set of indicator data to obtain a set of indicator data semantic embedding coding vectors; A semantic feature enhancement module, used for performing semantic feature enhancement processing on the set of indicator data semantic embedding coding vectors to obtain a set of indicator data enhanced semantic embedding coding vectors; An AI adaptive grading module, for generating an AI adaptive grading label based on the global context features of the set of semantic embedding encoding vectors enhanced by the indicator data; The early warning module is used to compare the human-judged adaptability grading label with the AI ​​adaptability grading label, and generate an early warning prompt if the two are inconsistent.

[0007] In the above-mentioned marine development suitability assessment system, the semantic feature enhancement module includes: a feature updating unit, which is used to update the set of indicator data semantic embedding coding vectors based on the feature field effect of the set of indicator data semantic embedding coding vectors to obtain an updated set of indicator data semantic embedding coding vectors; and a context association enhancement unit, which is used to perform bidirectional context association enhancement on the updated set of indicator data semantic embedding coding vectors to obtain the set of indicator data enhanced semantic embedding coding vectors.

[0008] According to another aspect of the present application, a method for evaluating the suitability of marine development is provided, which comprises: Extracting marine development suitability assessment documents; Extracting a set of human-judged adaptability classification labels and indicator data from the marine development suitability evaluation document, wherein the indicator data includes evaluation indicators, indicator explanations, and indicator values; Performing semantic embedding coding on each indicator data in the set of indicator data to obtain a set of indicator data semantic embedding coding vectors; Performing semantic feature enhancement processing on the set of indicator data semantic embedding coding vectors to obtain a set of indicator data enhanced semantic embedding coding vectors; Generate AI adaptive grading labels based on global context features of a set of semantic embedding encoding vectors reinforced by the indicator data; The human-judged adaptability grading label and the AI ​​adaptability grading label are compared, and if the two are inconsistent, an early warning prompt is generated.

[0009] Compared with the prior art, the method and system for evaluating the suitability of sea area development provided by this application uses natural language processing technology based on deep learning to process data of sea area development suitability evaluation documents, extracts human-judged adaptability classification labels and evaluation index-related information of sea area development, and realizes intelligent evaluation of the suitability of sea area development by semantically encoding and enhancing the relevant information of evaluation indexes, and then compares and analyzes the intelligent evaluation results with the human-judged adaptability classification labels to verify the reliability of human judgment. In this way, the subjective bias in the process of evaluating the suitability of sea area development can be effectively reduced, and the objectivity and accuracy of the evaluation can be improved, thereby providing a more reliable basis for subsequent decision-making on sea area development activities. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 The present invention is a block diagram of a system for evaluating the suitability of marine development according to an embodiment of the present application.

[0012] Figure 2 This is a data flow diagram of a sea area development suitability assessment system according to an embodiment of the present application.

[0013] Figure 3 It is a block diagram of a semantic feature enhancement module in a marine area development suitability evaluation system according to an embodiment of the present application.

[0014] Figure 4 This is a block diagram of the AI ​​adaptability grading module in the marine development suitability evaluation system according to an embodiment of the present application.

[0015] Figure 5 The present invention is a flowchart of a method for evaluating the suitability of sea area development according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] Below, the embodiments of the present application will be described in more detail in conjunction with the accompanying drawings, and the above and other purposes, features and advantages of the present application will become more apparent. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0017] In response to the technical problems described in the above background technology, this application proposes a marine development suitability evaluation system. Figure 1The present invention is a block diagram of a system for evaluating the suitability of marine development according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the marine development suitability evaluation system according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the marine development suitability evaluation system 100 includes: an evaluation document extraction module 110, which is used to extract marine development suitability evaluation documents; an information extraction module 120, which is used to extract a set of human-judged adaptability grading labels and indicator data from the marine development suitability evaluation documents, wherein the indicator data includes evaluation indicators, indicator explanations and indicator assignments; an indicator data semantic embedding coding module 130, which is used to perform semantic embedding coding on each indicator data in the set of indicator data to obtain a set of indicator data semantic embedding coding vectors; a semantic feature enhancement module 140, which is used to perform semantic feature enhancement processing on the set of indicator data semantic embedding coding vectors to obtain a set of indicator data enhanced semantic embedding coding vectors; an AI adaptability grading module 150, which is used to generate an AI adaptability grading label based on the global contextual features of the set of indicator data enhanced semantic embedding coding vectors; an early warning module 160, which is used to compare the human-judged adaptability grading label with the AI ​​adaptability grading label, and generate an early warning prompt if the two are inconsistent.

[0018] In the above-mentioned marine development suitability evaluation system 100, the evaluation document extraction module 110 is used to extract marine development suitability evaluation documents. It should be understood that the marine development suitability evaluation document is a document compiled by the user according to the actual marine conditions and relevant evaluation standards for marine development suitability, and includes a detailed description of the marine environment, human judgment adaptability classification labels made according to the actual marine environment and relevant evaluation standards, and evaluation index explanations and assignments.

[0019] In one embodiment of the present application, the relevant standards and implementation methods for marine development suitability assessment are as follows: 1. Determine the scope of evaluation. The evaluation scope should be the sea areas under the jurisdiction of coastal prefecture-level cities and above. Coastal provincial counties should be evaluated uniformly across the province. The land-sea boundary shall be based on the latest published coastline.

[0020] 2. Determine the evaluation object. According to the different purposes of the evaluation, the suitability evaluation of sea area development includes the suitability evaluation of marine aquaculture, the suitability evaluation of port construction, the suitability evaluation of offshore wind power, and the suitability evaluation of tourism and leisure.

[0021] 3. Divide the evaluation units. In the evaluation of suitability of marine aquaculture, offshore wind power, and tourism and leisure, the sea area within the evaluation scope should be divided into grids with an accuracy of not less than 1km*1km; in the evaluation of suitability of port construction, the coastline within the evaluation scope should be divided into shore sections with an accuracy of not less than 0.1km. Each grid or shore section is an evaluation unit. If the sea area within the evaluation scope is too small or too large, the size of the evaluation unit should be appropriately adjusted.

[0022] 4. Data collection and collation: Before conducting the evaluation, supporting data of the evaluation sea area and related supporting areas need to be collected and collated, which should meet the following requirements: Collect and evaluate data on the sea area's meteorology, oceanographic hydrology, seabed topography, marine life, marine ecology, social economy, and related planning and basic base maps, and classify, summarize and organize the information to ensure the authority, accuracy and timeliness of the data.

[0023] Vector data should adopt the "2000 National Geodetic Coordinate System" (CGCS 2000) and the "1985 National Height Datum" (Phase II), the projection system should adopt the "Gauss-Krüger" projection, and the zoning should adopt the "National Standard Zoning".

[0024] The accuracy of vector data should not be less than 1km*1km grid unit.

[0025] Analyze and judge based on the data collection status. If necessary, it is recommended to conduct supplementary investigation in accordance with the requirements of GB / T 12763 to ensure the reality of the data, as shown in Table A.

[0026] Table A Data Collection Checklist

[0027] 5. Construction of the index system. For the four types of sea area development, namely, marine aquaculture, port construction, offshore wind power, and tourism and leisure, indicators are selected from three aspects: marine natural conditions, regional social conditions, and ecological conditions to construct a suitability evaluation index system. See Tables 1 to 4.

[0028] Table 1 Evaluation index system for suitability of marine aquaculture

[0029] Table 2 Port construction suitability evaluation index system

[0030] Table 3 Offshore wind power suitability evaluation index system

[0031] Table 4 Tourism and leisure suitability evaluation index system

[0032] 6. Index assignment and weight (1) Assignment of indicators: According to the suitability of the four types of sea area development, assign a value of "1" (unsuitable), "3" (generally suitable), or "5" (suitable) to each evaluation indicator. For the specific assignment method, please refer to Tables 5 to 8.

[0033] (2) Index weights. The weights of various evaluation indicators should be calculated based on the evaluation area and the type of sea area development through methods such as the Delphi method and the analytic hierarchy process. The index weight range should refer to Tables 5 to 8.

[0034] Table 5 Explanation and assignment of marine aquaculture adaptability evaluation indicators

[0035] Table 6 Explanation and assignment of port construction suitability evaluation indicators

[0036] Table 7 Explanation and assignment of suitability evaluation indicators for offshore wind power construction

[0037] Table 8 Explanation and assignment of coastal tourism and leisure suitability evaluation indicators

[0038] 7. Suitability calculation and evaluation (1) Suitability calculation: The suitability index Su of the four types of sea area development in the evaluation area is obtained through raster operation. The calculation formula is as follows:

[0039] Where: represents the suitability index of marine development, Represents the total number of evaluation units, represents the weight of each indicator, Represents the total number of evaluation indicators, Indicates the value of each indicator.

[0040] (2) Suitability evaluation: based on the marine development suitability index Based on the calculation results, the evaluation units are divided into suitable, generally suitable and unsuitable. The classification method can be referred to Table 9. For the evaluation areas with low differentiation of suitability index, the natural breakpoint method and other methods can be used for reasonable classification.

[0041] Table 9 Evaluation of sea area development suitability by grade

[0042] In the above-mentioned marine development suitability evaluation system 100, the information extraction module 120 is used to extract a set of human-judged adaptability grading labels and indicator data from the marine development suitability evaluation document, and the indicator data includes evaluation indicators, indicator explanations and indicator assignments. It should be understood that evaluation indicators are specific indicators used to measure the suitability of marine development, such as seawater quality, marine disaster risk, and the richness of marine biological resources. The indicator explanation is a detailed description of each evaluation indicator, including the definition of the indicator, measurement methods, and evaluation criteria. The indicator assignment is a specific score or grade division for each indicator based on the actual marine conditions. In the technical solution of the present application, by extracting the above-mentioned set of indicator data, the specific conditions of various aspects of marine development can be fully understood, thereby realizing an intelligent evaluation of the suitability of marine development.

[0043] In specific implementation, named entity recognition and information extraction techniques can be used to parse documents, and a collection of human-judged adaptability grading labels and indicator data can be extracted in an automated manner. First, the marine development suitability assessment document needs to be preprocessed to ensure the consistency of the document format and the integrity of the content. The preprocessing steps include format conversion, text cleaning, and paragraph segmentation. Format conversion refers to converting documents into a unified electronic format, such as PDF, Word, etc., to facilitate subsequent automated processing. Text cleaning refers to removing irrelevant information in the document, such as headers, footers, comments, etc., to ensure that the extracted content is clean and tidy. Paragraph segmentation refers to dividing the document into paragraphs or chapters to facilitate subsequent refined processing.

[0044] After preprocessing, the information extraction phase begins. The core task of information extraction is to extract a set of human-judged adaptive classification labels and indicator data from documents. This process is mainly achieved through natural language processing (NLP) technology, which includes two main steps: named entity recognition (NER) and information extraction.

[0045] Named entity recognition (NER) is the first step in information extraction, and its purpose is to identify key entities from documents, such as evaluation indicators, indicator explanations, indicator assignments, etc. Named entity recognition models can be built based on rules and dictionaries. The rule and dictionary method identifies key entities in documents by matching predefined rules and dictionaries. For example, evaluation indicators such as "seawater quality" and "marine disaster risk" can be identified through keyword matching.

[0046] Next is the information extraction stage. The goal of information extraction is to accurately extract a set of adaptive grading labels and indicator data judged by humans from the document. To this end, a combination of multiple methods is usually adopted. The first is the rule matching method, that is, a series of rules are designed to match relevant information in the document based on known evaluation systems and standards. For example, for data tables and scoring rules in a specific format, regular expressions or other pattern matching algorithms can be written to extract the required data directly from the text. The second is the template matching method, which is suitable for information with a high degree of structure. By pre-defining data templates, such as table structures, field names, etc., qualified data can be quickly located and extracted from the document.

[0047] After the information extraction is completed, the data sorting stage will follow. The main task of this stage is to process the extracted raw data into a structured and easy-to-use format. First, the extracted information needs to be mapped according to the predetermined data structure. Common data structures include relational database tables, JSON objects, etc. This process requires the design of a suitable data model according to specific needs and the writing of corresponding data mapping rules. Secondly, data cleaning is also an indispensable link. In actual operation, due to various reasons, the extracted data may have problems such as inconsistent format, missing data, and outliers. These problems need to be solved through data cleaning. For example, for data with inconsistent formats, the format conversion can be performed through string processing functions; for missing data, it can be completed or deleted according to business logic; for outliers, it is necessary to combine domain knowledge for judgment and correction.

[0048] In the above-mentioned marine development suitability evaluation system 100, the indicator data semantic embedding coding module 130 is used to perform semantic embedding coding on each indicator data in the set of indicator data to obtain a set of indicator data semantic embedding coding vectors. It should be understood that, considering that the original indicator data is usually a description in text form, this unstructured information form is difficult to be directly processed by the machine learning model. Therefore, the present application further performs semantic embedding coding on each indicator data in the set of indicator data to capture the semantic information in the text data, and convert the text data into a numerical vector representation, thereby providing an effective data basis for subsequent machine learning models. In a specific example of the present application, a semantic embedding encoder based on the BERT model is used to perform semantic embedding coding on each indicator data in the set of indicator data to obtain a set of indicator data semantic embedding coding vectors. Those of ordinary skill in the art should know that the BERT model is a pre-trained language representation model that can capture contextual information in the text through a bidirectional Transformer structure, and effectively mine the semantic associations between indicator definitions, indicator interpretations, and indicator assignments in each indicator data, so as to fully understand the specific situation of marine development in various evaluation aspects and realize the semantic embedding representation of each indicator data.

[0049] In the above-mentioned marine development suitability evaluation system 100, the semantic feature enhancement module 140 is used to perform semantic feature enhancement processing on the set of indicator data semantic embedding coding vectors to obtain a set of indicator data enhanced semantic embedding coding vectors. It should be understood that, considering that in the marine development suitability evaluation, there is usually a certain dependency relationship between various indicator data, and each indicator data has different importance. Therefore, in order to further improve the accuracy of the evaluation, the present application proposes a semantic enhancement method based on feature field effects, which learns the semantic association between indicator data by analyzing the feature field effect distribution of the set of indicator data semantic embedding coding vectors, thereby utilizing the semantic association relationship between each indicator data to perform semantic enhancement processing on the set of indicator data semantic embedding coding vectors to improve the expressive power of the features.

[0050] Figure 3 FIG. 1 is a block diagram of a semantic feature enhancement module in a marine area development suitability evaluation system according to an embodiment of the present application. Figure 3As shown, the semantic feature enhancement module 140 includes: a feature updating unit 141, which is used to update the set of indicator data semantic embedding coding vectors based on the feature field effect of the set of indicator data semantic embedding coding vectors to obtain an updated set of indicator data semantic embedding coding vectors; a context association strengthening unit 142, which is used to perform bidirectional context association strengthening on the updated set of indicator data semantic embedding coding vectors to obtain the set of indicator data strengthened semantic embedding coding vectors.

[0051] Specifically, the feature updating unit 141 includes: a semantic reconstruction subunit, which is used to use a semantic reconstructor to semantically reconstruct each indicator data semantic embedding coding vector in the set of indicator data semantic embedding coding vectors to obtain a set of reconstructed indicator data semantic embedding coding vectors; a semantic depth calculation subunit, which is used to calculate the semantic depth implicit representation value of each reconstructed indicator data semantic embedding coding vector in the set of reconstructed indicator data semantic embedding coding vectors to obtain a set of indicator data semantic depth implicit representation values; a reordering subunit, which is used to reorder the set of indicator data semantic embedding coding vectors based on the size order of each indicator data semantic depth implicit representation value in the set of indicator data semantic depth implicit representation values ​​to obtain the set of updated indicator data semantic embedding coding vectors.

[0052] In a specific example of the present application, the semantic reconstruction subunit is used to: use the first weight parameter matrix and the second weight parameter matrix to continuously linearly transform the indicator data semantic embedding coding vector to obtain the reconstructed indicator data semantic embedding coding vector, which is expressed by the formula:

[0053]

[0054] in, represents the set of semantic embedding encoding vectors of the indicator data, , , and Respectively represent the first, second, and third in the set of semantic embedding coding vectors of the indicator data and The semantic embedding encoding vector of indicator data, is the number of semantic embedding coding vectors of the indicator data, and denote the first weight parameter matrix and the second weight parameter matrix respectively, For the That is, considering that each indicator data may have different feature dimensions, the semantic reconstructor is first used to perform domain transformation on the semantic embedding coding vector of each indicator data, map it to a unified feature space, unify the feature dimensions, and generate a set of semantic embedding coding vectors of reconstructed indicator data.

[0055] In a specific example of the present application, the semantic depth calculation subunit is used to: calculate the square of the Euclidean norm of the reconstructed indicator data semantic embedding coding vector to obtain the initial semantic depth implicit representation value of the indicator data; calculate the ratio of the sum of the initial semantic depth implicit representation value of the indicator data and the difference between the initial semantic depth implicit representation value of the indicator data to obtain the indicator data semantic depth representation modulation coefficient; calculate the logarithmic function value of the indicator data semantic depth representation modulation coefficient with base 2 to obtain the indicator data semantic depth implicit representation value, which is expressed as follows:

[0056] in, represents the calculation of the Euclidean norm of a vector, represents the logarithmic function with base 2, Indicates The semantic depth implicit representation value of each indicator data. That is, by calculating the semantic depth implicit representation value of the semantic embedding coding vector of each reconstructed indicator data, the information density or complexity carried by each indicator data in the overall data set is measured, and its importance and influence in the overall information are quantified.

[0057] Specifically, the context association strengthening unit 142 is used to: input the set of updated indicator data semantic embedding coding vectors into the first semantic association encoder based on the forward LSTM model and the second semantic association encoder based on the backward LSTM model to obtain a set of forward indicator data semantic association coding feature vectors and a set of backward indicator data semantic association coding feature vectors; perform weighted fusion on the set of forward indicator data semantic association coding feature vectors and the set of backward indicator data semantic association coding feature vectors to obtain the set of indicator data enhanced semantic embedding coding vectors, which is expressed as follows:

[0058]

[0059]

[0060]

[0061] in, represents the set of updated semantic embedding encoding vectors of indicator data, , , and respectively represent the first, second, and third in the set of updated indicator data semantic embedding coding vectors and The semantic embedding encoding vector of indicator data, represents the forward LSTM model, represents the backward LSTM model, represents the set of semantic association encoding feature vectors of forward indicator data, Represents the set of semantic association encoding feature vectors of backward indicator data, and denote the first weighting parameter and the second weighting parameter respectively, is the bias term, A set of semantic embedding encoding vectors is enhanced for the indicator data.

[0062] That is, in order to ensure that important indicator data information is given priority in the subsequent contextual semantic association enhancement encoding process, the feature order of each indicator data is further adjusted according to the semantic depth implicit representation value of each indicator data, and the information structure of the data set is reorganized. Then, on this basis, a bidirectional long short-term memory (LSTM) model is used to encode the updated indicator data semantic embedding encoding vector from the forward and backward directions respectively. Among them, the forward LSTM model can capture the information flow pattern from front to back in the indicator data set, while the backward LSTM model captures the information dependency from back to front through reverse operation. Through this bidirectional encoding mechanism, it can effectively capture the information dependency at different levels, and thereby enhance the semantic feature expression of each indicator data. Finally, by integrating the encoding results of the bidirectional LSTM model to integrate the bidirectional contextual semantic information of each indicator data, a richer and more comprehensive set of indicator data enhancement semantic embedding encoding vectors is formed, thereby achieving a deep understanding of each indicator data.

[0063] In the above-mentioned marine development suitability evaluation system 100, the AI ​​adaptability grading module 150 is used to generate AI adaptability grading labels based on the global context features of the set of semantic embedding coding vectors of the indicator data reinforcement. Figure 4 FIG. 1 is a block diagram of an AI adaptability grading module in a marine area development suitability evaluation system according to an embodiment of the present application. Figure 4As shown, the AI ​​adaptive grading module 150 includes: a global context semantic coding unit 151, which is used to perform global context semantic coding on the set of indicator data enhanced semantic embedded coding vectors to obtain an indicator global semantic coding vector; an adaptive grading decision unit 152, which is used to input the indicator global semantic coding vector into a classifier-based intelligent decision module to obtain the AI ​​adaptive grading label.

[0064] Specifically, the global context semantic encoding unit 151 is used to perform global context semantic encoding on the set of the indicator data enhanced semantic embedding encoding vectors to obtain the indicator global semantic encoding vector. In a specific example of the present application, the global context semantic encoding unit 151 is used to: input the set of the indicator data enhanced semantic embedding encoding vectors into a global semantic encoder based on a multi-head self-attention structure to obtain the indicator global semantic encoding vector. That is, in order to further integrate the global information of the indicator data and comprehensively evaluate the suitability of sea area development from a global perspective, the present application further uses a global semantic encoder based on a multi-head self-attention structure to globally encode the set of the indicator data enhanced semantic embedding encoding vectors to obtain the indicator global semantic encoding vector. Those of ordinary skill in the art should know that the multi-head self-attention mechanism can simultaneously focus on different positions in the set of the indicator data enhanced semantic embedding encoding vectors through multiple independent attention heads working in parallel, and capture the multi-dimensional associations between the indicator data from different subspaces. Integrating the global information of each indicator data in this way can significantly improve the model's understanding depth and accuracy of the sea area development suitability evaluation indicator data, thereby providing more reliable support for subsequent intelligent evaluation.

[0065] Specifically, the adaptive grading decision unit 152 is used to input the global semantic coding vector of the indicator into the intelligent decision module based on the classifier to obtain the AI ​​adaptive grading label. In a specific example of the present application, the classifier adopts a multi-layer perceptron structure, and performs multi-layer nonlinear transformation on the input global semantic coding vector of the indicator through a fully connected layer to capture the semantic characteristics of the marine development suitability evaluation index data contained therein, and designs neurons corresponding to the number of adaptive grading labels in the output layer, and uses the softmax function to perform probability calculation to obtain the possibility that the marine development suitability belongs to the corresponding grade, so as to determine the final AI adaptive grading label, such as "suitable", "relatively suitable", and "unsuitable".

[0066] In particular, in the technical solution of the present application, each indicator data enhanced semantic embedding coding vector in the set of indicator data enhanced semantic embedding coding vectors respectively represents the semantic enhanced features of the semantic embedding coding features of each indicator data after field effect analysis. In the process of inputting it into the global semantic encoder based on the multi-head self-attention structure, the multi-head self-attention structure is essentially used to capture the global contextual semantic association features between each indicator data. However, because the semantic feature enhancement based on field effect distribution analysis is essentially an attention mechanism, the global semantic encoder based on the multi-head self-attention structure is essentially a secondary attention mechanism, and although the attention mechanism will enhance the feature significance, it will also introduce attention interference noise, which may cause interference components in the indicator global semantic encoding vector. The interference component of the indicator global semantic encoding vector will ensure that the overall feature manifold of the indicator global semantic encoding vector in the high-dimensional feature space has fine-grained feature structure holes. The existence of fine-grained feature structure holes will not only make the semantic coverage of the features of the indicator global semantic encoding vector corresponding to the class probability label insufficient, but also cause the outlier class regression inference mapping to be off-target, affecting the accuracy of the AI ​​adaptive grading label obtained by the input of the indicator global semantic encoding vector into the intelligent decision module based on the classifier.

[0067] Based on this, in a preferred example of the present application, the indicator global semantic encoding vector is input into a classifier-based intelligent decision module to obtain an AI adaptive grading label, including: first, the indicator global semantic encoding vector is input into a pre-classifier based on a Softmax function to obtain an indicator global semantic fine-grained class probability label vector, as shown below:

[0068] in, The first eigenvalues, is the scale value of the global semantic encoding vector of the indicator, The first one represents the global semantic fine-grained class probability label vector of the indicator characteristic values.

[0069] Next, the product between the indicator global semantic fine-grained class probability label vector and its transposed vector is calculated to obtain the indicator global semantic fine-grained full label domain modulation matrix; the eigenvalues ​​of each position in the indicator global semantic fine-grained full label domain modulation matrix are divided by the square root of the scale value of the indicator global semantic encoding vector to obtain the indicator global semantic fine-grained full label domain scale modulation matrix, as shown below:

[0070] in, represents the global semantic fine-grained class probability label vector of the indicator, represents the transpose of the eigenvector, represents the matrix multiplication operation, Represents the global semantic fine-grained full-label domain scale modulation matrix of the indicator.

[0071] Secondly, the index global semantic fine-grained full label domain scale modulation matrix is ​​input into the class probability domain attention sparse module based on the multi-level gating function to obtain the sparse index global semantic fine-grained full label domain scale modulation matrix, as shown below:

[0072]

[0073] in, The first one represents the global semantic fine-grained full label domain scale modulation matrix of the indicator eigenvalues, represents a multi-level gating function, is the mask threshold of the multi-level gating function, is the scaling parameter, The first one represents the global semantic fine-grained full-label domain scale modulation matrix of the sparsification index feature values.

[0074] Furthermore, the indicator global semantic coding vector is used as the query feature vector, and the sparse indicator global semantic fine-grained full label domain scale modulation matrix is ​​matrix-multiplied with the indicator global semantic coding vector to obtain an optimized indicator global semantic coding vector, as shown below:

[0075] in, represents the global semantic encoding vector of the indicator, represents the global semantic fine-grained full-label domain scale modulation matrix of the sparsification index, A global semantic encoding vector representing the optimized indicator.

[0076] Finally, the optimized indicator global semantic coding vector is input into the classifier-based intelligent decision-making module to obtain the AI ​​adaptability grading label.

[0077] Therefore, by adding a class probability domain-level optimizable perturbation to the original feature vector to strengthen the dependency entanglement between different variables in the original feature vector based on label domain modulation, the source domain feature vector can be optimized based on class probability query-based scale modulation. In this way, the significant inherent modal information in the source domain features can be more effectively retained, and the interference components in the feature distribution can be effectively masked, thereby improving the adversarial robustness of the manifold expression of the feature vector.

[0078] In the above-mentioned marine development suitability evaluation system 100, the early warning module 160 is used to compare the manually judged adaptability grading label and the AI ​​adaptability grading label, and generate an early warning prompt if the two are inconsistent. That is, if the manually judged adaptability grading label and the AI ​​adaptability grading label are inconsistent, it indicates that there may be certain errors or deviations. At this time, a corresponding early warning prompt is further generated to remind relevant personnel to review and adjust. Through this combination of intelligent evaluation and manual review, the accuracy and reliability of marine development suitability evaluation can be effectively ensured, providing a scientific basis for the rational use and protection of marine resources.

[0079] In summary, the marine development suitability evaluation system according to the embodiment of the present application is explained, which uses natural language processing technology based on deep learning to process data of marine development suitability evaluation documents, extracts human-judgment adaptability grading labels and evaluation index-related information of marine development, and realizes intelligent evaluation of marine development suitability by semantically encoding and enhancing the relevant information of evaluation indexes, and then compares and analyzes the intelligent evaluation results with the human-judgment adaptability grading labels to verify the reliability of human judgment. In this way, the subjective bias in the marine development suitability evaluation process can be effectively reduced, and the objectivity and accuracy of the evaluation can be improved, thereby providing a more reliable basis for subsequent marine development activity decisions.

[0080] Figure 5 Flow chart of the method for evaluating the suitability of sea area development according to an embodiment of the present application. Figure 5As shown, according to the method for evaluating the suitability of sea area development in an embodiment of the present application, the method includes the following steps: S1, extracting a sea area development suitability evaluation document; S2, extracting a set of manually judged adaptability grading labels and indicator data from the sea area development suitability evaluation document, wherein the indicator data includes evaluation indicators, indicator explanations and indicator assignments; S3, performing semantic embedding coding on each indicator data in the set of indicator data to obtain a set of indicator data semantic embedding coding vectors; S4, performing semantic feature enhancement processing on the set of indicator data semantic embedding coding vectors to obtain a set of indicator data enhanced semantic embedding coding vectors; S5, generating an AI adaptability grading label based on the global contextual features of the set of indicator data enhanced semantic embedding coding vectors; S6, comparing the manually judged adaptability grading label with the AI ​​adaptability grading label, and generating an early warning prompt if the two are inconsistent.

[0081] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned marine development suitability evaluation method have been described in the above reference. Figures 1 to 4 It has been introduced in detail in the description of the marine development suitability evaluation system, and therefore, its repeated description will be omitted.

[0082] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to be implemented by adopting the above specific details.

Claims

1. A marine development suitability evaluation system, characterized in that: include: Evaluation document extraction module, used to extract marine development suitability evaluation documents; An information extraction module, used to extract a set of human-judged adaptability classification labels and indicator data from the marine development suitability evaluation document, wherein the indicator data includes evaluation indicators, indicator explanations and indicator assignments; An indicator data semantic embedding coding module, used for performing semantic embedding coding on each indicator data in the set of indicator data to obtain a set of indicator data semantic embedding coding vectors; A semantic feature enhancement module, used for performing semantic feature enhancement processing on the set of indicator data semantic embedding coding vectors to obtain a set of indicator data enhanced semantic embedding coding vectors; An AI adaptive grading module, for generating an AI adaptive grading label based on the global context features of the set of semantic embedding encoding vectors enhanced by the indicator data; An early warning module is used to compare the human-judged adaptability grading label with the AI ​​adaptability grading label, and generate an early warning prompt if the two are inconsistent; Among them, the semantic feature enhancement module includes: a feature updating unit, which is used to update the set of indicator data semantic embedding coding vectors based on the feature field effect of the set of indicator data semantic embedding coding vectors to obtain an updated set of indicator data semantic embedding coding vectors; a context association strengthening unit, which is used to perform bidirectional context association strengthening on the updated set of indicator data semantic embedding coding vectors to obtain the set of indicator data strengthened semantic embedding coding vectors.

2. The marine development suitability evaluation system according to claim 1 is characterized in that: The indicator data semantic embedding coding module is used to: Use a semantic embedding encoder based on the BERT model to perform semantic embedding encoding on each indicator data in the set of indicator data to obtain a set of semantic embedding encoding vectors of the indicator data.

3. The marine development suitability evaluation system according to claim 2 is characterized in that: The feature updating unit comprises: A semantic reconstruction subunit, used for using a semantic reconstructor to semantically reconstruct each indicator data semantic embedding coding vector in the set of indicator data semantic embedding coding vectors to obtain a set of reconstructed indicator data semantic embedding coding vectors; A semantic depth calculation subunit, used for calculating the semantic depth implicit representation value of each reconstructed indicator data semantic embedding coding vector in the set of reconstructed indicator data semantic embedding coding vectors to obtain a set of indicator data semantic depth implicit representation values; A reordering subunit is used to reorder the set of indicator data semantic embedding coding vectors based on the size order of each indicator data semantic depth implicit representation value in the set of indicator data semantic depth implicit representation values ​​to obtain the updated set of indicator data semantic embedding coding vectors.

4. The marine development suitability evaluation system according to claim 3 is characterized in that: The semantic reconstruction subunit is used to: The indicator data semantic embedding coding vector is continuously linearly transformed using the first weight parameter matrix and the second weight parameter matrix to obtain the reconstructed indicator data semantic embedding coding vector.

5. The marine development suitability evaluation system according to claim 4 is characterized in that: The semantic depth calculation subunit is used to: Calculating the square of the Euclidean norm of the reconstructed indicator data semantic embedding coding vector to obtain an initial semantic depth implicit representation value of the indicator data; Calculating a ratio of a sum of an initial semantic depth implicit representation value of the indicator data and a difference of an initial semantic depth implicit representation value of the indicator data to obtain an indicator data semantic depth representation modulation coefficient; The logarithmic function value with base 2 of the modulation coefficient of the semantic depth representation of the indicator data is calculated to obtain the implicit representation value of the semantic depth of the indicator data.

6. The marine development suitability evaluation system according to claim 5 is characterized in that: The context association reinforcement unit is used to: Inputting the updated set of indicator data semantic embedding coding vectors into a first semantic association encoder based on a forward LSTM model and a second semantic association encoder based on a backward LSTM model respectively to obtain a set of forward indicator data semantic association coding feature vectors and a set of backward indicator data semantic association coding feature vectors; The set of the forward indicator data semantic association coding feature vectors and the set of the backward indicator data semantic association coding feature vectors are weightedly fused to obtain the set of indicator data enhanced semantic embedding coding vectors.

7. The marine development suitability evaluation system according to claim 6 is characterized in that: The AI ​​adaptive grading module includes: A global context semantic encoding unit, used for performing global context semantic encoding on the set of the indicator data enhanced semantic embedding encoding vectors to obtain an indicator global semantic encoding vector; The adaptive grading decision unit is used to input the global semantic coding vector of the indicator into the classifier-based intelligent decision module to obtain the AI ​​adaptive grading label.

8. The marine development suitability evaluation system according to claim 7 is characterized in that: The global context semantic encoding unit is used to: The set of indicator data reinforced semantic embedding coding vectors is input into a global semantic encoder based on a multi-head self-attention structure to obtain the indicator global semantic coding vector.

9. A method for evaluating the suitability of marine development, characterized in that: include: Extracting marine development suitability assessment documents; Extracting a set of human-judged adaptability classification labels and indicator data from the marine development suitability evaluation document, wherein the indicator data includes evaluation indicators, indicator explanations, and indicator values; Performing semantic embedding coding on each indicator data in the set of indicator data to obtain a set of indicator data semantic embedding coding vectors; Performing semantic feature enhancement processing on the set of indicator data semantic embedding coding vectors to obtain a set of indicator data enhanced semantic embedding coding vectors; Generate AI adaptive grading labels based on global context features of a set of semantic embedding encoding vectors reinforced by the indicator data; The human-judged adaptability grading label and the AI ​​adaptability grading label are compared, and if the two are inconsistent, an early warning prompt is generated.

Images