Optimization method and system for water source conservation function evaluation based on self-supervised reconstruction

Abstract

The application discloses an optimization method and system for water source conservation function evaluation based on self-supervised reconstruction, and relates to the technical field of ecological environment data processing. The method comprises the following steps: acquiring ecological index data of a target evaluation area, and dividing the ecological index data into multiple ecological factor groups according to ecological function types; tokenizing and encoding each ecological factor group to obtain a group token sequence, masking part of the tokens according to a preset factor group level self-supervised masking strategy, inputting the obtained masked token sequence into a token reconstruction network to complete self-supervised pre-training, removing a reconstruction output layer to obtain an initial representation model; based on evaluation unit data labeled with water source conservation function, fine-tuning the initial representation model and synchronously training an initial evaluation model to obtain a target representation model and a target evaluation model; inputting ecological factor group token sequences of a region to be evaluated, and outputting water source conservation function evaluation results; and the method is suitable for different regional ecological conditions, and improves the utilization efficiency of multi-source ecological data and the stability of evaluation results.

Description

Technical Field

[0001] This application relates to the field of ecological and environmental data processing technology, and in particular to an optimization method and system for evaluating water conservation function based on self-supervised reconstruction. Background Technology

[0002] Water conservation function is the comprehensive capacity of an ecosystem to retain, regulate, and replenish regional water through processes such as vegetation interception, soil storage, topographic adjustment, precipitation infiltration, and runoff regulation. It is a crucial evaluation component in ecological environmental protection, water resource management, and the construction of ecological security patterns. With the development of remote sensing monitoring, meteorological observation, geographic information systems, and ecological environment databases, the types of data available for water conservation function evaluation are constantly expanding. Current evaluation processes typically unify data from different sources, resolutions, and time scales to the same evaluation unit scale. Compared to single-indicator evaluation methods, evaluation methods based on multi-source ecological environment data can more comprehensively reflect the influencing factors and spatial differences of water conservation function.

[0003] However, existing methods for evaluating water conservation functions have several drawbacks. Firstly, some methods rely primarily on hydrological models, ecological models, or weighted indicator rules, often depending on expert experience, parameter settings, and indicator weights, making it difficult to adequately adapt to the differences in ecological and environmental conditions across different regions. Secondly, while some data-driven methods can extract potential patterns from ecological and environmental data through model training, they typically process various ecological indicators directly into ordinary feature vectors without fully considering the differences in the roles, semantic structures, and inter-group relationships of different ecological indicators in the formation of water conservation functions. This results in insufficient utilization of the internal structure of multi-source ecological and environmental data by the evaluation models, affecting the stability and reliability of the evaluation results. Summary of the Invention

[0004] Based on this, this application provides an optimization method and system for evaluating water conservation function based on self-supervised reconstruction, aiming to solve the technical problems in related technologies such as insufficient structured expression of multi-source ecological environment data, insufficient utilization of semantic association between different ecological indicator groups, strong dependence of evaluation models on manually labeled samples, and insufficient stability and applicability of evaluation results.

[0005] In a first aspect, embodiments of this application provide an optimization method for evaluating water conservation function based on self-supervised reconstruction, comprising: Ecological index data of each evaluation unit in the target evaluation area are obtained and divided into multiple ecological factor groups according to the type of ecological role in the water conservation function evaluation. The ecological factor group is tokenized to obtain the group token sequence of each evaluation unit. Some tokens are masked according to the preset factor group-level self-supervised masking strategy to obtain the masked token sequence. The masking token sequence is input into the token reconstruction network, and the pre-training loss is output. Self-supervised pre-training is performed based on the pre-training loss. After pre-training is completed, the reconstruction output layer of the network is removed or not used, and the network body used to generate context representation is retained as the initial representation model of the ecological factor group. Based on the evaluation unit data labeled with water conservation function, the initial representation model is fine-tuned, and the initial evaluation model with preset water conservation function is trained simultaneously to obtain the target representation model and the target evaluation model respectively. The ecological factor group token sequence of the area to be evaluated is input into the target characterization model, and the functional evaluation characterization of each evaluation unit is output; the functional evaluation characterization is then input into the target evaluation model, and the water conservation function evaluation result is output.

[0006] In some embodiments, the step of tokenizing the ecological factor group to obtain a group token sequence for each evaluation unit includes: For each ecological factor group, perform intra-group feature mapping on the ecological indicators to obtain the corresponding intra-group embedding representation; Set factor group type codes for different ecological factor groups, and fuse the in-group embedding representation with the factor group type codes to obtain ecological factor group tokens; Multiple ecological factor group tokens are combined in a preset order to obtain the group token sequence of the ecological factor group corresponding to each evaluation unit.

[0007] In some embodiments, the intra-group feature mapping consists of a linear mapping layer, a multilayer perceptron, a normalization layer, an activation function layer, an attention mapping layer, or a combination thereof, used to map ecological factor groups with different dimensions, different scales, or different numbers of indicators to a token representation space of a unified dimension. In the group token sequence, the first The first evaluation unit The expression for the token of each ecological factor group is:

[0008] In the formula, For the first The first evaluation unit Each ecological factor group token. For the first The intragroup embedding mapping function corresponding to each ecological factor group Indicates the first The first evaluation unit One group of ecological factors For the first The type or location code corresponding to each ecological factor group.

[0009] In some embodiments, the step of masking a portion of the tokens according to a preset factor group-level self-supervised masking strategy to obtain a masked token sequence includes: The factor group-level self-supervised masking strategy includes at least one of the following: random masking, masking by factor group type, masking by missing indicators, masking by ecological importance, masking by differences in evaluation area, or masking by a preset masking ratio. The masked ecological factor group tokens satisfy:

[0010] In the formula, For the token of the ecological factor group after masking treatment. For the first The first evaluation unit Each ecological factor group token. To mask the indicator variable, A learnable masking token.

[0011] In some embodiments, in the step of inputting the masked token sequence into a token reconstruction network, outputting a pre-training loss, and performing self-supervised pre-training based on the pre-training loss, the token reconstruction network includes a context representation extraction layer, a factorial semantic association learning layer, and a reconstruction output layer; wherein... The context representation extraction layer is used to extract context features from the masked token sequence; the factor group semantic association learning layer is used to learn the semantic association between tokens of different ecological factor groups; and the reconstruction output layer is used to generate a reconstruction representation of the masked ecological factor group based on the context representation.

[0012] In some embodiments, the step of inputting the masked token sequence into a token reconstruction network, outputting a pre-training loss, and performing self-supervised pre-training based on the pre-training loss includes the following data processing procedures for the token reconstruction network: Based on the unmasked ecological factor group tokens, the semantic association between different ecological factor groups is learned, and the masked ecological factor group tokens are reconstructed to obtain the reconstructed ecological factor group representation. The association relationship of the original ecological factor group is calculated based on the token sequence of the complete ecological factor group, and the association relationship of the reconstructed ecological factor group is calculated based on the reconstructed ecological factor group representation. The self-supervised pre-training loss is determined based on the differences between the reconstructed ecological factor group representation and the corresponding original ecological factor group representation, as well as the differences between the reconstructed ecological factor group association relationship and the original ecological factor group association relationship.

[0013] In some embodiments, the pre-training loss includes ecological factor group reconstruction loss and factor group association reconstruction loss; The ecological factor set reconstruction loss is used to constrain the consistency between the reconstruction result of the masked ecological factor set and the original ecological factor set representation, and its expression is:

[0014] In the formula, For ecological factor set reconstruction loss, For the first The set of obscured ecological factors in each evaluation unit For the first In the evaluation unit, the first Reconstruction tokens for each ecological factor group For the first The first evaluation unit Each ecological factor group token. The total number of tokens for the obscured ecological factor group; The factor group association reconstruction loss is used to constrain the consistency between the reconstructed ecological factor group associations and the original ecological factor group associations, and its expression is:

[0015] In the formula, For factor group association reconstruction loss, To evaluate the number of units, For the first The reconstructed ecological factor group correlation relationship corresponding to each evaluation unit For the first The correlation of the original ecological factor groups corresponding to each evaluation unit. It is the Frobenius norm. This is a identifier for the Frobenius norm; The expression for the pre-training loss is:

[0016] In the formula, For self-supervised pre-training loss, For ecological factor set reconstruction loss, For factor group association reconstruction loss, This is the weighting adjustment coefficient.

[0017] Compared with the prior art, the technical solution provided in the first aspect of this application includes at least the following beneficial effects or advantages: The optimization method provided in this application divides multi-source ecological environment data into multiple ecological factor groups according to the ecological role types in water conservation function evaluation, and tokenizes the ecological factor groups so that the ecological index data of the evaluation unit can be expressed in the form of a structured token sequence. Through factor group-level self-supervised masking and token reconstruction network, the model can learn the semantic association and synergistic effect rules between different ecological factor groups under limited labeling conditions. By combining ecological factor group reconstruction loss and factor group association reconstruction loss for self-supervised pre-training, the learning ability of the ecological factor group representation model to the internal structure of multi-source ecological environment data is improved. Furthermore, by fine-tuning the pre-trained ecological factor group representation model and simultaneously training the water conservation function evaluation model during the evaluation task stage, the model's dependence on a large number of manually labeled samples is reduced, improving the stability, reliability, and applicability of the water conservation function evaluation results. Accurate water conservation function evaluation can still be achieved under limited labeled data scenarios.

[0018] Secondly, embodiments of this application provide an optimization system for evaluating water conservation function based on self-supervised reconstruction, comprising: The factor grouping module is designed to acquire ecological indicator data of each evaluation unit in the target evaluation area and divide them into multiple ecological factor groups according to the type of ecological role in the water conservation function evaluation. The token encoding module is configured to encode the ecological factor group into a token to obtain a group token sequence for each evaluation unit, and to mask some tokens according to a preset factor group-level self-supervised masking strategy to obtain a masked token sequence. The model building module is configured to input the masking token sequence into the token reconstruction network, output a pre-training loss, perform self-supervised pre-training based on the pre-training loss, and after pre-training is completed, remove or not use the reconstruction output layer of the network, retain the network body used to generate context representations as the initial representation model of the ecological factor group. The model training module is constructed based on evaluation unit data labeled with water conservation function, to fine-tune the initial representation model, and simultaneously train the initial evaluation model with preset water conservation function, so as to obtain the target representation model and the target evaluation model respectively. The evaluation output module is configured to input the ecological factor group token sequence of the area to be evaluated into the target representation model, output the functional evaluation representation of each evaluation unit, and input the functional evaluation representation into the target evaluation model to output the water conservation function evaluation result.

[0019] Thirdly, this application also provides an electronic device, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the steps of the optimization method for evaluating water conservation function based on self-supervised reconstruction provided in the first aspect above.

[0020] Fourthly, this application also provides a computer-readable storage medium storing computer program instructions thereon, which, when executed by a processor, implement the steps of the optimization method for evaluating water conservation function based on self-supervised reconstruction provided in the first aspect.

[0021] It is understood that the beneficial effects of the technical solutions provided in the second, third and fourth aspects above can be found in the relevant descriptions in the first aspect above, and will not be repeated here.

[0022] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating the optimization method for evaluating water conservation function based on self-supervised reconstruction provided in this application embodiment; Figure 2 A schematic diagram illustrating the process of generating ecological factor group tokens as provided in this application embodiment; Figure 3 A schematic diagram illustrating the process of self-supervised reconstruction of ecological factor group tokens and evaluation of water conservation function provided in the embodiments of this application; Figure 4 A structural block diagram of an optimization system for evaluating water conservation function based on self-supervised reconstruction, provided in an embodiment of this application. Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate several embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0026] It should be noted that, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0027] With the development of remote sensing monitoring, meteorological observation, geographic information systems, and ecological environment databases, the types of data available for water conservation function assessment are constantly being enriched. Current assessment processes typically require the integration of multi-source ecological environment data, including remote sensing imagery, meteorological data, topographic data, soil data, hydrological data, land use data, human activity data, and ecological survey data. Data from different sources, with different resolutions and time scales, are then processed at the same assessment unit scale.

[0028] In the evaluation of water conservation function, different ecological indicators typically correspond to different ecological processes. For example, vegetation status indicators can reflect the interception, cover protection, and transpiration regulation effects of vegetation; meteorological moisture indicators can reflect regional water supply and consumption conditions; soil storage indicators can reflect soil infiltration, water storage, and retention capacity; topographic indicators can reflect slope runoff, runoff, and water redistribution characteristics; and land use and human activity indicators can reflect the impact of surface cover changes and human disturbances on water conservation capacity. These different types of ecological indicators have both complementary and synergistic relationships. If all indicators are input into the model as undifferentiated features, it is easy to weaken the structured expressive power of the ecological indicator groups and reduce the model's representation effect on the formation mechanism of water conservation function.

[0029] Furthermore, in practical applications, obtaining water conservation function labels or high-quality historical evaluation results is often costly, and data completeness and labeling quality vary across different regions and years. Relying solely on labeled samples to train the evaluation model is susceptible to problems such as insufficient labeled samples, uneven sample distribution, and inadequate regional adaptability. Therefore, how to fully utilize the inherent correlations within multi-source ecological and environmental data under limited labeling conditions and construct data processing methods that reflect the semantic structure between ecological indicator groups is a key issue in improving the stability and applicability of intelligent evaluation of water conservation functions.

[0030] Based on this, the inventors propose an optimization method and system for evaluating water conservation function based on self-supervised reconstruction. This method categorizes multi-source ecological indicators into ecological factor groups with ecological semantics according to their water conservation function types. Through tokenization encoding of ecological factor groups, self-supervised masking at the factor group level, and reconstruction learning of ecological factor group representations and their correlations, it extracts ecological factor group representations suitable for water conservation function evaluation. This improves the model's ability to learn relationships between multi-source ecological environment indicator groups, enhancing the reliability and applicability of the water conservation function evaluation results. The following will further describe this technical solution with several embodiments: Please see Figures 1 to 3 , Figure 1 This is a flowchart of the optimization method for evaluating water conservation function based on self-supervised reconstruction provided in this embodiment. Figure 2 This is a schematic diagram of the ecological factor group token generation process provided in this embodiment; Figure 3 This is a schematic diagram of the process of self-supervised reconstruction of ecological factor group tokens and evaluation of water conservation function provided in this embodiment; this embodiment provides an optimization method for water conservation function evaluation based on self-supervised reconstruction, specifically including steps S10 to S50.

[0031] Step S10: Obtain the ecological index data of each evaluation unit in the target evaluation area, and divide it into multiple ecological factor groups according to the type of ecological role in the water conservation function evaluation. In this step, to obtain ecological indicator data, it is first necessary to collect multi-source ecological and environmental data of the target evaluation area. Multi-source ecological and environmental data includes remote sensing image data, meteorological data, topographic data, soil data, hydrological data, land use data, human activity data, ecological survey data, historical evaluation data, and other geospatial data that can be used to characterize the influencing factors of water conservation function.

[0032] Furthermore, spatial registration, scale unification, missing value handling, and standardization are performed on multi-source ecological and environmental data to unify data from different sources, with different spatial resolutions, and different dimensions to the scale of the preset evaluation units. These evaluation units can be raster evaluation units, sub-basin evaluation units, administrative evaluation units, or other preset spatial evaluation units. Ecological indicator values ​​are extracted from each evaluation unit to obtain the ecological indicator data for the target evaluation area. It should be understood that spatial registration, scale unification, missing value handling, and standardization are existing data preprocessing techniques and will not be elaborated upon here.

[0033] In some embodiments, ecological indicator data can be represented as an ecological indicator matrix:

[0034] in, This represents the ecological indicator matrix of the target evaluation area. Indicates the first Ecological indicator vectors for each evaluation unit.

[0035] No. The ecological indicator vector of each evaluation unit can be represented as:

[0036] in, Indicates the first The first evaluation unit One ecological indicator.

[0037] Optional ecological indicators may include multiple items from the following: vegetation cover, vegetation index, leaf area index, net primary productivity of vegetation, precipitation, evapotranspiration, temperature, humidity, solar radiation, soil moisture, plant available water content, root limiting layer depth, soil type, soil texture, soil organic matter, altitude, slope, aspect, topographic relief, surface runoff related indicators, land use type, land cover type, population density, construction intensity, and road density. The above ecological indicators are only examples, and specific indicators may be adjusted according to the data availability of the target evaluation area, the water conservation function evaluation needs, and the results of ecological factor group division.

[0038] In some embodiments, such as Figure 2 As shown, the division of multiple ecological factor groups can be based on the differences in the roles of ecological indicator data in the formation of water conservation functions. Ecological factor groups can include at least two of the following: vegetation factor group, meteorological and water factor group, soil water retention factor group, topographic regulation factor group, human activity factor group, hydrological response factor group, land use factor group, ecological protection factor group, land cover factor group, and at least two other factor groups used to characterize the influencing mechanisms of water conservation functions. The specific design can be selectively chosen based on different application scenarios.

[0039] Optional, the first The ecological indicator vector of each evaluation unit can be divided into multiple ecological factor groups:

[0040] in, Indicates the first The first evaluation unit One group of ecological factors.

[0041] For example, vegetation factor groups can be used to characterize ecological functions such as vegetation cover, vegetation growth, and vegetation interception; meteorological water factor groups can be used to characterize ecological functions such as precipitation supply, evapotranspiration, and heat conditions; soil storage factor groups can be used to characterize soil infiltration, soil water storage, and water retention capacity; topographic regulation factor groups can be used to characterize slope runoff, topographic runoff, and water redistribution; and human activity factor groups can be used to characterize the impact of land use change, construction disturbance, and human activities on water conservation functions. The names, quantities, and included indicators of the above ecological factor groups do not constitute a limitation on the scope of protection of this invention. In practical applications, they can be set according to the ecological environment characteristics, data types, and evaluation tasks of the target evaluation area.

[0042] Step S20: Tokenize the ecological factor group to obtain the group token sequence of each evaluation unit, and mask some tokens according to the preset factor group-level self-supervised masking strategy to obtain the masked token sequence. In some embodiments, the ecological indicators within each ecological factor group are mapped within the group to obtain the corresponding intra-group embedding representation; factor group type codes are set for different ecological factor groups, and the intra-group embedding representations are fused with the factor group type codes to obtain ecological factor group tokens; multiple ecological factor group tokens are combined in a preset order to obtain the group token sequence of the ecological factor group corresponding to each evaluation unit.

[0043] Specifically, in the group of tokens, the first The first evaluation unit The expression for the token of each ecological factor group is:

[0044] In the formula, Indicates the first The first evaluation unit Each ecological factor group token. Indicates the first The intragroup embedding mapping function corresponding to each ecological factor group Indicates the first The first evaluation unit One group of ecological factors Indicates the first The type or location code corresponding to each ecological factor group.

[0045] Optionally, the intra-group embedding mapping function can consist of a linear mapping layer, a multilayer perceptron, a normalization layer, an activation function layer, an attention mapping layer, or a combination thereof. Through intra-group embedding mapping, ecological factor groups with different dimensions, scales, and numbers of indicators can be mapped to a feature space of a unified dimension, thus facilitating subsequent learning of the contextual relationships between ecological factor groups by the neural network.

[0046] For example, the intra-group embedding mapping function can adopt a four-level cascaded architecture consisting of factor-wise normalization, dimension-aligned linear projection, intra-group attention weighted aggregation, and top-level feature calibration. First, layer normalization is performed independently on the original feature vectors of each heterogeneous ecological indicator within the ecological factor group to eliminate the differences in dimensionality and numerical scale between different indicators. Then, through a parameter-shared fully connected linear mapping layer combined with the ReLU activation function, the single-factor features with different dimensions are uniformly projected to a preset intermediate feature dimension. Subsequently, the importance weights of each ecological indicator are adaptively calculated through a single-head self-attention mechanism, and the weighted sum of all single-factor features within the group is used to obtain the initial group embedding vector. Finally, the initial group embedding vector is input into a multilayer perceptron consisting of two fully connected layers, a GELU activation function, and a layer normalization layer for nonlinear feature calibration, outputting a standard group embedding vector with a fixed dimension. This allows ecological factor groups with different numbers of indicators and different original dimensions to be mapped to a unified feature space, providing standardized input for subsequent neural network learning of the contextual relationships between ecological factor groups.

[0047] In some embodiments, the first The complete ecological factor group token sequence corresponding to each evaluation unit is represented as follows:

[0048] In the formula, Indicates the first Each evaluation unit corresponds to a complete set of ecological factor tokens. These represent ecological factor group tokens corresponding to different ecological factor groups. The complete ecological factor group token sequence is an unmasked ecological factor group token sequence, which can be used for subsequent model pre-training and water conservation function evaluation.

[0049] In some embodiments, such as Figure 3 As shown, during the token preparation and masking stage, a portion of the ecological factor group tokens in the complete ecological factor group's token sequence are masked to obtain a masked token sequence. Simultaneously, the complete ecological factor group token sequence before masking is retained for subsequent acquisition of the original ecological factor group representation and its association relationships.

[0050] Among them, the masked ecological factor group tokens satisfy:

[0051] In the formula, This indicates the token representing the ecological factor group after masking treatment. Indicates the occlusion indicator variable; when When, it indicates the first The tokens for each ecological factor group were not obscured; when When, it indicates the first The tokens for each ecological factor group were obscured; Indicates the first The first evaluation unit Each ecological factor group token; This indicates a learnable masking token.

[0052] Accordingly, the first The masking token sequence corresponding to each evaluation unit can be represented as:

[0053] in, Indicates the first The masking token sequence corresponding to each evaluation unit.

[0054] Optionally, the factor group-level self-supervised masking strategy includes at least one of the following: random masking, masking by factor group type, masking by missing indicators, masking by ecological importance, masking by differences in evaluation areas, and masking by a preset masking ratio.

[0055] In this embodiment, the masking process uses ecological factor group tokens as the basic unit, rather than individual ecological indicators or single values. Through factor group-level masking, the token reconstruction network can infer the representation of masked ecological factor groups based on the information of unmasked ecological factor groups, thereby learning the semantic relationships and synergistic patterns between different ecological factor groups.

[0056] Step S30: Input the masking token sequence into the token reconstruction network, output the pre-training loss, perform self-supervised pre-training based on the pre-training loss, and after the pre-training is completed, remove or do not use the reconstruction output layer of the network, retain the network main body used to generate context representations, as the initial representation model of the ecological factor group.

[0057] In some embodiments, such as Figure 3 As shown, the token reconstruction network includes a context representation extraction layer, a factor group semantic association learning layer, and a reconstruction output layer. The context representation extraction layer is used to extract context features from the masked token sequence; the factor group semantic association learning layer is used to learn the semantic association relationships between tokens of different ecological factor groups; and the reconstruction output layer is used to generate the reconstructed representation of the masked ecological factor groups based on the context representation.

[0058] Furthermore, the token reconstruction network can employ Transformer encoding structures, attention network structures, multilayer perceptron structures, encoder-decoder structures, or other neural network structures capable of processing token sequences and performing reconstruction learning. For example, a Transformer encoding structure can be used as the backbone architecture. Its input is an ordered sequence of tokens embedded in multiple ecological factor groups obtained through intra-group embedding mapping. The network mines global dependencies and semantic relationships between different ecological factor groups through stacked multi-head self-attention layers, and completes nonlinear transformation and deep aggregation of features with layer normalization layers and feedforward multilayer perceptron layers. In lightweight deployment scenarios with limited computing power, this token reconstruction network can be replaced with a pure attention network structure or a shallow multilayer perceptron structure to reduce computational overhead. In complex ecosystem analysis scenarios with higher requirements for reconstruction accuracy, an encoder-decoder structure can be used to achieve more refined token sequence reconstruction learning through encoding compression and decoding reconstruction, ultimately outputting an optimized token sequence that integrates global contextual information, providing a high-order feature foundation for subsequent ecological feature analysis and state assessment.

[0059] In some embodiments, the data processing procedure of the token reconstruction network includes: learning the semantic association between different ecological factor groups based on unmasked ecological factor group tokens, and reconstructing the masked ecological factor group tokens to obtain the reconstructed ecological factor group representation; calculating the original ecological factor group association based on the complete ecological factor group token sequence, and calculating the reconstructed ecological factor group association based on the reconstructed ecological factor group representation; and determining the self-supervised pre-training loss based on the difference between the reconstructed ecological factor group representation and the corresponding original ecological factor group representation, as well as the difference between the reconstructed ecological factor group association and the original ecological factor group association.

[0060] Specifically, reconstructing the masked ecological factor tokens includes: when the context representation is obtained:

[0061] In the formula, Indicates the first The context representation corresponding to each evaluation unit This indicates that the token restructures the network. This represents the network parameters for token-based network reconstruction.

[0062] Based on contextual representation, a reconstructed ecological factor set representation is generated by reconstructing the output layer:

[0063] In the formula, Indicates the first The reconstructed ecological factor group representation sequence corresponding to each evaluation unit This represents the reconstruction mapping function corresponding to the reconstructed output layer. This represents the network parameters of the reconstructed output layer.

[0064] Furthermore, the reconstructed ecological factor set representation sequence can be expressed as:

[0065] In the formula, Indicates the first The first evaluation unit Reconstructed representation of a group of ecological factors.

[0066] It should be noted that, in order to ensure that the original ecological factor group associations and the reconstructed ecological factor group associations have a consistent basis for comparison, this embodiment uses the same or structurally consistent ecological factor group association calculation function to calculate both. Specifically, the original ecological factor group associations are calculated from the complete ecological factor group token sequence, and the reconstructed ecological factor group associations are calculated from the reconstructed ecological factor group representation sequence.

[0067] The correlation between the original ecological factor groups can be represented as follows: , in,

[0068] In the formula, Indicates the first The correlation of the original ecological factor groups corresponding to each evaluation unit. This represents the complete token sequence of the ecological factor group. Functions representing the calculation of correlations among groups of ecological factors. Represents the first element in the original ecological factor group correlation matrix. Line 1 Column association weights Indicates the number of ecological factor groups. and These represent the indices of the ecological factor groups, and ; A unified representation dimension for ecological factor group tokens; , and They represent the first In the evaluation unit, the first , No. and the The original ecological factor group tokens corresponding to each ecological factor group. This represents the index for the summation of ecological factor groups in the Softmax normalization calculation, and , This represents the transpose of a vector.

[0069] The reconstructed relationships among ecological factor groups can be represented as follows: , in,

[0070] In the formula, Indicates the first The reconstructed ecological factor group correlation relationship corresponding to each evaluation unit This represents the sequence of reconstructed ecological factor groups. This refers to the function used to calculate the association relationship of ecological factor groups, which has the same calculation process or structure as the original ecological factor group. Indicating the reconstructed ecological factor group association matrix, the first... Line 1 Column association weights , and They represent the first In the evaluation unit, the first , No. and the Each ecological factor group corresponds to a reconstructed ecological factor group token. A unified representation dimension for ecological factor group tokens. Represents the transpose of a vector. This represents the index for the summation of ecological factor groups in the Softmax normalization calculation, and .

[0071] In this embodiment, the token reconstruction network learns the semantic associations and contextual dependencies between different ecological factor groups by masking token sequences, and generates reconstructed ecological factor group representations. Furthermore, it calculates the original ecological factor group associations and the reconstructed ecological factor group associations based on the complete ecological factor group token sequences and the reconstructed ecological factor group representation sequences, respectively, thereby providing a data foundation for the subsequent calculation of ecological factor group reconstruction loss and factor group association reconstruction loss.

[0072] In some embodiments, the pre-training loss includes an ecological factor group reconstruction loss and a factor group association reconstruction loss; wherein, the ecological factor group reconstruction loss is used to constrain the consistency between the reconstruction result of the masked ecological factor group and the original ecological factor group representation, and the expression for the ecological factor group reconstruction loss is:

[0073] In the formula, For ecological factor set reconstruction loss, For the first The set of obscured ecological factors in each evaluation unit For the first In the evaluation unit, the first Reconstruction tokens for each ecological factor group Indicates the first The first evaluation unit Each ecological factor group token. The total number of tokens for the obscured ecological factor group; The factor set association reconstruction loss is used to constrain the consistency between the reconstructed ecological factor set associations and the original ecological factor set associations. The expression for the factor set association reconstruction loss is:

[0074] In the formula, For factor group association reconstruction loss, To evaluate the number of units, For the first The reconstructed ecological factor group correlation relationship corresponding to each evaluation unit For the first The correlation of the original ecological factor groups corresponding to each evaluation unit. It is the Frobenius norm. This is a identifier for the Frobenius norm; The expression for the pre-training loss is:

[0075] In the formula, For self-supervised pre-training loss, For ecological factor set reconstruction loss, For factor group association reconstruction loss, This represents the weighted adjustment coefficient between the ecological factor group reconstruction loss and the factor group association reconstruction loss, and .

[0076] In some embodiments, after pre-training is completed, the reconstruction output layer in the token reconstruction network is removed or not used, and its network body used to generate context representations is retained as the initial representation model for the ecological factor group, wherein the initial representation model is expressed as:

[0077] in, Indicates the first Contextual representation of ecological factor groups corresponding to each evaluation unit This represents the ecological factor group representation model retained after pre-training. This represents the model parameters retained after pre-training.

[0078] Step S40: Based on the evaluation unit data labeled with water conservation function, the initial representation model is fine-tuned, and the initial evaluation model with preset water conservation function is trained simultaneously to obtain the target representation model and the target evaluation model respectively. In this step, such as Figure 3 As shown, the evaluation unit data can include, but is not limited to, data with water conservation function labels, historical evaluation results, or expert rule classification results. The complete ecological factor group token sequence corresponding to the evaluation unit data is input into the pre-trained ecological factor group initial representation model to obtain the water conservation function evaluation representation.

[0079] Specifically, the contextual representation of ecological factor groups is processed through pooling, aggregation, or feature selection to obtain the water conservation function evaluation representation:

[0080] in, Indicates the first Evaluation and characterization of water conservation function of each evaluation unit This represents the pooling function.

[0081] By inputting the water conservation function evaluation characteristics into the water conservation function evaluation model, the predicted results of water conservation function are obtained:

[0082] in, Indicates the first Prediction results of water conservation function of each evaluation unit This represents the water conservation function evaluation model. These represent the parameters of the water conservation function evaluation model.

[0083] In the classification and evaluation task, the water conservation function evaluation model outputs the first... The predicted probability that each evaluation unit belongs to each water source conservation function level. The loss of the classification evaluation task can be expressed as:

[0084] in, Indicates the loss of the classification and evaluation task. This indicates the number of labeled evaluation units participating in the training. Indicates the first The actual or reference rating label for each rating unit The model predicts the first... Each evaluation unit belongs to its true level. The probability of.

[0085] In the regression evaluation task, the water conservation function evaluation model outputs the first... The water conservation function score or continuous water conservation capacity value of each evaluation unit. The regression evaluation task loss can be expressed as:

[0086] in, This indicates the loss in the regression evaluation task. Indicates the first The predicted water conservation function score or continuous water conservation capacity value of each evaluation unit. This indicates the corresponding reference score or reference continuous value.

[0087] Evaluate task loss The assessment is determined based on the type of water conservation function evaluation task; when the evaluation task is a classification evaluation task... Loss of classification and evaluation task When the evaluation task is a regression evaluation task, Take the regression evaluation task loss .

[0088] Based on the evaluation task loss, the initial representation model of the pre-trained ecological factor group is fine-tuned, and the initial evaluation model of water conservation function is trained simultaneously. In other words, during the evaluation task training process, at least some parameters in the initial representation model of the ecological factor group and the parameters of the water conservation function evaluation model are updated simultaneously, so that the ecological factor group representation model is further adapted to the water conservation function evaluation task.

[0089] Step S50: Input the ecological factor group token sequence of the area to be evaluated into the target characterization model, and output the functional evaluation characterization of each evaluation unit; input the functional evaluation characterization into the target evaluation model, and output the water conservation function evaluation result.

[0090] In this step, it should be noted that for the token sequence of the ecological factor group in the area to be evaluated, the collected multi-source ecological environment data of the area to be evaluated can be processed according to the aforementioned steps S10 to S20 to generate the token sequence of the complete ecological factor group corresponding to the area to be evaluated. The complete ecological factor group token sequence is then input into the target representation model of the trained ecological factor group to obtain the water conservation function evaluation representation of each evaluation unit in the area to be evaluated. Finally, the water conservation function evaluation representation is input into the target evaluation model of the trained water conservation function to output the water conservation function evaluation result of the area to be evaluated.

[0091] Optionally, the evaluation results of water conservation function include at least one of the following: water conservation function level, water conservation function score, continuous water conservation capacity value, spatial distribution results of water conservation function, and evaluation reliability results.

[0092] In some embodiments, if the evaluation task is a classification task, the water conservation function level corresponding to each evaluation unit is output; if the evaluation task is a regression task, the water conservation function score or continuous water conservation capacity value corresponding to each evaluation unit is output; if spatial representation results are required, the evaluation results can be mapped to a spatial distribution map of water conservation function according to the spatial location of each evaluation unit.

[0093] In the method steps provided in the above embodiments, multi-source ecological environment data are divided into multiple ecological factor groups according to the ecological role type in water conservation function evaluation, and the ecological factor groups are tokenized and encoded so that the ecological index data of the evaluation unit can be expressed in the form of a structured token sequence. Through factor group-level self-supervised masking and token reconstruction network, the model can learn the semantic association and synergistic effect rules between different ecological factor groups under limited labeling conditions. By combining ecological factor group reconstruction loss and factor group association reconstruction loss for self-supervised pre-training, the learning ability of the ecological factor group representation model to the internal structure of multi-source ecological environment data is improved. Furthermore, by fine-tuning the pre-trained ecological factor group representation model and simultaneously training the water conservation function evaluation model during the evaluation task stage, the model's dependence on a large number of manually labeled samples is reduced, and the stability, reliability, and applicability of the water conservation function evaluation results are improved. Accurate water conservation function evaluation can still be achieved in scenarios with limited labeled data.

[0094] Please see Figure 4 , Figure 4 This embodiment illustrates an optimization system for evaluating water conservation function based on self-supervised reconstruction. In this embodiment, the optimization system for evaluating water conservation function based on self-supervised reconstruction is used to perform the above-mentioned... Figure 1 The steps in the corresponding embodiments. Please refer to the details. Figure 1 as well as Figure 1 The relevant descriptions in the corresponding embodiments are shown below. For ease of explanation, only the parts relevant to this embodiment are shown. See also... Figure 4 The optimization system 200 for evaluating water conservation function based on self-supervised reconstruction includes: The factor grouping module 210 is configured to acquire ecological index data of each evaluation unit in the target evaluation area, and divide the ecological index data into multiple ecological factor groups according to the ecological role type of the ecological index data in the water conservation function evaluation. The token encoding module 220 is configured to tokenize the ecological factor group to obtain a group token sequence for each evaluation unit, and to mask some tokens according to a preset factor group-level self-supervised masking strategy to obtain a masked token sequence. The model building module 230 is configured to input the masking token sequence into the token reconstruction network, output a pre-training loss, perform self-supervised pre-training based on the pre-training loss, and after pre-training is completed, remove or not use the reconstruction output layer of the network, retain the network body used to generate context representations as the initial representation model of the ecological factor group. The model training module 240 is constructed to fine-tune the initial representation model based on evaluation unit data labeled with water conservation function, and simultaneously train the initial evaluation model with preset water conservation function to obtain the target representation model and the target evaluation model respectively. The evaluation output module 250 is configured to input the ecological factor group token sequence of the area to be evaluated into the target representation model, output the functional evaluation representation of each evaluation unit, and input the functional evaluation representation into the target evaluation model to output the water conservation function evaluation result.

[0095] The optimized system for water conservation function evaluation provided in this embodiment divides multi-source ecological environment data into multiple ecological factor groups according to the ecological role types in water conservation function evaluation, and tokenizes the ecological factor groups so that the ecological index data of the evaluation unit can be expressed in the form of a structured token sequence. Through factor group-level self-supervised masking and token reconstruction network, the model can learn the semantic association and synergistic effect rules between different ecological factor groups under limited labeling conditions. By combining ecological factor group reconstruction loss and factor group association reconstruction loss for self-supervised pre-training, the learning ability of the ecological factor group representation model to the internal structure of multi-source ecological environment data is improved. Furthermore, by fine-tuning the pre-trained ecological factor group representation model and simultaneously training the water conservation function evaluation model during the evaluation task stage, the model's dependence on a large number of manually labeled samples is reduced, improving the stability, reliability, and applicability of the water conservation function evaluation results. It can still achieve accurate water conservation function evaluation in scenarios with limited labeled data.

[0096] It should be understood that the modules in the system provided in this embodiment are used to execute... Figure 1 The steps in the corresponding embodiments, and for Figure 1 The steps in the corresponding embodiments have been explained in detail in the above embodiments. Please refer to them for details. Figure 1 as well as Figure 1 The relevant descriptions in the corresponding embodiments will not be repeated here.

[0097] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5As shown, the electronic device provided in this embodiment includes at least one processor 901 and a memory 902. Optionally, the electronic device further includes a communication component 903. The processor 901, memory 902, and communication component 903 are connected via a bus 904. In the specific implementation process, at least one processor 901 executes computer execution instructions stored in memory 902, causing at least one processor 901 to execute the above-mentioned optimization method for water conservation function evaluation based on self-supervised reconstruction.

[0098] The specific implementation process of processor 901 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0099] In the above embodiments, it should be understood that the processor 901 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0100] The memory 902 can be an internal storage unit of an electronic device, such as a hard drive or memory in a server. The memory 902 can also be an external storage terminal device of an electronic device, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc.

[0101] Furthermore, the memory 902 may include both internal storage units of the electronic device and external storage terminal devices. The memory 902 is used to store computer programs and other programs and data required by the turntable terminal device. The memory 902 can also be used to temporarily store data that has been output or will be output.

[0102] Bus 904 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0103] Those skilled in the art will understand that Figure 5 This is merely an example of an electronic device and does not constitute a limitation on the electronic device. It may include more or fewer components than shown, or combine certain components, or different components. For example, a turntable terminal device may also include input / output terminal devices, network access terminal devices, etc.

[0104] In one embodiment, a computer-readable storage medium is also provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the optimization method for evaluating water conservation function based on self-supervised reconstruction as described in the above embodiments.

[0105] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0106] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in application-specific integrated circuits (ASICs). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0107] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0108] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0109] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be non-volatile or volatile. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0110] The terms "first," "second," "third," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects and not to describe a particular order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, it may include a series of steps or units, or optionally, steps or units not listed, or other steps or units inherent to these processes, methods, products, or devices.

[0111] The accompanying drawings show only the portions relevant to this application, not all of them. Before discussing exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations may be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations may be rearranged. The process may be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process may correspond to a method, function, procedure, subroutine, subprogram, etc.

[0112] The terms “component,” “module,” “system,” “unit,” etc., used in this specification are used to refer to computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a unit can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, a thread of execution, a program, and / or distributed between two or more computers. Furthermore, these units can be executed from various computer-readable media on which various data structures are stored. Units can communicate, for example, via signals having one or more data packets (e.g., data from a second unit interacting with another unit between a local system, a distributed system, and / or a network; for example, the Internet interacting with other systems via signals).

[0113] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example.

[0114] Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The reference to "embodiment" herein means that a specific feature, structure, or characteristic described in connection with an embodiment can be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily indicate the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

Claims

1. An optimization method for evaluating water conservation function based on self-supervised reconstruction, characterized in that, include: Ecological index data of each evaluation unit in the target evaluation area are obtained and divided into multiple ecological factor groups according to the type of ecological role in the water conservation function evaluation. The ecological factor groups include at least two of the following: vegetation factor group, meteorological water factor group, soil water retention factor group, topographic regulation factor group, human activity factor group, hydrological response factor group, land use factor group, ecological protection factor group, and land cover factor group. The ecological factor groups are tokenized to obtain a group token sequence for each evaluation unit. A portion of the tokens are then masked according to a preset factor group-level self-supervised masking strategy to obtain a masked token sequence. The masked ecological factor group tokens satisfy the following conditions: In the formula, For the token of the ecological factor group after masking treatment. To mask the indicator variable, For the first The first evaluation unit Each ecological factor group token. A learnable masking token; The masking token sequence is input into the token reconstruction network, and the output is a pre-training loss. Self-supervised pre-training is performed based on the pre-training loss. After pre-training is completed, the reconstruction output layer of the network is removed or not used, and the network body used to generate context representation is retained as the initial representation model of the ecological factor group. The pre-training loss includes ecological factor group reconstruction loss and factor group association reconstruction loss. The ecological factor set reconstruction loss is used to constrain the consistency between the reconstruction result of the masked ecological factor set and the original ecological factor set representation, and its expression is: In the formula, For ecological factor set reconstruction loss, For the first The set of obscured ecological factors in each evaluation unit For the first In the evaluation unit, the first Reconstruction tokens for each ecological factor group For the first In the evaluation unit, the first Original ecological factor group tokens for each ecological factor group. The total number of tokens for the obscured ecological factor group; The factor group association reconstruction loss is used to constrain the consistency between the reconstructed ecological factor group associations and the original ecological factor group associations, and its expression is: In the formula, For factor group association reconstruction loss, To evaluate the number of units, For the first The reconstructed ecological factor group correlation relationship corresponding to each evaluation unit For the first The correlation of the original ecological factor groups corresponding to each evaluation unit. It is the Frobenius norm. This is a identifier for the Frobenius norm; The expression for the pre-training loss is: In the formula, For self-supervised pre-training loss, For ecological factor set reconstruction loss, For factor group association reconstruction loss, This is the weighting adjustment coefficient; Based on the evaluation unit data labeled with water conservation function, the initial representation model is fine-tuned, and the initial evaluation model with preset water conservation function is trained simultaneously to obtain the target representation model and the target evaluation model respectively. The ecological factor group token sequence of the area to be evaluated is input into the target characterization model, and the functional evaluation characterization of each evaluation unit is output; the functional evaluation characterization is then input into the target evaluation model, and the water conservation function evaluation result is output.

2. The optimization method for evaluating water conservation function based on self-supervised reconstruction according to claim 1, characterized in that, The step of tokenizing the ecological factor group to obtain the group token sequence for each evaluation unit includes: For each ecological factor group, perform intra-group feature mapping on the ecological indicators to obtain the corresponding intra-group embedding representation; Set factor group type codes for different ecological factor groups, and fuse the in-group embedding representation with the factor group type codes to obtain ecological factor group tokens; Multiple ecological factor group tokens are combined in a preset order to obtain the group token sequence of the ecological factor group corresponding to each evaluation unit.

3. The optimization method for evaluating water conservation function based on self-supervised reconstruction according to claim 2, characterized in that, The intra-group feature mapping consists of a linear mapping layer, a multilayer perceptron, a normalization layer, an activation function layer, an attention mapping layer, or a combination thereof, and is used to map ecological factor groups with different dimensions, different scales, or different numbers of indicators to a token representation space with a unified dimension. In the group token sequence, the first The first evaluation unit The expression for the token of each ecological factor group is: In the formula, For the first The first evaluation unit Each ecological factor group token. Indicates the first The first evaluation unit One group of ecological factors For the first The intragroup embedding mapping function corresponding to each ecological factor group For the first The type or location code corresponding to each ecological factor group.

4. The optimization method for evaluating water conservation function based on self-supervised reconstruction according to claim 1, characterized in that, The step of masking a portion of the tokens according to a preset factor group-level self-supervised masking strategy to obtain a masked token sequence includes: The factor group-level self-supervised masking strategy includes at least one of the following: random masking, masking by factor group type, masking by missing indicators, masking by ecological importance, masking by differences in evaluation areas, or masking by a preset masking ratio.

5. The optimization method for evaluating water conservation function based on self-supervised reconstruction according to claim 1, characterized in that, In the step of inputting the masked token sequence into a token reconstruction network, outputting a pre-training loss, and performing self-supervised pre-training based on the pre-training loss, the token reconstruction network includes a context representation extraction layer, a factor group semantic association learning layer, and a reconstruction output layer; wherein... The context representation extraction layer is used to extract context features from the masked token sequence; the factor group semantic association learning layer is used to learn the semantic association between tokens of different ecological factor groups; and the reconstruction output layer is used to generate a reconstruction representation of the masked ecological factor group based on the context representation.

6. The optimization method for evaluating water conservation function based on self-supervised reconstruction according to claim 5, characterized in that, In the step of inputting the masked token sequence into the token reconstruction network, outputting the pre-training loss, and performing self-supervised pre-training based on the pre-training loss, the data processing of the token reconstruction network includes: Based on the unmasked ecological factor group tokens, the semantic association between different ecological factor groups is learned, and the masked ecological factor group tokens are reconstructed to obtain the reconstructed ecological factor group representation. The association relationship of the original ecological factor group is calculated based on the token sequence of the complete ecological factor group, and the association relationship of the reconstructed ecological factor group is calculated based on the reconstructed ecological factor group representation. The self-supervised pre-training loss is determined based on the differences between the reconstructed ecological factor group representation and the corresponding original ecological factor group representation, as well as the differences between the reconstructed ecological factor group association relationship and the original ecological factor group association relationship.

7. An optimization system for evaluating water conservation function based on self-supervised reconstruction, characterized in that, An optimization method applied to the water conservation function evaluation as described in any one of claims 1-6, comprising: The factor grouping module is designed to acquire ecological indicator data of each evaluation unit in the target evaluation area and divide them into multiple ecological factor groups according to the type of ecological role in the water conservation function evaluation. The token encoding module is configured to encode the ecological factor group into a token to obtain a group token sequence for each evaluation unit, and to mask some tokens according to a preset factor group-level self-supervised masking strategy to obtain a masked token sequence. The model building module is configured to input the masking token sequence into the token reconstruction network, output a pre-training loss, perform self-supervised pre-training based on the pre-training loss, and after pre-training is completed, remove or not use the reconstruction output layer of the network, retain the network body used to generate context representations as the initial representation model of the ecological factor group. The model training module is constructed based on evaluation unit data labeled with water conservation function, to fine-tune the initial representation model, and simultaneously train the initial evaluation model with preset water conservation function, so as to obtain the target representation model and the target evaluation model respectively. The evaluation output module is configured to input the ecological factor group token sequence of the area to be evaluated into the target representation model, output the functional evaluation representation of each evaluation unit, and input the functional evaluation representation into the target evaluation model to output the water conservation function evaluation result.

8. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the steps of the optimization method for evaluating water conservation function based on self-supervised reconstruction as described in any one of claims 1-6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the steps of the optimization method for evaluating water conservation function based on self-supervised reconstruction as described in any one of claims 1-6.

Images