Big data analysis method and system based on ecological environment monitoring

Abstract

The embodiment of the present application provides a big data analysis method and system based on ecological environment monitoring. By extracting the knowledge network topology of the sample data sequence from the ecological environment monitoring big data and using the neural network model to represent it by graph coding, the complex relationship between ecological factors and environmental monitoring indicators, including positive and negative correlation vectors, is effectively captured. This method not only generates positive and negative graph coding vectors for each network entity, but also accurately estimates the correlation vectors between ecological factors and environmental monitoring indicators based on these coding vectors. By optimizing the neural network model through parameter learning, it is possible to accurately predict the correlation between any ecological factor and environmental monitoring indicator in the sample ecological environment data sequence, significantly improving the accuracy and efficiency of ecological environment monitoring data analysis.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and specifically, to a big data analysis method and system based on ecological environment monitoring. Background Art

[0002] As the global ecological environment becomes increasingly severe, monitoring and protecting the ecological environment has become a global consensus. As an important basis for reflecting the status of the ecological environment, ecological environment monitoring data is huge and complex. How to effectively mine valuable information from these data has become an important topic in current ecological environment research.

[0003] Traditional ecological environment monitoring data analysis methods often only focus on the changing trends of a single ecological factor or environmental monitoring indicator, while ignoring the correlation between them. However, in the actual ecological environment, there are complex interactions and mutual influences between various ecological factors and environmental monitoring indicators. These correlations are crucial for a deep understanding of the evolution of the ecological environment and the formulation of effective protection measures.

[0004] In recent years, with the rapid development of big data technology, new ideas and methods have been provided for the analysis of ecological environment monitoring data. Through big data technology, massive amounts of ecological environment monitoring data can be efficiently stored, processed and analyzed, thereby mining the hidden association rules and patterns. However, how to accurately capture and represent these associations and make effective predictions and decisions based on these associations is still a challenge facing the current field of ecological environment monitoring big data analysis. Summary of the invention

[0005] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the present application is to provide a big data analysis method and system based on ecological environment monitoring.

[0006] In a first aspect, the present application provides a big data analysis method based on ecological environment monitoring, which is applied to a big data analysis system based on ecological environment monitoring, and the method comprises:

[0007] Obtaining a knowledge network topology corresponding to a sample ecological environment data sequence from ecological environment monitoring big data, wherein the sample ecological environment data sequence includes X ecological factors and Y environmental monitoring indicators; the knowledge network topology includes a plurality of network entities and links between network entities, wherein the network entities represent the ecological factors or the environmental monitoring indicators; the links represent association vectors between the ecological factors and the environmental monitoring indicators, wherein the association vectors include positive association vectors or negative association vectors;

[0008] Using a neural network model to perform graph coding representation on the knowledge network topology, generating a positive graph coding vector and a negative graph coding vector for each network entity, wherein the positive graph coding vector is obtained by performing graph coding representation based on the positive association vector in the knowledge network topology; and the negative graph coding vector is obtained by performing graph coding representation based on the negative association vector in the knowledge network topology;

[0009] Determine the target coding data corresponding to each network entity based on the positive graph coding vector and the negative graph coding vector of each network entity through the neural network model, and estimate the association vector between the ecological factors and the environmental monitoring indicators in the sample ecological environment data sequence based on the target coding data corresponding to each network entity through the neural network model;

[0010] The neural network model is subjected to parameter learning based on the correlation vector between the ecological factor and the environmental monitoring indicator. The neural network model after parameter learning is used to estimate the correlation vector between the target ecological factor and the target environmental monitoring indicator in the sample ecological environment data sequence, wherein the target ecological factor is used to represent any ecological factor in the sample ecological environment data sequence, and the target environmental monitoring indicator is used to represent any environmental monitoring indicator in the sample ecological environment data sequence.

[0011] In the second aspect, an embodiment of the present application also provides a big data analysis system based on ecological environment monitoring, the big data analysis system based on ecological environment monitoring includes a processor and a machine-readable storage medium, the machine-readable storage medium stores a computer program, and the computer program is loaded and executed in conjunction with the processor to implement the big data analysis method based on ecological environment monitoring of the above first aspect.

[0012] In combination with the above aspects, the embodiment of the present application effectively captures the complex associations between ecological factors and environmental monitoring indicators, including positive and negative association vectors, by extracting the knowledge network topology of the sample data sequence from the ecological environment monitoring big data and using the neural network model to represent it by graph coding. This method not only generates positive and negative graph coding vectors for each network entity, but also accurately estimates the association vectors between ecological factors and environmental monitoring indicators based on these coding vectors. By optimizing the neural network model through parameter learning, it is possible to accurately predict the association between any ecological factor and environmental monitoring indicator in the sample ecological environment data sequence, significantly improving the accuracy and efficiency of ecological environment monitoring data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 A schematic diagram of a big data analysis method based on ecological environment monitoring provided in an embodiment of the present application;

[0015] Figure 2 A structural schematic block diagram of a big data analysis system based on ecological environment monitoring for implementing the above-mentioned big data analysis method based on ecological environment monitoring provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The following description is intended to enable one of ordinary skill in the art to implement and utilize the present application, and the description is provided in the context of a specific application scenario and its requirements. It will be apparent to one of ordinary skill in the art that various changes may be made to the disclosed embodiments, and that the general principles defined in the present application may be applied to other embodiments and application scenarios without departing from the principles and scope of the present application. Therefore, the present application is not limited to the described embodiments, but should be given the broadest scope equal to the claims.

[0017] Step S110, obtaining a knowledge network topology corresponding to a sample ecological environment data sequence from the ecological environment monitoring big data, wherein the sample ecological environment data sequence includes X ecological factors and Y environmental monitoring indicators. The knowledge network topology includes multiple network entities and links between network entities, wherein the network entities represent the ecological factors or the environmental monitoring indicators. The links represent the association vectors between the ecological factors and the environmental monitoring indicators, wherein the association vectors include positive association vectors or negative association vectors.

[0018] In this embodiment, in an ecological environment monitoring project, the sample ecological environment data sequence covers multiple aspects of information. For example, X = 5 ecological factors, namely forest coverage, biodiversity (measured by the number of species in a specific area), soil fertility (indicator of nutrient content in the soil), water resources and wetland area; Y = 4 environmental monitoring indicators, including air quality index (AQI), water quality acidity (pH value), soil erosion rate and temperature change.

[0019] The server obtains the knowledge network topology corresponding to these data from the huge ecological environment monitoring big data repository. This knowledge network topology is like a complex relationship diagram. For example, the ecological factor of forest coverage, as a network entity, may have a positive correlation vector with the environmental monitoring indicator of air quality index. This means that the increase in forest coverage may have a positive impact on the air quality index, such as forests can absorb pollutants in the air, purify the air, and thus improve air quality.

[0020] For example, there may be a negative correlation vector between the ecological factor of biodiversity and the pH of water quality. This may be because changes in certain specific biological communities will lead to changes in the chemical composition of the water body, thereby affecting the pH of the water quality, and this impact is negative.

[0021] There is also a link between the network entity soil fertility and soil erosion rate. If soil fertility decreases, it may lead to poor vegetation growth, thereby increasing the risk of soil erosion, so there is a negative correlation vector between them. There may be a positive correlation vector between water resources and temperature changes, because changes in water resources (such as the expansion or reduction of water area) will affect the local climate and thus affect the temperature. The ecological factor of wetland area has a link relationship with multiple environmental monitoring indicators. For example, there may be a positive correlation vector with water quality pH, and wetlands can play a role in regulating the pH of water bodies; there is a negative correlation vector with soil erosion rate, and wetland vegetation can consolidate the soil and reduce soil erosion.

[0022] Step S120, using a neural network model to perform graph coding representation on the knowledge network topology, generating a positive graph coding vector and a negative graph coding vector for each network entity, wherein the positive graph coding vector is obtained by performing graph coding representation based on the positive association vector in the knowledge network topology, and the negative graph coding vector is obtained by performing graph coding representation based on the negative association vector in the knowledge network topology.

[0023] Taking the network entity "forest coverage rate" as an example, first, the server obtains the set sub-network extension depth Z=3.

[0024] Taking "forest coverage rate" as the basic network entity, according to the characteristic pointing information represented by the link representing the positive association vector in the knowledge network topology, the neural network model is used to parse out its three-level first extended association network entity from the knowledge network topology. Assume that the first-level extended association network entity contains the air quality index, an environmental monitoring indicator, because the forest coverage rate has a positive association vector with the air quality index, and the existence of forests has an improving effect on air quality. The second-level extended association network entity may include other entities that are associated with the air quality index and indirectly related to the forest coverage rate, such as the improvement of the living environment of some organisms related to the improvement of air quality. The third-level extended association network entity may be some more indirect entities, such as the development of ecotourism brought about by the improvement of the living environment of organisms.

[0025] Based on the "forest coverage rate" and the parsed first extended associated network entities, an active sub-network topology corresponding to the "forest coverage rate" is generated.

[0026] Based on the positive association vectors between the network entities in this positive subnetwork topology, the positive association array corresponding to the "forest coverage rate" is configured. For example, if the positive impact of the forest coverage rate on the air quality index is expressed as 0.8, this value will be recorded in the positive association array.

[0027] Based on the positive association array corresponding to the "forest coverage rate", the positive connectivity array and the positive unit array corresponding to the "forest coverage rate" are configured. Assuming that the positive connectivity array represents the comprehensive relationship between the connectivity and the degree of association between network entities, the positive unit array is used for some basic settings of standardization. Then, based on the positive unit array, the positive connectivity array is standardized and transformed to generate the positive connectivity array after the standardized transformation.

[0028] A basic association array is obtained, and a positive graph encoding vector of "forest coverage" is determined according to the basic association array and the positive connectivity array after normalization transformation.

[0029] For negative correlation vectors, we also use "forest coverage rate" as the basic network entity, and based on the characteristic pointing information represented by the link representing the negative correlation vector in the knowledge network topology, assuming Z=3, and use the neural network model to parse out its three-level second extended correlation network entity from the knowledge network topology. For example, it may be found that forest coverage rate and soil erosion rate have negative correlation vectors in some special cases, such as unreasonable forest development leading to soil exposure and increasing the risk of soil erosion. The parsed second extended correlation network entity may include some factors related to soil erosion rate, such as rainfall intensity (because rainfall intensity affects the degree of soil erosion and has an indirect connection with forest coverage rate in this negative relationship).

[0030] A negative sub-network topology corresponding to the "forest coverage rate" is generated based on the "forest coverage rate" and each of the parsed second extended associated network entities.

[0031] Based on the negative correlation vectors between the network entities in the negative sub-network topology corresponding to the "forest coverage rate", the negative correlation array corresponding to the "forest coverage rate" is configured. If the negative impact of the forest coverage rate on the soil erosion rate is numerically 0.3 in a specific case, it is recorded in the negative correlation array.

[0032] The positive association array corresponding to the "forest coverage rate" is obtained, and the negative image coding vector of the "forest coverage rate" is generated based on the positive association array and the negative association array corresponding to the "forest coverage rate".

[0033] Taking the network entity “biodiversity” as an example, the subnetwork expansion depth Z=3 is set.

[0034] For positive association vectors, taking "biodiversity" as the basic network entity, according to the positive association vector link feature pointing information in the knowledge network topology, the first extended association network entity parsed may include entities related to water quality pH (assuming that the increase in biodiversity helps maintain the ecological balance of water bodies, thus having a positive impact on water quality pH). With the expansion of the hierarchy, the second layer may be the reproduction of aquatic organisms related to water quality, and the third layer may be fishery resources related to these aquatic organisms.

[0035] After generating the positive subnetwork topology corresponding to “biodiversity”, the process of configuring the positive association array, positive connectivity array, and positive unit array, and determining the positive graph encoding vector after standardized transformation is similar to that of “forest coverage rate”.

[0036] For the negative association vector, the parsed second extended association network entity may be related to temperature change (assuming that in some cases changes in biodiversity will affect local microclimates, which in turn have a negative indirect impact on temperature). Construct the negative subnetwork topology, configure the negative association array, and generate the negative graph encoding vector in combination with the positive association array.

[0037] Taking other network entities as an example, for the network entity of soil fertility, in terms of positive association, there may be positive association vectors with environmental monitoring indicators such as vegetation coverage. By analyzing and extending the associated network entity, a positive sub-network topology is constructed, and the related array is configured to determine the positive graph coding vector; in terms of negative association, there may be negative association vectors with soil pollution indicators, and the negative sub-network topology is also constructed and the negative graph coding vector is determined.

[0038] For the network entity of water resources, in terms of positive association vectors, there may be a positive association with aquatic biodiversity. According to the steps, a positive subnetwork topology is constructed and a positive graph coding vector is determined. In terms of negative association vectors, there is a negative association with indicators such as the decline in groundwater levels caused by excessive exploitation of water resources. A negative subnetwork topology is constructed and a negative graph coding vector is determined.

[0039] For the network entity of wetland area, in terms of positive association vectors, there may be a positive association with the number of migratory birds (as an ecological factor), and the positive graph coding vector is determined by constructing a positive subnetwork topology. In terms of negative association vectors, there may be a negative association with agricultural non-point source pollution around the wetland, and the negative subnetwork topology is constructed and the negative graph coding vector is determined.

[0040] Step S130, determining the target coding data corresponding to each network entity based on the positive graph coding vector and the negative graph coding vector of each network entity through the neural network model, and estimating the association vector between the ecological factors and the environmental monitoring indicators in the sample ecological environment data sequence based on the target coding data corresponding to each network entity through the neural network model.

[0041] Taking the network entity "forest coverage rate" as an example, the server determines the attention coefficient of "forest coverage rate" based on the positive graph coding vector and negative graph coding vector of "forest coverage rate" through the neural network model. Assume that the positive graph coding vector represents a coding form of the positive impact of forest coverage rate on the relevant network entity, and the negative graph coding vector represents the coding form of the negative impact. If the value of the positive graph coding vector is large in some dimensions, it means that the positive impact of forest coverage rate accounts for a large proportion, which may cause the attention coefficient to tilt towards the positive side.

[0042] If the attention coefficient of "forest coverage rate" is used as the weighting factor of the negative image coding vector, for example, the attention coefficient is 0.6, the elements in the negative image coding vector are multiplied by 0.6 respectively, and then the product calculation result between the weighting factor and the negative image coding vector is added to the positive image coding vector to generate the target coding data of "forest coverage rate". Alternatively, the attention coefficient is used as the weighting factor of the positive image coding vector, such as multiplying the positive image coding vector and adding it to the negative image coding vector to obtain the target coding data.

[0043] The neural network model estimates the association vector between the forest coverage rate and the air quality index based on the target coded data of "forest coverage rate" and the target coded data of the environmental monitoring index associated with it (such as the air quality index). For example, if there is a certain mathematical relationship (such as product, difference, etc.) between certain elements in the target coded data of "forest coverage rate" and the corresponding elements in the target coded data of the air quality index, the estimated value of the association vector between them can be obtained through the learning and calculation of the neural network.

[0044] Taking the network entity “biodiversity” as an example, the neural network model determines the attention coefficient of “biodiversity”, assuming that the attention coefficient is 0.7 based on the calculation of the positive graph encoding vector and the negative graph encoding vector.

[0045] The target coding data of “biodiversity” is calculated and generated in one of the above two ways (using the attention coefficient as a negative graph coding vector weighting factor or a positive graph coding vector weighting factor).

[0046] Then, based on the target coded data of "biodiversity" and the target coded data of environmental monitoring indicators related to it (such as water quality pH), the correlation vector between biodiversity and water quality pH is estimated. If the elements in the target coded data of biodiversity represent the comprehensive impact of biodiversity on the ecosystem, and the target coded data of water quality pH represents the comprehensive impact related to the chemical properties of the water body, the degree and direction of their correlation (i.e., the correlation vector) can be analyzed through the neural network algorithm.

[0047] Taking other network entities as an example, for the network entity of soil fertility, after determining its attention coefficient, the target coding data is generated, and then the association vector between the target coding data and environmental monitoring indicators such as soil erosion rate is estimated based on the target coding data.

[0048] For the network entity of water resources, after generating the target coded data, the association vectors between it and environmental monitoring indicators such as aquatic biodiversity and temperature change are estimated.

[0049] For the network entity of wetland area, after generating the target coded data, the association vectors between relevant indicators such as the number of migratory birds and soil pollution around wetlands were estimated.

[0050] Step S140, parameters of the neural network model are learned based on the association vector between the ecological factors and the environmental monitoring indicators. The neural network model after parameter learning is used to estimate the association vector between the target ecological factors and the target environmental monitoring indicators in the sample ecological environment data sequence, wherein the target ecological factor is used to represent any ecological factor in the sample ecological environment data sequence, and the target environmental monitoring indicator is used to represent any environmental monitoring indicator in the sample ecological environment data sequence.

[0051] Taking "forest coverage rate" and related network entities as an example, in the knowledge network topology, the "forest coverage rate" that represents ecological factors is the first network entity, and network entities such as the air quality index that have positive correlation vectors with it are extended related network entities. Network entities that have negative correlation vectors with it (such as soil erosion rate in some special cases) or do not have correlation vectors (such as there may be no direct correlation with the number of biological populations in some remote areas) are non-extended related network entities.

[0052] The server obtains the first estimated correlation index between "forest coverage" and the extended associated network entity (such as air quality index). Assuming the value of this index is 0.8, it means that the forest coverage has a strong positive impact on the air quality index. The server obtains the second estimated correlation index between "forest coverage" and the non-extended associated network entity (such as soil erosion rate in special cases). Assuming it is 0.3, it means that the forest coverage has a certain negative impact on the soil erosion rate in special cases.

[0053] The target training error is configured based on the first associated estimation indicator and the second associated estimation indicator. For example, the target training error can be a function of the difference between the two indicators and the prediction result of the neural network model (such as mean square error, etc.).

[0054] The neural network model is optimized with the goal of minimizing the target training error. The server continuously adjusts the parameters of the neural network model, such as weights and biases. If the initial neural network model predicts the correlation vector between forest coverage and air quality index to be 0.7, which is different from the actual first correlation estimation index of 0.8, the model parameters will be adjusted. Similarly, if the correlation vector prediction between forest coverage and soil erosion rate is different from the second correlation estimation index, it will also be adjusted. This continues until the neural network model meets the training termination conditions, such as the target training error is less than a set threshold (such as 0.01).

[0055] Taking "biodiversity" and related network entities as an example, for the first network entity "biodiversity", those that have positive associations with water quality pH, etc. are extended related network entities, and those that have negative associations with temperature changes, etc. or have no direct associations are non-extended related network entities.

[0056] The first estimated index of the association between biodiversity and the extended network entity (such as water quality pH) is obtained, and it is assumed to be 0.7, indicating that biodiversity has a significant positive impact on water quality pH. The second estimated index of the association between biodiversity and the non-extended network entity (such as temperature change) is obtained, and it is assumed to be -0.2, indicating that biodiversity has a slight negative impact on temperature change in some cases.

[0057] Configure the target training error and optimize the neural network model with the goal of minimizing the error. If the neural network model initially predicts the association vector between biodiversity and water quality pH inaccurately, adjust the parameters to gradually approach the actual first association estimation index. The same is true for the prediction of the association vector between biodiversity and temperature change until the training termination condition is met.

[0058] Taking other network entities as examples, for the network entity of soil fertility, it is an extended associated network entity with vegetation coverage, etc., and a non-extended associated network entity with soil pollution indicators, etc. The associated estimation indicators are obtained, the target training error is configured, and the neural network model is optimized in the above manner until the training termination conditions are met.

[0059] For the network entity water resources, the same operation is performed with aquatic biodiversity as an extended associated network entity and groundwater level decline as a non-extended associated network entity.

[0060] For the network entity of wetland area, the number of migratory birds is an extended associated network entity, and the agricultural non-point source pollution around the wetland is a non-extended associated network entity, and relevant parameter learning operations are performed.

[0061] Through such parameter learning, the neural network model after parameter learning can be used to estimate the association vector between the target ecological factor and the target environmental monitoring index in the sample ecological environment data sequence. For example, the association vector between forest coverage and other environmental monitoring indicators that have not participated in the training, or the association vector between biodiversity and new ecological environment related indicators can be accurately estimated, thereby providing valuable analytical basis for the research, protection and management of the ecological environment.

[0062] Based on the above steps, the embodiment of the present application effectively captures the complex associations between ecological factors and environmental monitoring indicators, including positive and negative association vectors, by extracting the knowledge network topology of the sample data sequence from the ecological environment monitoring big data and using the neural network model to represent it by graph coding. This method not only generates positive and negative graph coding vectors for each network entity, but also accurately estimates the association vectors between ecological factors and environmental monitoring indicators based on these coding vectors. By optimizing the neural network model through parameter learning, it is possible to accurately predict the association between any ecological factor and environmental monitoring indicator in the sample ecological environment data sequence, significantly improving the accuracy and efficiency of ecological environment monitoring data analysis.

[0063] In a possible implementation manner, any network entity in the knowledge network topology is represented as network entity i. Step S120 includes:

[0064] Step S121, obtaining a set sub-network extension depth Z, where Z is a positive integer.

[0065] Step S122, taking network entity i as the basic network entity, according to the characteristic pointing information represented by the link representing the positive association vector in the knowledge network topology, using the neural network model to parse out the Z-level first extended associated network entity of the network entity i from the knowledge network topology.

[0066] Step S123: generating an active sub-network topology corresponding to the network entity i based on the network entity i and each of the parsed first extended associated network entities.

[0067] Step S124, configuring the active graph coding vector of the network entity i based on the active sub-network topology corresponding to the network entity i. And,

[0068] Step S125, obtaining a preset sub-network expansion depth Z, where Z is a positive integer.

[0069] Step S126, taking the network entity i as the basic network entity, according to the characteristic pointing information represented by the link representing the negative association vector in the knowledge network topology, using the neural network model to parse out the Z-level second extended associated network entity of the network entity i from the knowledge network topology, and generating the negative sub-network topology corresponding to the network entity i based on the network entity i and each of the parsed second extended associated network entities.

[0070] Step S127: configuring a passive graph coding vector of the network entity i based on the passive sub-network topology corresponding to the network entity i.

[0071] In this embodiment, the server first determines a suitable sub-network extension depth Z, assuming that Z=3. The determination of this value is based on the server's preliminary assessment of the complexity of the ecological environment data relationship and the consideration of computing resources. If the Z value is too large, although the relationship between network entities can be considered more comprehensively, it will increase the computational complexity and time cost; if the Z value is too small, it may not be possible to fully explore the deep-level associations between network entities.

[0072] Taking forest coverage rate as the basic network entity, the server uses a neural network model to parse out the three-level first extended associated network entity from the knowledge network topology based on the feature pointing information represented by the links representing the positive association vectors in the knowledge network topology.

[0073] At the first level, the direct network entity with a positive correlation vector with forest coverage may be the air quality index (AQI). Because forests have functions such as absorbing harmful gases and releasing oxygen, they have a direct positive impact on air quality. Therefore, when forest coverage increases, the air quality index tends to improve, and this association is represented in the knowledge network topology through a positive correlation vector.

[0074] At the second level, the network entities associated with the air quality index and indirectly related to the forest coverage rate may be the population of some biological species that are sensitive to air quality. For example, some birds have high requirements for air quality. When the forest coverage rate increases and improves the air quality, the living environment of these birds becomes better and the population may increase. This indirect association between the forest coverage rate and the bird population established through the air quality index is also parsed under the link feature pointing of the positive association vector.

[0075] At the third level, ecotourism revenue related to the number of bird populations may become an extended association network entity at the third level. When the number of bird populations increases, it may attract more tourists to watch the birds, thereby driving local ecotourism revenue. This series of network entities are gradually expanded and analyzed based on forest coverage through positive association vectors.

[0076] The server generates the positive subnetwork topology corresponding to the forest coverage rate based on the basic network entity of forest coverage rate and the parsed first extended associated network entities (air quality index, bird population, ecotourism income, etc.). This positive subnetwork topology is a relatively independent relationship graph, in which the forest coverage rate is the core node and is connected to other nodes through links representing positive associations. For example, the link weight between the forest coverage rate and the air quality index may be expressed as 0.8 (indicating the relative strength of the positive impact of the forest coverage rate on the air quality index), the link weight between the air quality index and the bird population may be 0.6 (indicating the relative strength of the impact of air quality on the bird population), and the link weight between the bird population and ecotourism income may be 0.5 (indicating the relative strength of the impact of the bird population on ecotourism income). These link weights are determined based on historical data, expert knowledge or previous data analysis, and can be adjusted in the subsequent learning process.

[0077] Based on the positive correlation vectors between the network entities in the positive subnetwork topology corresponding to the forest coverage rate, the server configures the positive correlation array corresponding to the forest coverage rate. For example, in this array, the correlation value between the forest coverage rate and the air quality index is 0.8, the indirect correlation value with the bird population in the second layer may be obtained through matrix operation (assuming it is 0.48, i.e. 0.8×0.6), and the indirect correlation value with the ecotourism income in the third layer may be 0.24 (i.e. 0.8×0.6×0.5), etc.

[0078] Then, the server configures the positive connectivity array and the positive unit array corresponding to the forest coverage rate based on this positive association array. The positive connectivity array represents the comprehensive relationship between the connectivity and the degree of association between network entities, and the positive unit array is used for some basic settings of standardization. For example, the positive unit array may be a square matrix with 1 elements on the diagonal, which is used to standardize the elements in the positive connectivity array to make them comparable and have a suitable value range.

[0079] The server performs a standardized transformation on the positive connection array based on the positive unit array to generate a standardized transformed positive connection array. For example, if an element in the positive connection array indicates that the comprehensive correlation between forest coverage and ecotourism income is 100 (hypothetical value), after the standardized transformation of the positive unit array, it may become 0.5 (the specific value depends on the standardized algorithm and the setting of the data range).

[0080] Finally, the server obtains the basic association array and determines the positive graph coding vector of the forest coverage rate based on the basic association array and the positive connection array after the standardization conversion. The basic association array may contain some pre-set basic association relationship values ​​related to the ecological environment. By performing specific mathematical operations (such as matrix multiplication, addition, etc.) with the positive connection array after the standardization conversion, the positive graph coding vector of the forest coverage rate is obtained. This vector represents various information about the forest coverage rate in terms of positive associations in a form suitable for neural network processing.

[0081] Similarly, the preset sub-network extension depth Z = 3 is obtained. Taking the forest coverage rate as the basic network entity, according to the characteristic pointing information represented by the link representing the negative association vector in the knowledge network topology, the neural network model is used to parse the 3-level second extended association network entity from the knowledge network topology.

[0082] At the first level, the soil erosion rate may have a negative correlation vector with the forest cover rate. In some special cases, such as unreasonable forest development (such as excessive logging, unreasonable road construction, etc.), forest vegetation will be destroyed, resulting in soil exposure, thereby increasing the risk of soil erosion. Therefore, there is a negative correlation vector between forest cover and soil erosion rate.

[0083] At the second level, a network entity that is associated with soil erosion rate and indirectly related to forest cover may be river sediment content. When soil erosion rate increases, more soil will be washed into the river, resulting in an increase in river sediment content.

[0084] At the third level, the siltation of rivers and reservoirs related to the sediment content of rivers may become an extended associated network entity at the third level. The increase in the sediment content of rivers will make rivers and reservoirs more susceptible to siltation, affecting the water storage capacity and service life of the reservoirs.

[0085] The server generates a negative sub-network topology corresponding to the forest coverage rate based on the basic network entity of forest coverage rate and the parsed second extended associated network entities (soil erosion rate, river sediment content, river reservoir siltation, etc.). In this negative sub-network topology, the forest coverage rate is connected to other nodes through links representing negative associations. For example, the link weight between the forest coverage rate and the soil erosion rate may be expressed as -0.3 (indicating the relative strength of the negative impact of the forest coverage rate on the soil erosion rate), the link weight between the soil erosion rate and the river sediment content may be 0.7 (indicating the relative strength of the impact of the soil erosion rate on the river sediment content), and the link weight between the river sediment content and the river reservoir siltation may be 0.6 (indicating the relative strength of the impact of the river sediment content on the river reservoir siltation), etc.

[0086] Based on the negative correlation vectors between each network entity in the negative subnetwork topology corresponding to the forest coverage rate, the server configures the negative correlation array corresponding to the forest coverage rate. For example, in this array, the correlation value between the forest coverage rate and the soil erosion rate is -0.3, the indirect correlation value with the river sediment content in the second layer may be obtained through matrix operation (assuming it is -0.21, that is, -0.3×0.7), and the indirect correlation value with the siltation of the river reservoir in the third layer may be -0.126 (that is, -0.3×0.7×0.6), etc.

[0087] The server obtains the positive association array corresponding to the forest coverage rate (which has been obtained previously in the positive graph coding vector configuration process).

[0088] The negative map coding vector of forest coverage is generated based on the positive correlation array and the negative correlation array corresponding to the forest coverage. This process may involve complex mathematical operations, such as combining the corresponding elements in the positive correlation array and the negative correlation array (such as addition, subtraction or more complex function operations) to obtain a negative map coding vector that can fully reflect the comprehensive impact of forest coverage on positive and negative correlations.

[0089] In a possible implementation, step S124 includes:

[0090] Step S1241 : configuring an active association array corresponding to the network entity i based on active association vectors between network entities in an active sub-network topology corresponding to the network entity i.

[0091] Step S1242, configuring the active connectivity array and the active unit array corresponding to the network entity i based on the active association array corresponding to the network entity i, performing standardization conversion on the active connectivity array based on the active unit array, and generating the active connectivity array after standardization conversion.

[0092] Step S1243, obtaining a basic association array, and determining an active graph coding vector of the network entity i according to the basic association array and the actively connected array after normalization conversion.

[0093] Taking the network entity “biodiversity” as an example, the server takes “biodiversity” as network entity i, and its corresponding active subnetwork topology is a relationship graph constructed based on biodiversity and other related network entities through active association vectors.

[0094] In this positive subnetwork topology, the first-layer network entities with a positive correlation vector with biodiversity may be entities related to water quality pH. This is because the increase in biodiversity helps maintain the ecological balance of water bodies, which has a positive impact on water quality pH. Assuming that the positive impact of biodiversity on water quality pH is expressed as 0.7, then in the positive association array, the value corresponding to the association between biodiversity and water quality pH is 0.7.

[0095] As the levels expand, the second layer may be the reproduction of aquatic organisms related to water quality. Since the improvement of water pH is conducive to the reproduction of aquatic organisms, biodiversity indirectly affects the reproduction of aquatic organisms through the positive impact on water pH. Assuming that the indirect impact of biodiversity on the reproduction of aquatic organisms through water pH is 0.5 (this value is based on relevant ecological knowledge and data analysis), then in the positive association array, the value corresponding to the indirect association between biodiversity and the reproduction of aquatic organisms is 0.7×0.5=0.35.

[0096] The third layer may be fishery resources related to these aquatic organisms. The improvement of aquatic organism reproduction will promote the growth of fishery resources. Assuming that the indirect impact of biodiversity on fishery resources through the previous association is 0.4, then in the positive association array, the value corresponding to the indirect association between biodiversity and fishery resources is 0.7×0.5×0.4=0.14.

[0097] In this way, the server constructs a positive association array corresponding to biodiversity based on the positive association vectors between each network entity in the positive sub-network topology corresponding to biodiversity. This array contains the association values ​​between biodiversity and related network entities at different levels, and comprehensively reflects the impact of biodiversity on other network entities in terms of positive association.

[0098] Next, the server configures the positive connectivity array and the positive unit array based on the positive association array of the biodiversity. The positive connectivity array is intended to represent the comprehensive relationship of connectivity and association degree between network entities.

[0099] For example, in addition to the previously calculated correlation values ​​between biodiversity and water pH, aquatic organism reproduction, and fishery resources, the positive connectivity array may also include a comprehensive representation of other correlations. Assuming that there is another indirect correlation between biodiversity and water pH (through other ecological factors), the correlation value is 0.3, which will also be included in the positive connectivity array.

[0100] The active unit array is used in some basic settings of standardization, which may be a square matrix with all elements on the diagonal set to 1. For example, the dimension of this square matrix is ​​determined by the number of network entities in the active connectivity array. If there are 4 network entities in the active connectivity array (biodiversity, water pH, aquatic biological reproduction, fishery resources), then the active unit array is a 4×4 square matrix with all elements on the diagonal set to 1 and all other elements set to 0.

[0101] Then, the server performs a standardized transformation on the positive connectivity array based on the positive unit array. For example, an element in the positive connectivity array indicates that the comprehensive correlation between biodiversity and fishery resources is 50 (hypothetical value). After the standardized transformation of the positive unit array, it may become 0.2 (the specific value depends on the standardized algorithm and the data range setting). Through such a standardized transformation, all elements in the positive connectivity array are adjusted to a suitable numerical range, generating a standardized transformed positive connectivity array. This standardized transformed positive connectivity array can be more conveniently processed by the neural network model, and also makes the correlation between different network entities comparable in terms of numerical values.

[0102] Then, the server obtains a basic association array, which may contain some pre-set basic association relationship values ​​related to the ecological environment. These values ​​are determined based on a large amount of ecological research, historical data or expert knowledge.

[0103] For example, the basic correlation array may contain the basic correlation value between biodiversity and the stability of the entire ecosystem, assuming it is 0.6. This value reflects the fundamental positive role of biodiversity in the ecosystem, which is different from the correlation relationship obtained through the positive subnetwork topology analysis above. It is a more macro and basic correlation relationship.

[0104] The server determines the biodiversity positive graph encoding vector according to the basic association array and the normalized positive connectivity array. This process may involve specific mathematical operations, such as matrix multiplication, addition, etc.

[0105] Assume that the positively connected array after normalization transformation is represented as matrix A, and the basic association array is represented as matrix B. Through matrix multiplication A×B (here is just an example of a possible operation method), a new matrix is ​​obtained. Then the elements in this new matrix are adjusted or combined (for example, some elements are added, the average is taken, etc.), and finally the positive graph encoding vector of biodiversity is obtained. This positive graph encoding vector represents various information about biodiversity in terms of positive association in a form suitable for neural network processing. It integrates the information of biodiversity at different levels and different types of associations, and provides comprehensive input data for subsequent neural network model processing.

[0106] Through the above steps, this embodiment takes biodiversity as an example to show in detail how to configure the active graph coding vector based on the active sub-network topology corresponding to the network entity. For other network entities, such as forest coverage, soil fertility, water resources, wetland area, etc., the active graph coding vector can also be configured in a similar manner, but there will be differences in specific association relationships, association values, and network entities, which depends on their respective characteristics and relationships in the ecological environment.

[0107] In a possible implementation, step S127 includes:

[0108] Step S1271: configuring a negative association array corresponding to the network entity i based on the negative association vectors between the network entities in the negative sub-network topology corresponding to the network entity i.

[0109] Step S1272, obtaining the active association array corresponding to the network entity i.

[0110] Step S1273: Generate a negative graph coding vector of the network entity i based on the positive association array and the negative association array corresponding to the network entity i.

[0111] In this embodiment, “biodiversity” is taken as network entity i, and its corresponding negative sub-network topology is a relationship graph constructed based on biodiversity and other related network entities through negative association vectors.

[0112] In this negative subnetwork topology, the first-layer network entities with negative correlation vectors with biodiversity may be entities related to temperature changes. This is because in some cases, changes in biodiversity can affect local microclimates, which in turn have a negative indirect impact on temperature. For example, when biodiversity decreases, it may disrupt the balance of the ecosystem, causing changes in vegetation coverage, etc., which in turn affects the surface's absorption and reflection of solar radiation, leading to temperature changes.

[0113] Based on this correlation, the server will determine the numerical value corresponding to the negative correlation vector between biodiversity and temperature change. This value is based on ecological research, historical data and analysis of the ecosystem, and put it into the negative correlation array.

[0114] As the layers expand, the second layer may be the survival adaptability of other organisms in the ecosystem related to temperature changes. Due to temperature changes, the living environment of some organisms has changed, and they may have difficulty adapting to new temperature conditions. Biodiversity indirectly affects the survival adaptability of these organisms through the negative impact on temperature. The server determines a value based on the degree of this indirect impact and puts it into the negative association array. This value reflects the negative impact of biodiversity on the survival adaptability of other organisms through temperature changes.

[0115] The third layer may be the service functions of the ecosystem related to these organisms. When the survival adaptability of organisms is affected, the service functions of the entire ecosystem will also be implicated. For example, some organisms may undertake functions such as pollination and decomposition of organic matter in the ecosystem. The impact on their survival will lead to a decrease in the efficiency of these functions. The server determines a value based on the degree of this indirect impact and puts it into the negative association array.

[0116] In this way, the server constructs a negative correlation array corresponding to biodiversity based on the negative correlation vectors between each network entity in the negative sub-network topology corresponding to biodiversity. This array contains the negative correlation values ​​between biodiversity and related network entities at different levels, and fully reflects the impact of biodiversity on other network entities in terms of negative correlation.

[0117] For the network entity "biodiversity", the server has obtained its corresponding positive association array in the process of constructing the positive graph coding vector. This positive association array reflects the impact of biodiversity on other network entities in terms of positive association.

[0118] For example, the positive association array contains the positive association values ​​between entities related to biodiversity and water quality pH. Because the increase in biodiversity helps maintain the ecological balance of water bodies, it has a positive impact on water quality pH, and this positive impact value is recorded in the positive association array.

[0119] It also includes the indirect positive impact of biodiversity on the reproduction of aquatic organisms through water pH, as well as the further indirect positive impact on fishery resources. This positive correlation array is constructed by comprehensively considering the positive impact of biodiversity at different levels and in different aspects. It is an array that comprehensively reflects the positive correlation of biodiversity.

[0120] After obtaining the positive association array and the negative association array of biodiversity, the server starts to generate a negative graph coding vector.

[0121] This process is a complex operation that takes into account both the positive and negative associations of biodiversity. For example, some combination of operations may be performed on the corresponding elements in the positive association array and the negative association array. If an element in the positive association array represents the positive impact of biodiversity on water pH, and the corresponding element in the negative association array represents the negative impact of biodiversity on temperature change, the server may perform weighted addition or other function operations on these two elements based on the overall logic of the ecosystem and the needs of the neural network model.

[0122] From a more macroscopic perspective, the positive association array reflects the positive promotion of biodiversity to some network entities in the ecosystem, while the negative association array reflects the negative impact of biodiversity on other network entities. By combining these two arrays, a comprehensive vector that fully reflects the positive and negative impacts of biodiversity can be obtained, namely the negative graph coding vector.

[0123] This negative graph coding vector enables the neural network model to more fully understand the complex role of biodiversity in the entire ecological environment. For other network entities, such as forest coverage, soil fertility, water resources, wetland area, etc., the server can also build a negative association array based on the negative association vectors in the corresponding negative sub-network topology, obtain the positive association array, and then generate a negative graph coding vector in a similar way, but there will be differences in the specific association relationship, association value, and network entity, which depends on their respective characteristics and relationships in the ecological environment.

[0124] Thus, through the above steps, this embodiment takes biodiversity as an example to illustrate in detail the complete process of configuring a negative graph coding vector based on a negative sub-network topology corresponding to a network entity.

[0125] In a possible implementation manner, the expansion depth of the active sub-network topology corresponding to the network entity i is Z, where Z is a positive integer. Step S1243 includes:

[0126] Step S1243-1, taking j=0, determining the j+1th level graph coding vector of the network entity j based on the basic association array and the normalized positive connectivity array. j is an integer less than Z.

[0127] Step S1243-2, take j=j+1, if j=j+1 and is less than Z, then optimize the j+1th level graph coding vector according to the normalized transformed active connectivity array. If j=j+1 and is equal to Z, then determine the active graph coding vector of the network entity i based on the optimized Zth level graph coding vector.

[0128] This embodiment continues to use "forest coverage rate" as network entity i and sets the sub-network expansion depth Z (where Z is a certain positive integer, for example, a value determined based on a previous evaluation of the complexity of the ecological environment data relationship and factors such as computing resources).

[0129] When j=0, the first level graph encoding vector is constructed. The basic association array contains some pre-set basic association relationship values ​​related to the ecological environment, which reflect the fundamental positive role of forest coverage in the entire ecosystem. For example, the basic association array may contain the correlation relationship value between forest coverage and the stability of the entire ecosystem, which is based on a large amount of ecological research and historical data, reflecting the basic contribution of forests to the stability of the ecosystem in terms of maintaining water and soil, regulating climate, etc.

[0130] The standardized transformed positive connectivity array represents the comprehensive relationship between the connectivity and correlation between forest coverage and other network entities (such as the air quality index, the living environment of certain organisms related to the improvement of air quality, etc.), and these relationships have been standardized to make them comparable and have an appropriate numerical range.

[0131] The server determines the first-level graph encoding vector based on specific mathematical operations. For example, the basic association array may be operated on by matrix multiplication with some elements in the positively connected array after normalization transformation. If a row element in the basic association array represents a basic association relationship such as forest coverage and ecosystem stability, and a column element in the positively connected array after normalization transformation represents an association relationship between forest coverage and first-level extended association network entities such as air quality index, then a new numerical vector is obtained through the matrix multiplication operation of the two, and this vector is part of the first-level graph encoding vector. This process comprehensively considers the basic positive role of forest coverage and its association relationship with the first-level extended association network entities to construct the first-level graph encoding vector.

[0132] When j=1 (because j=0 before and j=j+1 operation was performed), the optimization of the second-level graph encoding vector begins. At this time, the elements in the normalized positively connected array reflect the association between the forest coverage rate and the entities of the second-level extended association network (such as other entities related to the air quality index and indirectly related to the forest coverage rate, such as entities related to the improvement of the living environment of some organisms related to the improvement of air quality).

[0133] The server optimizes the second-level graph encoding vector based on these associations. For example, suppose that some elements in the first-level graph encoding vector reflect the positive impact of forest coverage on the air quality index, and in the second level, the improvement in air quality brings about a better living environment for organisms, which in turn affects the relationship between forest coverage and the entire ecosystem. The server adjusts the corresponding elements in the first-level graph encoding vector by analyzing the elements related to this indirect relationship in the positively connected array after the standardized transformation. This adjustment may be achieved through weighted calculations or other mathematical operations. For example, if the degree of impact of improved air quality on the living environment of organisms is represented as a certain value in the positively connected array after the standardized transformation, then this value can be multiplied by the relevant elements in the first-level graph encoding vector, and then other relevant values ​​can be added or subtracted to optimize the second-level graph encoding vector.

[0134] As j increases (as long as j=j+1 and less than Z), the server will continue to optimize the graph encoding vectors at each level based on the standardized positively connected array. For example, when j=2, for the third-level extended association network entities (such as the development of ecotourism due to the improvement of the biological living environment), the server will further analyze the indirect impact of forest coverage on the development of ecotourism through the previous two levels of association, and optimize the third-level graph encoding vector based on the corresponding elements in the standardized positively connected array. This process will continue to dig deeper into the complex relationship between forest coverage and different levels of extended association network entities, and gradually optimize the graph encoding vector to make it more comprehensive and accurate. Reflect the various information of forest coverage in terms of positive associations.

[0135] When the value of j increases continuously and reaches Z (for example, Z=3), the construction and optimization of the map coding vectors of all levels (from level 1 to level Z) have been completed. The server determines the positive map coding vector of "forest coverage" based on the optimized map coding vector of level Z.

[0136] This process may be the final integration and adjustment of the Z-th level graph coding vector. For example, the elements in the Z-th level graph coding vector may be combined, such as adding some elements, taking the average value, or performing other more complex function operations. Because the Z-th level graph coding vector has integrated all the positive association information from the basic association relationship to the extended association network entities at each level, through this final operation, a vector that can fully and accurately represent the information of forest coverage in terms of positive association can be obtained. This vector is the positive graph coding vector of "forest coverage". This positive graph coding vector represents the positive association relationship of forest coverage in the entire ecological environment in a form suitable for neural network processing, providing an important data basis for subsequent neural network model processing (such as estimating the association vector between forest coverage and other network entities).

[0137] For other network entities, such as biodiversity, soil fertility, water resources, wetland area, etc., the server can also follow a similar method, according to their respective basic association arrays, the actively connected arrays after standardized conversion, and the set sub-network extension depth Z, by constructing and optimizing graph coding vectors at different levels, and finally determine their respective active graph coding vectors. The only difference is the specific association relationship, network entity level content, and calculation value, which depends on their respective characteristics and relationships in the ecological environment.

[0138] In a possible implementation, step S1273 includes:

[0139] Step S1273-1, obtaining a normalized converted active connectivity array corresponding to the network entity i. The normalized converted active connectivity array is determined based on the active association array.

[0140] Step S1273-2: configuring the negative connectivity array and the negative unit array corresponding to the network entity i based on the negative association array corresponding to the network entity i.

[0141] Step S1273-3, performing a normalization conversion on the negative connected array based on the negative unit array to generate a normalized negative connected array.

[0142] Step S1273-4, obtaining a basic association array, and determining a negative graph coding vector of the network entity i according to the basic association array, the positive connection array after the normalization conversion, and the negative connection array after the normalization conversion.

[0143] This embodiment continues to use "forest coverage rate" as the network entity i. Previously, in the process of constructing the active graph coding vector of forest coverage rate, the active connection array and the active unit array have been configured based on the active association array, and the active connection array has been standardized to obtain the standardized transformed active connection array.

[0144] This standardized positive connectivity array fully reflects the comprehensive relationship between the connectivity and degree of association between forest coverage and other network entities in terms of positive association. For example, it includes the positive association between forest coverage and air quality index (the comprehensive impact of increased forest coverage on the improvement of air quality index), as well as the positive association between forest coverage and other indirectly related entities (such as the increase in the number of certain air quality-sensitive biological populations, which in turn drives ecotourism benefits, etc.) through the air quality index. These relationships have been standardized and the values ​​are within an appropriate range, which is convenient for subsequent calculations and analysis.

[0145] For forest coverage, the corresponding negative correlation array has been constructed, which contains the relationship values ​​between forest coverage and other network entities in terms of negative correlation.

[0146] Based on this negative association array, the server starts to configure the negative connectivity array. The negative connectivity array is intended to represent the connectivity and comprehensive association between forest coverage and other network entities in terms of negative association. For example, if forest coverage and soil erosion rate are negatively correlated in some special cases (such as unreasonable forest development leading to soil exposure and increased soil erosion risk), then this direct negative association value will be reflected in the negative connectivity array. At the same time, if an increase in soil erosion rate leads to an increase in river sediment content, and an increase in river sediment content leads to an increase in river reservoir siltation, these indirect negative association values ​​will also be included in the negative connectivity array.

[0147] The negative unit array is used as the basic setting for standardizing the negative connected array. Similar to the positive unit array, the negative unit array may be a square matrix with specific values ​​(such as 1) on the diagonal, and its dimension is determined by the number of network entities in the negative connected array. For example, if the negative connected array contains negative correlations related to the four network entities of forest coverage, soil erosion rate, river sediment content, and river reservoir siltation, then the negative unit array is a 4×4 square matrix with 1 on the diagonal and 0 on the other elements. This negative unit array will be used for the subsequent standardization transformation operation of the negative connected array.

[0148] The server uses the negative unit array to perform a standardized conversion on the negative connected array. Taking the relationship between forest coverage and soil erosion rate as an example, assume that in the negative connected array, the value of the direct negative impact of forest coverage on soil erosion rate is a certain original value (this value is based on ecological research, historical data, etc.). Under the action of the negative unit array, this original value will be converted to a standardized value.

[0149] For example, the original value may not be directly comparable in different correlation comparisons due to data magnitude or other factors. Through the standardization transformation of the negative unit array (which may involve multiplication, division, etc., depending on the set standardization algorithm), this value is converted to a value within the range of 0 to 1 or other appropriate values. Similar standardization transformation operations are also performed for the values ​​of forest coverage that are indirectly related to river sediment content, river reservoir siltation, etc. through soil erosion rate. Finally, the negative connection array after standardization transformation is obtained. After standardization, the values ​​in this array can be more conveniently combined with other arrays (such as positive connection arrays, basic association arrays, etc.), and at the same time, the relationship between forest coverage and other network entities in terms of negative association is more comparable in terms of values.

[0150] The server obtains a basic association array, which includes some pre-set basic association relationship values ​​related to the ecological environment. These values ​​reflect the basic association characteristics of the forest coverage rate in the entire ecosystem, such as the relationship value between the forest coverage rate and the basic structural stability of the ecosystem.

[0151] Then, the negative map encoding vector of forest cover is determined based on the basic correlation array, the normalized positive connection array and the normalized negative connection array. This process involves complex mathematical operations.

[0152] For example, the corresponding elements in the normalized positive connected array and the normalized negative connected array may be combined (such as addition, subtraction, or addition according to a certain weight, etc.) to obtain an intermediate result array. Then, the intermediate result array is further operated with the basic association array, such as matrix multiplication or other custom function operations. The purpose of these operations is to comprehensively consider the basic association characteristics of forest coverage, the comprehensive relationship of positive associations, and the comprehensive relationship of negative associations, so as to obtain a vector that can fully reflect the complex association relationship (including positive and negative aspects) of forest coverage in the entire ecosystem. This vector is the negative map encoding vector of forest coverage.

[0153] For other network entities, such as biodiversity, soil fertility, water resources, wetland area, etc., the server can also follow similar steps to determine their respective negative graph coding vectors through corresponding operations based on their respective positive association arrays, negative association arrays, positive connectivity arrays after standardized conversion, negative unit arrays, and basic association arrays. The only difference is the specific association relationships, association values, and network entity level content, which depends on their respective characteristics and relationships in the ecological environment.

[0154] In a possible implementation manner, the extended depth of the passive sub-network topology corresponding to the network entity i is Z, where Z is a positive integer. Step S1273-4 includes:

[0155] Step S1273-41, taking j=0, determining the j+1th level graph coding vector of the network entity j based on the basic association array and the normalized positive connectivity array. j is an integer less than Z.

[0156] Step S1273-42, take j=j+1, if j=j+1 and less than Z-1, optimize the j+1th level graph coding vector according to the positive connected array after the normalization conversion. If j=j+1 and equal to Z-1, optimize the Z-1th level graph coding vector based on the negative connected array after the normalization conversion to generate the negative graph coding vector of the network entity i.

[0157] This embodiment continues to use "forest coverage rate" as the network entity i, and determines the expansion depth Z of the negative sub-network topology (Z is a positive integer, and this value is determined based on the server's assessment of the complexity of the ecological environment data relationship related to forest coverage rate and computing resources and other factors).

[0158] When j=0, the server starts to build the first-level graph encoding vector. The basic association array contains some basic correlation values ​​of forest coverage in the entire ecosystem, which reflect the fundamental role of forest coverage. For example, the basic association array may contain the relationship between forest coverage and the stability of the entire ecosystem, which is determined by a large amount of ecological research, historical data and expert knowledge. Forest coverage has a fundamental impact on maintaining water and soil, regulating climate, etc., and these impacts are reflected in the basic association array.

[0159] The standardized transformed positive connectivity array was obtained in the previous process of constructing the forest coverage positive map encoding vector. It reflects the connectivity and comprehensive correlation between forest coverage and other network entities in terms of positive correlation. These relationships have been standardized and have a suitable numerical range for easy calculation and comparison.

[0160] The server determines the first-level graph coding vector through specific mathematical operations. For example, matrix multiplication is used to multiply the elements in the basic association array with the corresponding elements in the normalized positively connected array, and then these product results are combined according to certain rules (such as addition or other custom combinations) to obtain a new numerical vector, which is the first-level graph coding vector. This process comprehensively considers the basic association characteristics of forest coverage and its relationship with other network entities in terms of positive association, thereby constructing the first-level graph coding vector.

[0161] When j=1 (because j=0 before and j=j+1 operation was performed), the optimization of the second-level graph encoding vector begins. At this time, since j=1 is less than Z-1 (assuming Z is large enough, such as greater than 2), the server performs optimization operations based on the actively connected array after normalization transformation.

[0162] In the case of forest cover, the normalized positive connectivity array contains the multi-level relationships between forest cover and other network entities in terms of positive associations. For example, forest cover is positively associated with the air quality index, and the improvement of the air quality index will affect the living environment of certain organisms, which in turn affects related situations such as ecotourism.

[0163] For the optimization of the second-level graph encoding vector, the server analyzes the relationship between the elements in the first-level graph encoding vector and the elements related to the second level in the positively connected array after the normalization transformation. Assume that some elements in the first-level graph encoding vector reflect the positive impact of forest coverage on the air quality index, and in the second level, the improvement of the air quality index on the biological living environment is indirectly related to the forest coverage. Based on this indirect correlation, the server obtains the relevant elements from the positively connected array after the normalization transformation and adjusts the corresponding elements in the first-level graph encoding vector. This adjustment may be achieved through weighted calculation or other mathematical operations. For example, if the degree of impact of air quality on the biological living environment is represented as a certain value in the positively connected array after the normalization transformation, then this value can be multiplied with the relevant elements in the first-level graph encoding vector, and then other relevant values ​​can be added or subtracted to optimize the second-level graph encoding vector.

[0164] As j increases (as long as j=j+1 and less than Z-1), the server will continue to optimize the graph encoding vectors of each level based on the standardized positively connected array. For example, when j=2 (assuming Z is greater than 3), for the optimization of the third-level graph encoding vector, the server will further analyze the impact of forest coverage on more indirect network entities (such as other situations related to ecotourism due to the improvement of the living environment of organisms) through the previous two levels of association, and optimize the third-level graph encoding vector based on the corresponding elements in the standardized positively connected array. This process continuously digs deeper into the complex relationship between forest coverage and different levels of extended associated network entities, and gradually optimizes the graph encoding vector to make it more comprehensive and accurate. Reflect the various information of forest coverage in terms of positive associations.

[0165] When j increases continuously to be equal to Z-1 (for example, assuming Z=3, this step is reached when j=2), the server starts to optimize the Z-1 level graph coding vector based on the normalized transformed negative connectivity array to generate a negative graph coding vector of forest coverage.

[0166] The negative connectivity array after standardized transformation reflects the connectivity and comprehensive correlation between forest coverage and other network entities in terms of negative correlation, and these relationships have been standardized. For example, in the negative subnetwork topology of forest coverage, there is a negative correlation between forest coverage and soil erosion rate (such as unreasonable forest development leading to soil exposure, increasing the risk of soil erosion), soil erosion rate and river sediment content are related, and river sediment content is related to river reservoir siltation. These relationships are represented by corresponding numerical values ​​in the negative connectivity array after standardized transformation.

[0167] The server uses this standardized negative connectivity array to optimize the Z-1 level graph coding vector. For example, specific mathematical operations (such as multiplication, addition, or other operations defined according to ecological relationships) are performed on the elements in the Z-1 level graph coding vector and the corresponding elements in the standardized negative connectivity array. Assuming that some elements in the Z-1 level graph coding vector reflect the comprehensive impact of forest coverage on a certain network entity through the previous level association relationship, and the elements in the standardized negative connectivity array reflect the indirect impact of negative associations such as forest coverage and soil erosion rate on this network entity, through these operations, information on negative associations can be integrated into the graph coding vector.

[0168] After such optimization operation, the result is the negative graph coding vector of forest coverage. This negative graph coding vector integrates the basic correlation characteristics of forest coverage, the information of positive correlation (through the previous operation based on the positive connection array) and the information of negative correlation (through the final optimization operation based on the negative connection array), and represents the complex correlation relationship of forest coverage in the entire ecological environment in a comprehensive way.

[0169] For other network entities, such as biodiversity, soil fertility, water resources, wetland area, etc., the server can also follow similar steps, according to their respective basic association arrays, positive connectivity arrays after standardization conversion, negative connectivity arrays after standardization conversion, and the set negative sub-network topology expansion depth Z, by constructing and optimizing graph coding vectors at different levels, and finally determining their respective negative graph coding vectors, but there will be differences in specific association relationships, association values, and network entity level content, which depends on their respective characteristics and relationships in the ecological environment.

[0170] In a possible implementation manner, any network entity in the knowledge network topology is represented as network entity i. Step S130 includes:

[0171] Step S131, determining the attention coefficient of the network entity i according to the positive graph coding vector and the negative graph coding vector of the network entity i through the neural network model.

[0172] Step S132, using the attention coefficient of the network entity i as a weighting factor of the negative graph coding vector of the network entity i, and adding the product calculation result between the weighting factor and the negative graph coding vector of the network entity i to the positive graph coding vector to generate the target coding data of the network entity i. Or,

[0173] Step S133, using the attention coefficient of the network entity i as a weighting factor of the active graph coding vector of the network entity i, and adding the product calculation result between the weighting factor and the active graph coding vector of the network entity i and the passive graph coding vector to generate the target coding data of the network entity i.

[0174] In this embodiment, "forest coverage rate" is used as network entity i. The positive graph coding vector of forest coverage rate is obtained through a series of complex construction processes. This vector comprehensively represents the impact of forest coverage rate on other network entities in terms of positive association. For example, it includes information such as the positive impact of forest coverage rate on air quality index (such as the improvement of air quality by forests absorbing harmful gases and releasing oxygen), as well as the positive chain reaction generated by the improvement of air quality index on the living environment of related organisms, ecotourism, etc.

[0175] The negative map encoding vector of forest cover is also constructed through multiple steps, which reflects the relationship between forest cover and other network entities in terms of negative associations. For example, when forests are not rationally developed, the reduction of forest cover will lead to an increase in soil erosion rate, which in turn affects the negative relationship such as river sediment content and river reservoir siltation.

[0176] The neural network model determines the attention coefficient of forest coverage based on these two graph coding vectors. The neural network model has been previously learned and trained with a large amount of data and can understand various ecological association information in positive and negative graph coding vectors. For example, the model may analyze the positive contribution of forest coverage to ecosystem stability in the positive graph coding vector, and the negative impact of reduced forest coverage on soil erosion risk in the negative graph coding vector.

[0177] If the elements in the positive graph coding vector indicate that forest coverage has a strong positive impact in many aspects, such as maintaining biodiversity and regulating local climate, while the elements in the negative graph coding vector show that forest coverage has a relatively small negative impact on soil erosion and other aspects, and the circumstances in which it occurs are relatively special (such as only occurring in large-scale unreasonable development), then the neural network model may determine a larger attention coefficient that leans toward the positive side. This attention coefficient is a trade-off that comprehensively considers the positive and negative effects of forest coverage in the entire ecological environment, reflecting the dominant nature (positive or negative) tendency of forest coverage in the overall correlation relationship.

[0178] Assume that the neural network model determines the attention coefficient of the forest coverage rate to be a certain value (the specific number is not given here). The server uses this attention coefficient as a weighting factor for the forest coverage rate negative map encoding vector.

[0179] The negative graph coding vector of forest cover contains various relationship values ​​of forest cover and other network entities in terms of negative association. For example, an element in the negative graph coding vector represents the negative correlation value between forest cover and soil erosion rate. When the attention coefficient is used as a weighting factor and multiplied by the negative graph coding vector, this negative correlation value will be adjusted according to the size of the attention coefficient. If the attention coefficient is large (indicating that positive influences dominate), then the elements in the negative graph coding vector will become relatively small after multiplication, which means that the weight of negative influences in the whole is reduced.

[0180] Then the product of the weighted factor and the negative map coding vector is added to the positive map coding vector. The elements in the positive map coding vector represent the various relationship values ​​of the forest coverage in the positive association, such as the positive impact value on the air quality index. The addition process is to combine the negative impact and positive impact after weighted adjustment. For example, the positive impact value of the forest coverage on the air quality index is added to the negative impact value of the forest coverage on the soil erosion rate after weighted adjustment. This process will comprehensively consider the positive and negative impacts of the forest coverage in the entire ecological environment, and the result is the target coding data of the forest coverage. This target coding data represents the overall correlation of the forest coverage in the ecological environment in a comprehensive way, taking into account both positive and negative impacts, and reasonably weighting the two according to the attention coefficient.

[0181] Taking the forest coverage rate as an example, the server uses the attention coefficient determined by the neural network model as a weighting factor of the forest coverage rate positive graph encoding vector.

[0182] The elements in the positive graph coding vector of forest coverage reflect the various relationship values ​​of forest coverage in terms of positive associations. When the attention coefficient is multiplied by the positive graph coding vector, the elements in the positive graph coding vector are adjusted according to the attention coefficient. If the attention coefficient is large, the elements in the positive graph coding vector will become relatively large after multiplication, which means that the weight of the positive impact in the whole increases.

[0183] Then, the product calculation result between the weighting factor and the positive graph coding vector is added to the negative graph coding vector. The elements in the negative graph coding vector represent the relationship values ​​of the forest coverage rate in the negative correlation, such as the negative correlation value with the soil erosion rate. This addition process also combines the adjusted positive and negative effects. For example, the positive impact value of the forest coverage rate on the air quality index after weighted adjustment is added to the negative impact value of the forest coverage rate on the soil erosion rate to obtain the target coding data of the forest coverage rate. This method also expresses the overall correlation relationship of the forest coverage rate in the ecological environment in a comprehensive way, and reasonably distributes the weights of the positive and negative effects through the attention coefficient. Although the calculation order is different from the previous method, the ultimate goal is to generate a target coding data that can fully reflect the comprehensive correlation relationship of network entities in the ecological environment.

[0184] For other network entities, such as biodiversity, soil fertility, water resources, wetland area, etc., the server can also follow similar steps to determine their respective attention coefficients through the neural network model, and then calculate the target coding data according to one of the two methods mentioned above, but there will be differences in the specific positive graph coding vectors, the content of the negative graph coding vectors, and the determination of the attention coefficients, which depends on their respective characteristics and relationships in the ecological environment.

[0185] In a possible implementation, the knowledge network topology includes a first network entity representing ecological factors and a second network entity representing environmental monitoring indicators, wherein the second network entity includes an extended associated network entity and a non-extended associated network entity of the first network entity. The extended associated network entity includes a network entity having a positive association vector with the first network entity, and the non-extended associated network entity includes a network entity having a negative association vector with the first network entity or having no association vector. The association vector is characterized by an association estimation indicator.

[0186] Step S140 includes:

[0187] Step S141: acquiring a first association estimation indicator between the first network entity and the extended association network entity, and a second association estimation indicator between the first network entity and the non-extended association network entity.

[0188] Step S142: configuring a target training error based on the first association estimation indicator and the second association estimation indicator.

[0189] Step S143, optimizing the neural network model with minimizing the target training error as the training goal until the neural network model meets the training termination condition.

[0190] In this embodiment, "forest coverage rate" is used as the first network entity. In the knowledge network topology related to forest coverage rate, there is a second network entity representing the environmental monitoring index, wherein the extended associated network entity is a network entity with a positive association vector with the forest coverage rate, and the non-extended associated network entity is a network entity with a negative association vector with the forest coverage rate or does not have an association vector.

[0191] For the extended association network entity, taking the air quality index as an example, the server obtains the first association estimation index between the forest coverage rate and the air quality index. There is a positive association vector between the forest coverage rate and the air quality index, which means that changes in the forest coverage rate will have a positive impact on the air quality index. The first association estimation index is used to characterize the degree of this positive association vector. For example, forests have a significant improvement effect on air quality by absorbing harmful gases and releasing oxygen, and the degree of this improvement is represented by the first association estimation index. This indicator is determined based on a large amount of ecological environment monitoring data, long-term research results, and professional ecological analysis. It may comprehensively consider the corresponding changes in the air quality index under different levels of forest coverage. For example, in areas with high forest coverage, the air quality index is usually at a good level. The quantitative result of this corresponding relationship is the first association estimation index.

[0192] For non-extended associated network entities, taking soil erosion rate as an example (assuming that there is a negative correlation vector between forest coverage and soil erosion rate in some special cases), the server obtains the second correlation estimation index between forest coverage and soil erosion rate. When forest coverage decreases due to unreasonable development (such as excessive logging, unreasonable road construction, etc.), soil erosion rate will increase, and the degree of this correlation is expressed by the second correlation estimation index. This indicator is also based on actual ecological monitoring data. For example, by conducting long-term monitoring of soil erosion in areas with different forest coverage rates, analyzing the relationship between changes in forest coverage and changes in soil erosion rate, the second correlation estimation index is determined. In addition, there are some network entities that do not have correlation vectors with forest coverage. For example, the number of biological populations in some remote areas may not be directly related to forest coverage, but within the scope of the overall non-extended associated network entity, this situation will also be comprehensively considered when determining the second correlation estimation index.

[0193] The target training error is configured based on the first association estimation index (the association index between the forest coverage rate and the extended association network entity, such as the air quality index) and the second association estimation index (the association index between the forest coverage rate and the non-extended association network entity, such as the soil erosion rate) obtained previously.

[0194] The target training error is a measure of the difference between the prediction of the neural network model and the actual association (represented by the first association estimate and the second association estimate). For example, the mean square error (MSE) can be used as a calculation method for the target training error. Assuming that the neural network model predicts one value for the association vector between forest coverage and air quality index, and the actual first association estimate is another value, the square of the difference between the two is part of the calculation of the mean square error. Similarly, the square of the difference between the prediction of the association vector between forest coverage and soil erosion rate and the second association estimate is also part of the calculation of the mean square error. The target training error is obtained by adding these two parts according to certain weights (the weights can be determined according to the actual situation, such as the relative importance of the two associations in the entire ecosystem). This configuration method comprehensively considers the difference between the actual situation and the model prediction results in terms of positive association (with extended association network entities) and negative association (with non-extended association network entities) of forest coverage, thereby providing an accurate target for the optimization of the neural network model.

[0195] Finally, the neural network model is optimized with the goal of minimizing the target training error. In the initial state, the neural network model may not accurately predict the association vector between forest coverage and air quality index, soil erosion rate and other network entities. For example, the model may overestimate or underestimate the positive impact of forest coverage on air quality index, or make inaccurate judgments on the negative association between forest coverage and soil erosion rate in special cases.

[0196] To optimize the model, adjust the parameters of the neural network model, such as weights and biases. If the model's prediction of the association vector between forest coverage and air quality index results in a large target training error, the server will adjust the relevant weights and biases based on the direction (overestimation or underestimation) and magnitude of the error. For example, if the positive impact is overestimated, the weights of some features connecting forest coverage and air quality index may be reduced.

[0197] With continuous optimization and adjustment, the target training error will gradually decrease. The training termination condition can be that the target training error is less than a certain set threshold. For example, when the target training error is reduced to a very small value (such as indicating that the model prediction result is very close to the actual correlation situation), the neural network model is considered to have met the training termination condition. In this process, the neural network model continuously learns the true correlation between forest coverage and extended associated network entities (such as air quality index) and non-extended associated network entities (such as soil erosion rate, etc.), so that it can more accurately predict the association vector between forest coverage and other network entities.

[0198] For other first network entities, such as biodiversity, soil fertility, water resources, wetland area, etc., the server can also follow similar steps to obtain the first association estimation index between each of them and the extended associated network entity, the second association estimation index between each of them and the non-extended associated network entity, configure the target training error, and optimize the neural network model with the goal of minimizing the target training error until the training termination condition is met. It is just that there will be differences in the specific association relationship, the determination of the association estimation index, and the parameter adjustment of the neural network model, which depends on their respective characteristics and mutual relationships in the ecological environment.

[0199] For example, in a possible implementation, the method further includes:

[0200] Step S150, obtaining target ecological factors and target environmental monitoring indicators in the sample ecological environment data sequence.

[0201] Step S160, determining the association vector between the target ecological factor and the target environmental monitoring indicator according to the target coding data of the target ecological factor and the target coding data of the target environmental monitoring indicator through the neural network model after completing parameter learning.

[0202] Among them, the target ecological factor is used to represent any ecological factor in the sample ecological environment data sequence, and the target environmental monitoring indicator is used to represent any environmental monitoring indicator in the sample ecological environment data sequence.

[0203] For example, in a possible implementation, the association vector is indicated by an association estimation index, and the method further includes:

[0204] Obtain the correlation estimation indicators respectively corresponding to the target ecological factor and at least one target environmental monitoring indicator.

[0205] The target environmental monitoring indicators are sorted according to the descending order of the correlation estimation indicators respectively corresponding to the target ecological factors and the target environmental monitoring indicators to generate sorted target environmental monitoring indicators.

[0206] The first Q target environmental monitoring indicators are extracted from the sorted target environmental monitoring indicators according to the associated estimation index, and the Q target environmental monitoring indicators are associated with the target ecological factors.

[0207] In the sample ecological environment data sequence, there are multiple ecological factors (such as forest coverage, biodiversity, soil fertility, etc.) and multiple environmental monitoring indicators (such as air quality index, water pH, soil erosion rate, etc.). The server determines "forest coverage" as the target ecological factor, which is selected based on specific analysis needs or research purposes. At the same time, a certain indicator is determined as the target environmental monitoring indicator from among many environmental monitoring indicators, such as "air quality index". This selection can be random or determined based on the focus of the current ecological environment research or a specific task.

[0208] The server has completed the parameter learning of the neural network model through the previous steps. For the target ecological factor "forest coverage", its target coding data is the result of integrating various information on forest coverage in terms of positive and negative associations. For example, the positive impact of forest coverage on air quality (such as absorbing pollutants, releasing oxygen, etc.) and the negative impact on soil erosion under special circumstances (such as increased soil erosion risk caused by unreasonable development) are all integrated into this target coding data.

[0209] For the target environmental monitoring indicator “Air Quality Index”, its target coding data is also obtained through a series of complex construction processes, reflecting the various correlations between the Air Quality Index and other ecological factors and environmental monitoring indicators.

[0210] After completing parameter learning, the neural network model uses these two target coded data to determine the association vector between forest coverage and air quality index. The neural network model analyzes and calculates the target coded data of forest coverage and the target coded data of air quality index based on the various association patterns and relationship weights between ecological factors and environmental monitoring indicators learned previously. For example, the model may identify that there is a specific logical relationship between certain elements in the target coded data of forest coverage and specific elements in the target coded data of air quality index. Through the analysis and calculation of these relationships, the association vector between forest coverage and air quality index is obtained. This association vector represents the direction and degree of the comprehensive impact of forest coverage on air quality index. For example, a strong positive correlation indicates that an increase in forest coverage will significantly improve the air quality index.

[0211] The server obtains the correlation estimation indicators corresponding to the forest coverage rate and multiple target environmental monitoring indicators. In addition to the correlation with the air quality index mentioned above, it may also include correlation estimation indicators with other environmental monitoring indicators such as soil erosion rate, water quality pH, etc.

[0212] The estimated correlation index between forest coverage and soil erosion rate is obtained based on the analysis of a large amount of actual ecological and environmental data. In the actual ecological environment, when the forest coverage changes, the soil erosion rate will also change accordingly. For example, when the forest is cut down and the forest coverage decreases, the soil loses the protection of vegetation and is more susceptible to factors such as rain erosion, which leads to an increase in soil erosion rate. The degree of this correlation is expressed by the estimated correlation index, which reflects the size and direction of the impact of forest coverage on soil erosion rate (in this case, it is a positive impact, that is, a decrease in forest coverage will increase soil erosion rate).

[0213] For the correlation estimation index between forest coverage and water pH, the presence of forests will affect the surrounding ecological environment, which in turn affects the water pH. For example, factors such as vegetation and soil microorganisms in forests will affect the chemical cycle in water bodies, thereby affecting the water pH. The correlation estimation index is a numerical value that quantifies the degree and direction of this influence.

[0214] The server estimates the size of the indicator based on the correlation between the forest coverage rate and various target environmental monitoring indicators (such as air quality index, soil erosion rate, water pH, etc.), and arranges these target environmental monitoring indicators in descending order.

[0215] Assume that the estimated index of association between forest coverage and air quality index shows that forest coverage has a strong impact on air quality index, while the estimated index of association with water quality pH shows that its impact is relatively weak, and the estimated index of association with soil erosion rate is between the two. Then, after sorting in descending order, the first is air quality index, followed by soil erosion rate, and finally water quality pH. This sorting order reflects the order of the degree of influence of forest coverage on different target environmental monitoring indicators.

[0216] The server extracts the first Q target environmental monitoring indicators from the sorted target environmental monitoring indicators based on the associated estimation indicators. Assuming Q=2, then according to the sorted order, the two target environmental monitoring indicators of air quality index and soil erosion rate will be extracted.

[0217] Then, these Q target environmental monitoring indicators are associated with the target ecological factor "forest coverage". This association means that when conducting ecological and environmental analysis or research on forest coverage, these Q target environmental monitoring indicators are the indicators that are most closely related to forest coverage, have the greatest impact on forest coverage, or are most affected by forest coverage. In subsequent ecological and environmental research, protection, and management, we can focus on the relationship between these related indicators so that more targeted measures can be taken. For example, in forest management, we can evaluate the impact of changes in forest coverage by monitoring changes in air quality index and soil erosion rate, or improve air quality index and control soil erosion rate by adjusting forest coverage.

[0218] For other target ecological factors, such as biodiversity, soil fertility, water resources, wetland area, etc., the server can also follow similar steps to obtain the correlation estimation indicators between the target environmental monitoring indicators, and perform indicator sorting, extraction and association operations. The only difference is in the specific correlation relationship, the determination of the correlation estimation indicators, and the selection of the target environmental monitoring indicators, which depends on their respective characteristics and relationships in the ecological environment.

[0219] Figure 2 In an embodiment of the present application, a big data analysis system 100 based on ecological environment monitoring is provided, including a processor 1001 and a memory 1003 and an application code stored in the memory 1003. The processor 1001 executes the above application code to implement the steps of the sensor data processing method applied to smart photovoltaics.

[0220] Figure 2 The big data analysis system 100 based on ecological environment monitoring shown includes: a processor 1001 and a memory 1003. Among them, the processor 1001 and the memory 1003 are connected, such as through a bus 1002. Optionally, the big data analysis system 100 based on ecological environment monitoring may also include a transceiver 1004, which can be used for data interaction between the server and other servers, such as data sending and / or data receiving. It should be noted that the transceiver 1004 is not limited to one in actual scheduling, and the structure of the big data analysis system 100 based on ecological environment monitoring does not constitute a limitation on the embodiments of the present application.

[0221] Processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 1001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0222] The bus 1002 may include a path to transmit information between the above components. The bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 1002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0223] The memory 1003 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), magnetic disk storage medium, other magnetic storage devices, or any other medium that can be used to carry or store application code and can be read by a computer, without limitation herein.

[0224] The memory 1003 is used to store application code for executing the embodiment of the present application, and the execution is controlled by the processor 1001. The processor 1001 is used to execute the application code stored in the memory 1003 to implement the steps shown in the above method embodiment.

[0225] The server includes but is not limited to: mobile terminals such as mobile phones, notebook computers, PADs, etc. and fixed terminals such as digital TVs, desktop computers, etc.

[0226] An embodiment of the present application provides a computer-readable storage medium on which application code is stored. When the application code is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0227] It should be understood that, although each operation step is indicated by arrows in the flowchart of the embodiment of the present application, the implementation order of these steps is not limited to the order indicated by the arrows. Unless there is clear explanation in this article, in some implementation scenarios of the embodiment of the present application, the implementation steps in each flowchart can be performed in other orders according to demand. In addition, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages according to actual implementation scenarios, and some or all of these sub-steps or stages may be executed at the same time, and each sub-step or stage in these sub-steps or stages may also be executed at different times respectively. In different scenarios at the execution time, the execution order of these sub-steps or stages can be flexibly configured based on demand, and the embodiment of the present application does not limit this.

[0228] The above is only an optional implementation method for some implementation scenarios of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the scheme of the present application, other similar implementation methods based on the technical ideas of the present application are also within the protection scope of the embodiments of the present application.

Claims

1. A big data analysis method based on ecological environment monitoring, characterized in that: The method includes: Obtaining a knowledge network topology corresponding to a sample ecological environment data sequence from ecological environment monitoring big data, wherein the sample ecological environment data sequence includes X ecological factors and Y environmental monitoring indicators; the knowledge network topology includes a plurality of network entities and links between network entities, wherein the network entities represent the ecological factors or the environmental monitoring indicators; the links represent association vectors between the ecological factors and the environmental monitoring indicators, wherein the association vectors include positive association vectors or negative association vectors; Using a neural network model to perform graph coding representation on the knowledge network topology, generating a positive graph coding vector and a negative graph coding vector for each network entity, wherein the positive graph coding vector is obtained by performing graph coding representation based on the positive association vector in the knowledge network topology; and the negative graph coding vector is obtained by performing graph coding representation based on the negative association vector in the knowledge network topology; Determine the target coding data corresponding to each network entity based on the positive graph coding vector and the negative graph coding vector of each network entity through the neural network model, and estimate the association vector between the ecological factors and the environmental monitoring indicators in the sample ecological environment data sequence based on the target coding data corresponding to each network entity through the neural network model; The neural network model is subjected to parameter learning based on the correlation vector between the ecological factor and the environmental monitoring indicator. The neural network model after parameter learning is used to estimate the correlation vector between the target ecological factor and the target environmental monitoring indicator in the sample ecological environment data sequence, wherein the target ecological factor is used to represent any ecological factor in the sample ecological environment data sequence, and the target environmental monitoring indicator is used to represent any environmental monitoring indicator in the sample ecological environment data sequence.

2. The big data analysis method based on ecological environment monitoring according to claim 1 is characterized in that: Any network entity in the knowledge network topology is represented as network entity i; the use of a neural network model to represent the knowledge network topology by graph coding to generate a positive graph coding vector and a negative graph coding vector for each network entity includes: Get the set sub-network expansion depth Z, where Z is a positive integer; Taking network entity i as a basic network entity, according to the characteristic pointing information represented by the link representing the positive association vector in the knowledge network topology, using a neural network model to parse out the Z-level first extended association network entity of the network entity i from the knowledge network topology; Generate an active sub-network topology corresponding to the network entity i based on the network entity i and each of the parsed first extended associated network entities; Configure an active graph coding vector of the network entity i based on the active sub-network topology corresponding to the network entity i; and, Get the preset sub-network expansion depth Z, where Z is a positive integer; Taking the network entity i as the basic network entity, according to the characteristic pointing information represented by the link representing the negative association vector in the knowledge network topology, using a neural network model to parse out the Z-level second extended associated network entity of the network entity i from the knowledge network topology, and generating the negative sub-network topology corresponding to the network entity i based on the network entity i and each of the parsed second extended associated network entities; A negative graph coding vector of the network entity i is configured based on the negative sub-network topology corresponding to the network entity i.

3. The big data analysis method based on ecological environment monitoring according to claim 2 is characterized in that: The configuring the active graph coding vector of the network entity i based on the active sub-network topology corresponding to the network entity i includes: Based on the active association vectors between the network entities in the active sub-network topology corresponding to the network entity i, configuring the active association array corresponding to the network entity i; Based on the active association array corresponding to the network entity i, the active connection array and the active unit array corresponding to the network entity i are configured; based on the active unit array, the active connection array is standardized and converted to generate an active connection array after the standardized conversion; A basic association array is obtained, and an active graph coding vector of the network entity i is determined according to the basic association array and the normalized active connectivity array.

4. The big data analysis method based on ecological environment monitoring according to claim 2 or 3 is characterized in that: The configuring the passive graph coding vector of the network entity i based on the passive sub-network topology corresponding to the network entity i includes: Based on the negative association vectors between the network entities in the negative sub-network topology corresponding to the network entity i, configuring the negative association array corresponding to the network entity i; Obtaining an active association array corresponding to the network entity i; A negative graph encoding vector of the network entity i is generated based on the positive association array and the negative association array corresponding to the network entity i.

5. The big data analysis method based on ecological environment monitoring according to claim 3 is characterized in that: The expansion depth of the active subnetwork topology corresponding to the network entity i is Z, where Z is a positive integer; the active graph encoding vector of the network entity i is determined based on the basic association array and the normalized active connectivity array, including: Take j=0, determine the j+1th level graph encoding vector of the network entity j based on the basic association array and the normalized positive connectivity array; j is an integer less than Z; Take j=j+1. If j=j+1 and is less than Z, optimize the j+1th level graph coding vector according to the actively connected array after normalization transformation; if j=j+1 and is equal to Z, determine the active graph coding vector of the network entity i based on the optimized Zth level graph coding vector.

6. The big data analysis method based on ecological environment monitoring according to claim 4 is characterized in that: The generating the negative graph coding vector of the network entity i based on the positive association array and the negative association array corresponding to the network entity i comprises: Acquire a normalized converted positively connected array corresponding to the network entity i; the normalized converted positively connected array is determined based on the positively associated array; Based on the negative association array corresponding to the network entity i, configure the negative connectivity array and the negative unit array corresponding to the network entity i; Performing a normalization transformation on the negative connected array based on the negative unit array to generate a normalized negative connected array; A basic association array is obtained, and a negative graph coding vector of the network entity i is determined according to the basic association array, the positive connection array after the normalization conversion, and the negative connection array after the normalization conversion.

7. The big data analysis method based on ecological environment monitoring according to claim 6 is characterized in that: The extended depth of the passive sub-network topology corresponding to the network entity i is Z, where Z is a positive integer; the passive graph coding vector of the network entity i is determined according to the basic association array, the positive connection array after the normalization conversion, and the negative connection array after the normalization conversion, including: Take j=0, determine the j+1th level graph encoding vector of the network entity j based on the basic association array and the normalized positive connectivity array; j is an integer less than Z; Take j=j+1. If j=j+1 and is less than Z-1, optimize the j+1th level graph coding vector according to the positive connected array after the normalization conversion; if j=j+1 and is equal to Z-1, optimize the Z-1th level graph coding vector based on the negative connected array after the normalization conversion to generate the negative graph coding vector of the network entity i.

8. The big data analysis method based on ecological environment monitoring according to claim 1 is characterized in that: Any network entity in the knowledge network topology is represented as network entity i; determining the target coding data corresponding to each network entity based on the positive graph coding vector and the negative graph coding vector of each network entity through the neural network model includes: Determine the attention coefficient of the network entity i according to the positive graph coding vector and the negative graph coding vector of the network entity i through the neural network model; Using the attention coefficient of the network entity i as a weighting factor of the negative graph coding vector of the network entity i, and adding the product calculation result between the weighting factor and the negative graph coding vector of the network entity i to the positive graph coding vector to generate target coding data of the network entity i; or, The attention coefficient of the network entity i is used as a weighting factor of the active graph coding vector of the network entity i, and the product calculation result between the weighting factor and the active graph coding vector of the network entity i is added to the passive graph coding vector to generate the target coding data of the network entity i.

9. The big data analysis method based on ecological environment monitoring according to claim 1 is characterized in that: The knowledge network topology includes a first network entity representing ecological factors and a second network entity representing environmental monitoring indicators, wherein the second network entity includes an extended associated network entity and a non-extended associated network entity of the first network entity; wherein the extended associated network entity includes a network entity having a positive association vector with the first network entity, and the non-extended associated network entity includes a network entity having a negative association vector with the first network entity or having no association vector; the association vector is characterized by an association estimation indicator; The performing parameter learning on the neural network model based on the association vector between the ecological factors and the environmental monitoring indicators includes: Acquire a first association estimation index between the first network entity and the extended association network entity, and a second association estimation index between the first network entity and the non-extended association network entity; configuring a target training error based on the first association estimation indicator and the second association estimation indicator; Taking minimizing the target training error as the training goal, the neural network model is optimized until the neural network model meets the training termination condition.

10. A big data analysis system based on ecological environment monitoring, characterized in that: The big data analysis system based on ecological environment monitoring includes a processor and a machine-readable storage medium, in which machine-executable instructions are stored. The machine-executable instructions are loaded and executed by the processor to implement the big data analysis method based on ecological environment monitoring of any one of claims 1-9.

Images