CN-122022331-A - Multi-dimensional index associated natural resource intelligent auxiliary decision-making method and system

CN122022331ACN 122022331 ACN122022331 ACN 122022331ACN-122022331-A

Abstract

The invention relates to the technical field of intelligent decision making of natural resources, and discloses a method and a system for generating intelligent auxiliary decision making of natural resources by multi-dimensional index association, wherein the method acquires multi-dimensional drainage basin state index data, performs heterogeneous identification and normalization processing, and obtains a feature vector set; the method comprises the steps of generating an index association strength matrix and embedded representation by utilizing a dynamic graph neural network and contrast learning, constructing a decision knowledge map based on scene clustering and map embedding, combining scene matching retrieval and reinforcement learning fusing knowledge prior to form a decision generation model and a resource allocation scheme, outputting a comprehensive decision report by utilizing digital twin and multi-scene simulation, providing an interactive interface and an interpretable report by means of attention visualization and anti-facts reasoning, and realizing dynamic updating of the decision and continuous optimization of the model by utilizing model prediction control and online learning. The invention realizes the intelligent mining and knowledge-guided decision generation of index association, and improves the scientificity and the interpretability of resource configuration decisions.

Inventors

LIU ENGUO
LI QUANXIU
HU AIHUI
CHEN CHANGSHENG
JI NA
SU WEIQIANG
ZHANG YUFEI
ZHOU YUNXIA
HUANG KUNKUN
GUO CAIFANG
JIN CHENJIAN
JIN FEI
GAO GAN

Assignees

杭州臻善信息技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. A natural resource intelligent auxiliary decision-making generation method associated with multidimensional indexes is characterized by comprising the following steps: S1, acquiring original data of a multidimensional drainage basin state index, and adopting a heterogeneous data type identification and multilevel normalization processing method to obtain a multidimensional index feature vector set; S2, acquiring a multidimensional index feature vector set, and acquiring an index association intensity matrix and an index embedded representation by adopting a dynamic graph neural network and contrast learning; s3, receiving an index association intensity matrix and an index embedding representation, and obtaining a decision knowledge graph by adopting a scene clustering and graph embedding method; S4, receiving an index association strength matrix, an index embedding representation and a decision knowledge graph, and obtaining a decision generation model and a resource allocation decision scheme by adopting a scene matching retrieval and a reinforcement learning framework fused with knowledge prior; s5, receiving a resource allocation decision scheme, and obtaining a comprehensive decision report by adopting a digital twin technology and a multi-scenario simulation method; S6, receiving an index association strength matrix and an index embedding representation, a decision generation model and a decision scheme, and a comprehensive decision report, and obtaining an interactive visual interface and an interpretability report by adopting attention visualization and counterfactual reasoning; And S7, receiving real-time monitoring data and effect feedback in the execution process of the decision scheme, and obtaining a dynamically updated decision scheme and iteratively optimized model parameters by adopting a model prediction control and online learning method.
2. The method for generating a natural resource intelligent auxiliary decision associated with a multidimensional index according to claim 1, wherein S1 comprises: Acquiring index data of a hydrological dimension, a resource supply and demand dimension, an ecological environment dimension, a socioeconomic dimension and a policy constraint dimension of a multi-source monitoring network, complementing the index data by adopting an interpolation method according to the deficiency rate, and identifying an abnormal value by adopting a standard deviation criterion; Constructing an index type identification model to divide the index into a numerical continuous index, a numerical discrete index and a category index, and respectively adopting a maximum and minimum value normalization, a standard deviation normalization and a single-heat coding method for treatment; Determining initial weights of indexes based on the information entropy values and the experience scores of field experts, and dynamically adjusting by a coefficient of variation method; extracting time sequence features and spatial distribution features, and splicing to form a multidimensional index feature vector set.
3. The method for generating a natural resource intelligent auxiliary decision associated with a multidimensional index according to claim 1, wherein S2 comprises: Taking each index as a node in the map, establishing a directed edge for index pairs with physical causal relation based on domain knowledge, calculating the pearson correlation coefficient of the index time sequence, and establishing an undirected edge when the absolute value of the correlation coefficient is larger than a preset correlation threshold; designing a multi-layer dynamic graph rolling network, calculating the aggregation weight of neighbor nodes by adopting an attention mechanism, and realizing information layer-by-layer propagation by stacking multi-layer graph rolling operation; modeling the time sequence by adopting a gating circulation unit, and outputting a hidden state fusing the historical association mode and the current association mode; Designing an association mutation detection module, calculating the distance between the current moment and a historical average association strength matrix, and triggering an association mutation identifier when the distance exceeds a dynamic threshold value; And constructing a positive sample pair and a negative sample pair for comparison learning training, and outputting an index association strength matrix and an index embedded representation.
4. The method for generating a natural resource intelligent auxiliary decision associated with a multidimensional index according to claim 1, wherein S3 comprises: extracting drainage basin state characteristics, decision measure content and effect evaluation data from each historical case, and constructing scene feature vectors by combining an index association strength matrix and node embedding vectors; performing cluster analysis on the scene feature vectors by adopting a hierarchical clustering algorithm, and dividing the historical cases into a plurality of typical scene categories; Defining decision scene entities, index feature entities, decision behavior entities and effect evaluation entities, wherein the design scene comprises relations, scene applicable relations, behavior generation relations and index influence relations, and converting historical cases into entities and relations in the map; and (3) training by adopting an embedded model based on translation, and outputting vector representation of the decision knowledge graph.
5. The method for generating a natural resource intelligent auxiliary decision associated with a multidimensional index according to claim 1, wherein S4 comprises: inputting the characteristic vector of the current scene into a scene encoder to output an embedded vector, searching a historical scene with highest cosine similarity in a decision knowledge graph as a candidate matching scene, and searching a decision behavior entity associated with the candidate scene as a reference strategy; Defining a state space and an action space of sequence decision, and adopting a hierarchical decision strategy; defining a reward function comprising a water supply satisfaction reward item, an ecological protection achievement scale reward item, an economic benefit reward item and a decision stability penalty item; Constructing an actor network and a critic network, designing a knowledge guiding attention module, and enabling a final output strategy of the actor network to be a weighted fusion of a self learning strategy and a priori strategy according to knowledge guiding weight; Training by adopting a near-end strategy optimization algorithm, and outputting a decision generation model and a resource allocation decision scheme.
6. The method for generating a natural resource intelligent auxiliary decision associated with a multidimensional index according to claim 1, wherein S5 comprises: constructing a digital twin model comprising a hydrological sub-model, a hydrodynamic sub-model, a water quality sub-model, an ecological sub-model and a socioeconomic sub-model, constructing an error correction model by adopting a data driving method, and assimilating data by adopting a set Kalman filtering method; inputting a resource allocation decision scheme into a digital twin model for coupling simulation; Setting various scenes aiming at weather uncertainty, water-demand uncertainty and emergency uncertainty; And calculating a sensitivity index of the uncertain factors to the decision effect by adopting a sensitivity analysis method, and adjusting decision parameters aiming at weak links to generate a comprehensive decision report.
7. The method for generating a natural resource intelligent auxiliary decision associated with a multidimensional index according to claim 1, wherein S6 comprises: Visualizing the index association map by adopting a force-directed layout algorithm; extracting attention weight vectors of the actor network and drawing the attention weight vectors into a thermodynamic diagram; constructing a counter fact scene for the key decision variables, respectively inputting an original scheme and the counter fact scheme into a digital twin model for simulation, and quantifying the influence brought by the adjustment of the decision variables; and generating a natural language decision interpretation text, and constructing an interactive decision interpretation system.
8. The method for generating a natural resource intelligent auxiliary decision associated with a multidimensional index according to claim 1, wherein S7 comprises: collecting real-time monitoring data through a river basin distributed monitoring network; comparing the real-time monitoring data with the expected state of the decision scheme to calculate the deviation of the key index, and triggering early warning when the deviation exceeds a preset early warning threshold; and (3) performing rolling optimization by adopting a model predictive control method, and regenerating an optimization decision scheme based on the current real-time monitored river basin state and the latest prediction.
9. The method for generating a natural resource intelligent auxiliary decision associated with a multidimensional index according to claim 8, wherein S7 further comprises: supplementing new decision cases formed by input conditions, scheme parameters and actual execution effects of the decision scheme to a historical decision case database; updating the decision knowledge graph based on the increment of the extended case library; and fine adjustment and updating are carried out on the decision generation model parameters by adopting an online learning method.
10. A multi-dimensional index-associated natural resource intelligent decision-making system for performing the steps of a multi-dimensional index-associated natural resource intelligent decision-making method according to any one of claims 1 to 9, comprising: The data acquisition and preprocessing module is used for acquiring the original data of the multidimensional drainage basin state indexes, and acquiring a multidimensional index feature vector set by adopting a heterogeneous data type identification and multilevel normalization processing method; The index association mining module is used for acquiring a multidimensional index feature vector set, and acquiring an index association intensity matrix and an index embedding representation by adopting a dynamic graph neural network and contrast learning; The knowledge graph construction module is used for receiving the index association intensity matrix and the index embedding representation, and obtaining a decision knowledge graph by adopting a scene clustering and graph embedding method; The decision generation module is used for receiving the index association strength matrix, the index embedded representation and the decision knowledge map, and obtaining a decision generation model and a resource allocation decision scheme by adopting a scene matching retrieval and a reinforcement learning framework fused with knowledge prior; the simulation evaluation module is used for receiving a resource allocation decision scheme and obtaining a comprehensive decision report by adopting a digital twin technology and a multi-scenario simulation method; the interpretive analysis module is used for receiving the index association intensity matrix and index embedded representation, a decision generation model and decision scheme and a comprehensive decision report, and adopting attention visualization and anti-facts reasoning to obtain an interactive visual interface and an interpretive report; the dynamic optimization module is used for receiving real-time monitoring data and effect feedback in the execution process of the decision scheme, and obtaining a dynamically updated decision scheme and iteratively optimized model parameters by adopting a model prediction control and online learning method.

Description

Multi-dimensional index associated natural resource intelligent auxiliary decision-making method and system Technical Field The invention relates to the technical field of intelligent decision making of natural resources, in particular to a method and a system for generating intelligent auxiliary decision making of natural resources associated with multidimensional indexes. Background The current natural resource supply and demand contradiction is prominent, and the river basin water resource allocation faces complex challenges such as multidimensional index coupling, space-time non-uniformity, high uncertainty and the like. The traditional method relies on experience rules and a simplified model, is difficult to describe dynamic association among multi-dimensional indexes, has limited capability of coping with systematic risks, and has insufficient interpretability while the traditional data driving method can process a large amount of data, so that practical management application is restricted. In the prior art, the index association analysis method mostly adopts static correlation calculation, so that the characteristic that an index association mode dynamically evolves along with time and a drainage basin state cannot be captured, and the prediction accuracy is reduced when the association mode is mutated under extreme climate conditions. The existing decision generation method lacks systematic organization and effective utilization of historical decision experience, and fails to organically integrate the domain knowledge with data-driven learning, so that the generated decision scheme is insufficient in rationality and feasibility in a complex scene. The existing decision evaluation method is mostly based on a simplified effect prediction model, the influence of multi-process coupling and uncertainty factors is not fully considered, and the robustness of a decision scheme in actual execution is difficult to evaluate accurately. Therefore, an intelligent auxiliary decision-making method capable of dynamically associating and fusing knowledge-guided decisions, supporting comprehensive simulation and interpretable analysis is needed to improve the scientificity, rationality and credibility of resource allocation, and support sustainable utilization and fine management of natural resources. Disclosure of Invention The invention provides a natural resource intelligent auxiliary decision generation method and system for multidimensional index association, which solve the technical problems of inaccurate multidimensional index association relation mining, lack of knowledge guidance for decision generation and poor interpretation of a decision scheme in the related technology. The invention provides a natural resource intelligent auxiliary decision generation method associated with multidimensional indexes, which comprises the following steps: S1, acquiring original data of a multidimensional drainage basin state index, and adopting a heterogeneous data type identification and multilevel normalization processing method to obtain a multidimensional index feature vector set; S2, acquiring a multidimensional index feature vector set, and acquiring an index association intensity matrix and an index embedded representation by adopting a dynamic graph neural network and contrast learning; s3, receiving an index association intensity matrix and an index embedding representation, and obtaining a decision knowledge graph by adopting a scene clustering and graph embedding method; S4, receiving an index association strength matrix, an index embedding representation and a decision knowledge graph, and obtaining a decision generation model and a resource allocation decision scheme by adopting a scene matching retrieval and a reinforcement learning framework fused with knowledge prior; s5, receiving a resource allocation decision scheme, and obtaining a comprehensive decision report by adopting a digital twin technology and a multi-scenario simulation method; S6, receiving an index association strength matrix and an index embedding representation, a decision generation model and a decision scheme, and a comprehensive decision report, and obtaining an interactive visual interface and an interpretability report by adopting attention visualization and counterfactual reasoning; And S7, receiving real-time monitoring data and effect feedback in the execution process of the decision scheme, and obtaining a dynamically updated decision scheme and iteratively optimized model parameters by adopting a model prediction control and online learning method. In a preferred embodiment, the S1 includes: Acquiring index data of a hydrological dimension, a resource supply and demand dimension, an ecological environment dimension, a socioeconomic dimension and a policy constraint dimension of a multi-source monitoring network, complementing the index data by adopting an interpolation method according to the deficiency rate, and identifying an