Environmental protection service management method, management system, medium and program product

Abstract

The invention relates to the technical field of environmental protection data analysis, in particular to an environmental protection service management method and management system, a medium and a program product, and the method comprises the steps: collecting multi-dimensional environmental data, determining basic features, spatial-temporal features and derivative features corresponding to the multi-dimensional environmental data, and determining the spatial-temporal features of the multi-dimensional environmental data according to the basic features, the spatial-temporal features and the derivative features; determining the layer weight of each functional layer in the preset machine learning model; optimizing an initial loss function used by the preset machine learning model in the training stage based on each layer weight corresponding to the preset machine learning model to obtain a target loss function; training a preset machine learning model based on the target loss function to obtain a target machine learning model, and importing the multi-dimensional environment data into the target machine learning model to obtain an environment evaluation report corresponding to the multi-dimensional environment data; and obtaining an environmental protection demand provided by the user, and generating an environmental protection service feedback scheme based on the environmental protection demand and the environmental evaluation report. The evaluation precision of the environmental protection evaluation service can be improved.

Description

Technical Field

[0001] This application relates to the field of environmental data analysis technology, and in particular to an environmental service management method, management system, medium, and program product. Background Technology

[0002] With the continued acceleration of global industrialization and urbanization, various environmental problems such as air pollution, water eutrophication, and heavy metal pollution in soil are becoming increasingly prominent. These problems not only seriously threaten the stability and balance of ecosystems but also directly affect human health and safety, hindering the sustainable development of society and the economy. Against this backdrop, environmental data, as a core support for understanding environmental conditions, formulating scientific governance strategies, and accurately assessing governance effectiveness, has become a key research direction urgently needing breakthroughs in the environmental field through its in-depth analysis and efficient utilization. Furthermore, it is a crucial foundation for improving the quality and efficiency of ecological and environmental governance.

[0003] However, current conventional environmental assessment methods fail to fully unleash the core value of environmental data and have significant limitations: on the one hand, they mostly focus on the superficial interpretation of basic environmental data, lacking in-depth mining of the complex relationships behind the data, resulting in the failure to effectively capture the potential environmental patterns contained in the data, and the core value of the data being seriously underestimated; on the other hand, conventional analysis techniques mostly use fixed analytical models for training and evaluation, without adapting and optimizing them for the complex distribution characteristics of environmental data, making it difficult for the model to accurately fit the dynamic changes and complex relationships of environmental data during the training process, thus frequently resulting in problems such as slow convergence speed, overfitting or underfitting, ultimately directly reducing the reliability and accuracy of environmental assessment results and failing to provide strong support for environmental decision-making. Summary of the Invention

[0004] To improve the accuracy of environmental assessment services, this application provides an environmental service management method, management system, medium, and program product.

[0005] Firstly, this application provides an environmental service management method, which adopts the following technical solution: An environmental service management method includes: Collect multi-dimensional environmental data and determine the basic features, spatiotemporal features, and derived features corresponding to the multi-dimensional environmental data; The layer weights of each functional layer in the preset machine learning model are determined based on the basic features, the spatiotemporal features, and the derived features. Based on the weights of each layer corresponding to the preset machine learning model, the initial loss function used by the preset machine learning model during the training phase is optimized to obtain the target loss function; The preset machine learning model is trained based on the target loss function to obtain the target machine learning model. The multi-dimensional environmental data is then imported into the target machine learning model to obtain the environmental assessment report corresponding to the multi-dimensional environmental data. Obtain environmental protection needs provided by users, and generate environmental service feedback schemes based on the environmental protection needs and the environmental assessment report.

[0006] By adopting the above technical solutions and analyzing multi-dimensional environmental data, the problem of underestimated data value can be addressed from multiple perspectives, including static attributes, dynamic changes, and intrinsic relationships. This facilitates improving the accuracy of environmental assessment results from the source. Furthermore, based on fundamental, spatiotemporal, and derived features, the weights of each layer of the pre-defined machine learning model are specifically determined. This allows the model to focus on high-value features and minimize interference from non-critical features, thereby enhancing the model's ability to fit complex environmental situations and avoiding the weakening of key information due to unreasonable weights. This improves the accuracy of the resulting environmental assessment report. Finally, by deeply integrating personalized environmental needs with accurate environmental assessment reports based on user-provided environmental requirements, customized service feedback solutions are generated, improving the fit and accuracy between the environmental service feedback solutions and the user's environmental needs.

[0007] In one possible implementation, determining the layer weights of each functional layer in the preset machine learning model based on the basic features, the spatiotemporal features, and the derived features includes: The basic features, the spatiotemporal features, and the derived features are evaluated using normalized quantification to determine their respective feature contribution. Based on the preset layer examination mapping relationship, the examination features corresponding to each functional layer in the preset machine learning model and the examination attention corresponding to each examination feature are determined. The feature contribution and the attention corresponding to each feature are calculated to obtain the layer weight of each functional layer.

[0008] By adopting the above technical solution, the contribution of each of the three types of features is determined by quantitative evaluation. This facilitates the objective and accurate measurement of the actual impact value of each type of feature and individual features on the environmental assessment results. It avoids the limitations of relying on subjective experience to judge the importance of features in the traditional model weight setting. Based on the preset layer examination mapping relationship, the examination features and examination attention corresponding to each functional layer are determined. This makes it easier for the weight setting of different functional layers of the model to be accurately adapted to their own core task requirements. It avoids the problem of function and feature disconnect caused by the one-size-fits-all allocation of weights in traditional models. Finally, the layer weight is obtained by calculating the feature contribution and examination attention of the corresponding examination features of each functional layer. This facilitates the dual adaptation and deep integration of the value of the feature itself and the task requirements of the functional layer.

[0009] In one possible implementation, optimizing the initial loss function used by the preset machine learning model during the training phase based on the layer weights corresponding to the preset machine learning model to obtain the target loss function includes: Calculate the sum of squares of the weights of each layer as a weight penalty term; The optimization coefficient term is determined based on the derived features; The target loss function is obtained based on the weight penalty term, the optimization coefficient term, the preset regularization penalty coefficient, the preset initial loss function used by the preset machine learning model during the training phase, and the preset loss function calculation formula. The formula for calculating the preset loss function is as follows: ; in, The target loss function; This is the initial loss function used by the preset machine learning model during the training phase; λ is the preset regularization penalty coefficient; k is the optimization coefficient term; This refers to the weight penalty term corresponding to the weights of all layers. Let be the weight of the i-th layer.

[0010] By adopting the above technical solution, the weight penalty term is determined by calculating the sum of squares of the weights of each layer, which effectively limits the excessive growth of the weights of each layer and ensures that the weight distribution is always maintained within a reasonable range. The importance of derived features is quantified to determine the optimization coefficient term, and a stronger penalty is applied to the prediction error corresponding to the derived features based on the optimization coefficient term and the preset regularization penalty coefficient. This avoids evaluation bias caused by the insufficient correction of deep feature errors, thereby improving the accuracy of the model's fitting to the essential laws of the environmental system. Furthermore, the weight penalty term corresponding to the weights of all layers is used to constrain the weight of the loss function, which helps to prevent the model from losing control of weights due to excessive focus on derived features, thereby improving the accuracy of the model training results.

[0011] In one possible implementation, determining the optimization coefficient term based on the derived features includes: The environmental adaptability coefficient is determined based on the multi-dimensional environmental data. Obtain the feature contribution degree corresponding to the derived feature, and determine the optimization weight based on the feature contribution degree and the environment adaptation coefficient; Identify the interaction feature group and environmental interference features corresponding to the derived features, wherein the interaction feature group contains at least two interaction features, and the at least two interaction features are basic features and / or spatiotemporal features; The mutual information value corresponding to the interactive feature group is determined based on the preset mutual information method, and the partial correlation coefficient corresponding to the environmental interference feature is determined based on the preset partial correlation correction algorithm. The feature coupling degree corresponding to the derived feature is determined based on the mutual information value and the partial correlation coefficient. The optimization coefficient term is determined based on the optimization weight and the feature coupling degree.

[0012] By adopting the above technical solution, the weights are optimized through a two-dimensional weighted calculation of feature contribution and environmental adaptation coefficient. This allows for the retention of the inherent value of derived features while avoiding weight bias caused by single-dimensional quantification through dynamic adjustment of the environmental adaptation coefficient. In addition, by clearly distinguishing between interactive feature groups and environmental interference features, it is easier to identify the core components of derived features, ensuring that the coupling degree calculation focuses on the correlation that truly drives the derived features and avoids interference from irrelevant features, thereby improving the accuracy of feature coupling degree. Finally, by optimizing the weights and feature coupling degree, the optimization coefficient term is determined, which ensures that the optimization coefficient term can accurately map the comprehensive contribution of derived features to environmental assessment.

[0013] In one possible implementation, the generation of an environmental service feedback scheme based on the environmental protection needs and the environmental assessment report includes: Identify explicit demand dimension parameters from the environmental protection demands, and determine implicit demand dimension parameters based on the environmental protection demands, the multi-dimensional environmental data, and the environmental assessment report; Based on the explicit demand dimension parameters and the implicit demand dimension parameters, determine the demand priority vector corresponding to the environmental protection demand and the environmental assessment report; An environmental problem label matrix is ​​determined based on the environmental assessment report, and an environmental problem structure vector is determined based on the environmental problem label matrix. The environmental problem structure vector includes problem labels, impact scores, and improvement potential values. Based on the demand priority vector and the environmental problem structure vector, the correlation matching degree between each problem label and the demand indicator dimension is determined; The target issue label is determined based on the preset matching degree threshold and the correlation matching degree corresponding to each issue label. The target feedback template corresponding to the target issue label is obtained from the feedback template library based on the target issue label. An environmental service feedback scheme is generated based on the target feedback template and the target issue label.

[0014] By adopting the above technical solution, explicit demand dimension parameters are directly identified from environmental protection needs, ensuring the directness and accuracy of demand capture. At the same time, by combining environmental protection needs, multi-dimensional environmental data, and high-precision environmental assessment reports, implicit demand dimension parameters are derived in reverse, which helps to fill the blind spots of traditional demand mining. The determination of dual-dimensional demand parameters facilitates the improvement of the comprehensiveness of environmental protection service solutions. In addition, by analyzing the correlation and matching degree between problem tags and demand indicator dimensions, target problem tags with high correlation and matching degree are selected, while problem tags with low matching degree are weakened, avoiding the generalization and lack of focus in environmental protection service feedback solutions.

[0015] In one possible implementation, determining the environmental protection requirement and the corresponding requirement priority vector for the environmental assessment report based on the explicit requirement dimension parameters and the implicit requirement dimension parameters includes: The parameters of each explicit demand dimension are quantified into explicit indicator values, and the parameters of each implicit demand dimension are quantified into implicit indicator values. Identify environmental characteristic parameters from the multi-dimensional environmental data; The environmental characteristic parameters are matched with each explicit demand dimension parameter to obtain the first demand matching value corresponding to each explicit demand dimension parameter. Based on the explicit index value corresponding to each explicit demand dimension parameter and the first demand matching value, the explicit priority vector is determined. The environmental feature parameters are matched with each implicit demand dimension parameter to obtain the second demand matching value corresponding to each implicit demand dimension parameter. Based on the implicit index value and the second demand matching value corresponding to each implicit demand dimension parameter, the implicit priority vector is determined. Based on the explicit priority vector and the implicit priority vector, the requirement priority vector corresponding to the environmental protection requirement and the environmental assessment report is determined.

[0016] By adopting the above technical solution, explicit demand dimension parameters are quantified into explicit indicator values, and implicit demand dimension parameters are quantified into implicit indicator values. This facilitates the transformation of vague demand requests into clear quantitative data, avoiding the problem of immeasurable demand importance. In addition, by extracting environmental feature parameters from multi-dimensional environmental data and then accurately matching them with explicit and implicit demand dimension parameters to determine the demand priority vector, the adaptability of the demand priority vector to the current environmental scenario can be significantly enhanced, ensuring that the priority ranking conforms to the actual environmental conditions.

[0017] Secondly, this application provides a management system, which adopts the following technical solution: A management system comprising: At least one processor; Memory; At least one application, wherein the at least one application is stored in memory and configured to be executed by at least one processor, the at least one application being configured to: execute the above-described environmental service management method.

[0018] Thirdly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium includes: a computer program stored thereon that can be loaded by a processor and execute the above-described environmental service management method.

[0019] Fourthly, this application provides a computer program product, which adopts the following technical solution: A computer program product includes a computer program that, when executed by a processor, implements the aforementioned environmental service management method.

[0020] In summary, this application includes at least one of the following beneficial technical effects: By analyzing multi-dimensional environmental data, the problem of underestimated data value can be addressed from multiple perspectives, including static attributes, dynamic changes, and intrinsic relationships. This facilitates improving the accuracy of environmental assessment results from the source. Furthermore, based on basic features, spatiotemporal features, and derived features, the weights of each layer of the preset machine learning model are determined in a targeted manner. This allows the model to focus on high-value features and weaken the interference of non-critical features, thereby improving the fitting ability of the preset machine learning model to complex environmental situations and avoiding the weakening of key information due to unreasonable weights. This improves the accuracy of the determined environmental assessment report. Finally, by deeply integrating personalized environmental needs with accurate environmental assessment reports based on user-provided environmental requirements, customized service feedback solutions are generated, improving the adaptability and accuracy between the environmental service feedback solutions and the environmental needs provided by users.

[0021] By directly identifying explicit demand dimensions from environmental protection needs, the directness and accuracy of demand capture are ensured. At the same time, by combining environmental protection needs, multi-dimensional environmental data, and high-precision environmental assessment reports, implicit demand dimensions are derived in reverse, which helps to fill the blind spots of traditional demand mining. The determination of dual-dimensional demand parameters facilitates the improvement of the comprehensiveness of environmental protection service solutions. In addition, by analyzing the correlation and matching degree between problem tags and demand indicator dimensions, target problem tags with high correlation and matching degree are selected, while problem tags with low matching degree are weakened, avoiding the generalization and lack of focus in environmental protection service feedback solutions. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating an environmental service management method according to an embodiment of this application; Figure 2 This is a schematic diagram of a process for generating an environmental service feedback scheme in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a management system according to an embodiment of this application. Detailed Implementation

[0023] The following is in conjunction with the appendix Figures 1 to 3 This application will be described in further detail.

[0024] After reading this specification, those skilled in the art may make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] It should be noted that, in the optional embodiments of this application, the data related to object information, when applied to specific products or technologies, requires the permission or consent of the object. Furthermore, the collection, use, and processing of this data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. In other words, if the embodiments of this application involve data related to an object, it must be obtained with the object's authorization and consent, the authorization and consent of relevant departments, and in accordance with the relevant laws, regulations, and standards of the country and region. If the embodiments involve personal information, the acquisition of all personal information requires the individual's consent. If sensitive information is involved, the separate consent of the information subject is required. The embodiments also need to be implemented with the object's authorization and consent.

[0027] Specifically, this application provides an environmental service management method executed by a management system, which can be a server or a terminal device. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. The terminal device and the server can be directly or indirectly connected via wired or wireless communication, and this application does not impose any limitations on this connection.

[0028] refer to Figure 1 , Figure 1 This is a flowchart illustrating an environmental service management method according to an embodiment of this application. The method includes steps S110-S150, wherein: Step S110: Collect multi-dimensional environmental data and determine the basic features, spatiotemporal features, and derived features corresponding to the multi-dimensional environmental data.

[0029] Specifically, multi-dimensional environmental data includes, but is not limited to, atmospheric environmental data, water environmental data, soil environmental data, ecological environment data, human activity data, and geospatial data. These can be acquired through fixed monitoring, mobile monitoring, and data access methods to ensure the breadth and accuracy of the multi-dimensional environmental data coverage. Ground monitoring stations or sensor arrays can be deployed in key areas such as industrial zones, residential areas, and ecological protection zones to facilitate the collection of relevant environmental data. Additionally, drone patrols equipped with gas sensors, hyperspectral cameras, and other devices can be used to conduct targeted and precise data collection in suspected pollution areas and remote areas. Furthermore, publicly available data from relevant environmental protection departments, such as the national real-time air quality release platform and the pollutant discharge permit management information platform, can be accessed. The specific methods for acquiring multi-dimensional environmental data are not limited in this application embodiment.

[0030] Based on a preset feature recognition algorithm, the system can identify monitoring indicators contained in multi-dimensional environmental data, and then determine the basic, spatiotemporal, and derived features corresponding to all monitoring indicators. These monitoring indicators can include: atmospheric environment indicators (e.g., dust concentration, air pollutant concentration, temperature, humidity, light intensity); water environment indicators (e.g., dissolved oxygen, pH value, water pollutant concentration, water level, water temperature, flow velocity); soil environment indicators (e.g., soil moisture content, soil pollutant concentration); ecological environment indicators (e.g., vegetation coverage, green area, wetland area, species richness); human activity indicators (e.g., exhaust emissions, carbon sequestration, electricity consumption, pesticide use); and geospatial indicators (e.g., regional functional zoning, landforms).

[0031] Basic features can be statistical characteristics such as real-time values, mean, variance, extreme values, and rate of change of multiple monitoring indicators. The basic features can be obtained by quantifying each monitoring indicator. Since there are many sources of multi-dimensional environmental data and a large number of monitoring indicators, obtaining basic features by quantifying multi-dimensional environmental data can help solve the problem of the original data being messy and unusable directly.

[0032] Spatiotemporal features can be extracted based on the time series and spatial distribution of various monitoring indicators, such as the seasonal variation trend of O3 concentration, the PM10 concentration gradient in different regions, the decreasing trend of ammonia nitrogen concentration in upstream and downstream sections of rivers, and the decay pattern of pollution concentration around pollution sources. These features are used to capture the dynamic evolution and spatial correlation characteristics of multidimensional environmental data, making up for the limitation that basic features can only reflect static statistical states.

[0033] Derived features are implicit patterns generated through linear or nonlinear coupling and multidimensional correlation analysis of basic features, spatiotemporal features, or basic features and spatiotemporal features that are inherently related. The core is the correlation analysis of multiple monitoring indicators. Based on basic features and spatiotemporal features, they can be extracted through linear coupling, nonlinear correlation and scenario-based adaptation analysis of multidimensional monitoring indicators. For example, the coupling index between the average PM2.5 concentration and relative humidity, and the adsorption coupling coefficient between soil heavy metal content and organic matter content, etc., are used to explore the deep intrinsic correlation and implicit patterns behind multidimensional environmental data, and make up for the limitations of static description of basic features by a single indicator and dynamic or spatial correlation of spatiotemporal features by a single dimension.

[0034] Step S120: Determine the layer weights of each functional layer in the preset machine learning model based on basic features, spatiotemporal features, and derived features.

[0035] Specifically, the method for determining the corresponding basic features, spatiotemporal features, and derived features based on multi-dimensional environmental data is not specifically limited in this application embodiment. The key point is to use the determined basic features, spatiotemporal features, and derived features to participate in the training process of the preset machine learning model. By improving the layer weights of each functional layer in the preset machine learning model, the loss function used in the model training process is optimized, rather than using a fixed loss function for model training. By improving the fit between the layer weights and the multi-dimensional environmental data, the fit between the loss function and the actual evaluation environment is improved, thereby facilitating the improvement of the accuracy of the model training results.

[0036] Furthermore, to ensure that the weight settings of different functional layers in the preset machine learning model are accurately adapted to the core task requirements, the layer weights of each functional layer in the preset machine learning model are determined based on basic features, spatiotemporal features, and derived features. Specifically, this may include: The basic features, spatiotemporal features, and derived features are evaluated using normalized quantification to determine their respective feature contributions. Based on the preset layer examination mapping relationship, the examination features corresponding to each functional layer in the preset machine learning model and the examination attention corresponding to each examination feature are determined. The feature contributions and examination attention corresponding to each examination feature of each functional layer are calculated to obtain the layer weights corresponding to each functional layer.

[0037] Specifically, a pre-defined analysis of variance and a pre-defined random forest model can be used to determine the feature contribution of basic features when the model completes training. First, the F-value between a single basic feature and its corresponding target task indicator can be calculated; a larger F-value indicates a stronger ability of the basic feature to distinguish the target task indicator. Then, all basic features are input into the pre-defined random forest model, and the feature importance score output by the model is extracted. Finally, the F-values ​​of each basic feature and their corresponding feature importance scores are weighted to obtain the feature contribution of each basic feature. The target task indicator corresponding to the basic feature is the output result when the model completes training, and the feature contribution of the basic feature represents the degree to which the basic feature helps the model capture the basic static attributes and key statistical information of environmental data, thereby accurately completing the core environmental protection task.

[0038] The feature contribution degree corresponding to the spatiotemporal feature can be determined by using the preset mutual information method. Specifically, the mutual information value between the spatiotemporal feature and the corresponding target task indicator can be calculated first to quantify the nonlinear correlation strength. The larger the mutual information value, the more target information the spatiotemporal feature contains. Then, its spatial correlation coefficient with the distribution of pollution sources can be calculated. After normalizing the mutual information value and the spatial correlation coefficient, they are summed to obtain the feature contribution degree of each spatiotemporal feature. The feature contribution degree of the spatiotemporal feature is the degree to which the spatiotemporal feature helps the model capture the dynamic evolution law and spatial correlation characteristics of environmental data, and thus accurately complete the core environmental protection task.

[0039] The derived features can be input into the trained coupled model using a preset SHAP value analysis method and a preset feature replacement method. The marginal contribution of the derived features to the coupled model's output is quantified using the preset SHAP value analysis method. Then, the values ​​of the derived features are randomly shuffled, and the percentage decrease in the coupled model's prediction accuracy is calculated. Based on the marginal contribution percentage and the percentage decrease percentage, the feature contribution degree of the derived features is determined. The feature contribution degree of the derived features represents the degree to which the model helps capture the multi-dimensional intrinsic correlations and implicit coupling patterns of environmental data, thereby accurately completing core environmental protection tasks.

[0040] The preset machine learning model contains multiple functional layers, each with different evaluation features. That is, changes in these evaluation features to varying degrees have different impacts on each functional layer. Therefore, the evaluation features and their corresponding evaluation attention levels can be determined based on the preset layer evaluation mapping relationship. For example, functional layer a might have evaluation features of basic features and derived features, with the basic features having an evaluation attention level of 20% and the derived features having an evaluation attention level of 50%. Functional layer b might have evaluation features of basic features and spatiotemporal features, with the basic features having an evaluation attention level of 30% and the derived features having an evaluation attention level of 10%. The preset layer evaluation mapping relationship includes the evaluation features and their corresponding evaluation attention levels for each functional layer. Specific details are not limited in this embodiment and can be determined by relevant personnel based on historical experimental data and uploaded to the management system.

[0041] For any functional layer, the layer weight is calculated based on the feature contribution of the corresponding features and the attention level of each feature. For example, if the feature contribution of the basic feature is 30%, the feature contribution of the spatiotemporal feature is 40%, and the feature contribution of the derived feature is 30%, and the features for functional layer a are both basic and derived features, with the attention level of the basic feature being 20% ​​and the attention level of the derived feature being 50%, then the layer weight for functional layer a is 20%*30% + 50%*30% = 21%. The layer weight for each functional layer can be obtained using the above method.

[0042] Step S130: Optimize the initial loss function used by the preset machine learning model during the training phase based on the weights of each layer corresponding to the preset machine learning model to obtain the target loss function.

[0043] Specifically, the preset machine learning model is used to analyze multi-dimensional environmental data. To improve the accuracy of the preset machine learning model in the process of analyzing multi-dimensional environmental data, it is necessary to optimize the initial loss function based on the determined weights of each layer, and then train the preset machine learning model based on the target loss function and a large number of samples of multi-dimensional environmental data. When optimizing the initial loss function based on the weights of each layer, the target penalty term corresponding to any layer weight can be determined according to the preset penalty term mapping relationship, and then the initial loss function is optimized based on the target penalty term. The preset penalty term mapping relationship is the correspondence between each layer weight and the target penalty term. The specific content is not specifically limited in this embodiment of the application, and can be determined by relevant personnel based on historical experimental data and then uploaded to the management system.

[0044] Furthermore, to improve the accuracy of the model training results, the initial loss function used by the preset machine learning model during the training phase is optimized based on the weights of each layer corresponding to the preset machine learning model, resulting in a target loss function. Specifically, this may include: calculating the sum of squares of the weights of each layer as a weight penalty term; determining optimization coefficients based on derived features; and obtaining the target loss function based on the weight penalty term, optimization coefficients, preset regularization penalty coefficients, the initial loss function used by the preset machine learning model during the training phase, and the preset loss function calculation formula. The preset loss function calculation formula is as follows: ; in, The target loss function; This is the initial loss function used by the preset machine learning model during the training phase. λ is the preset regularization penalty coefficient; k is the optimization coefficient term; This refers to the weight penalty term corresponding to the weights of all layers. Let be the weight of the i-th layer.

[0045] Specifically, the weight penalty term is obtained by calculating the sum of squares of the weights of each layer. Weight penalty term = ,in, To predetermine the layer weights of the i-th functional layer in the machine learning model, a weight penalty term is determined by calculating the sum of the squares of the weights of each layer. This effectively limits the excessive growth of the weights of each layer, ensuring that the weight distribution remains within a reasonable range. By introducing a pre-defined regularization penalty coefficient λ during the optimization of the initial loss function, the core purpose is to address the model overfitting problem, balancing fitting accuracy and model complexity. This ensures that the model performs well on training data while possessing strong generalization ability. A larger pre-defined regularization penalty coefficient λ results in a heavier penalty. The specific pre-defined regularization penalty coefficient is not specifically limited in this embodiment and can be set by relevant personnel according to actual needs. An optimization coefficient term k is determined by analyzing the feature coupling degree corresponding to the derived features and considering the interference caused by irrelevant features in the environment. The importance of the derived features is quantified to determine the optimization coefficient term. Based on the optimization coefficient term and the pre-defined regularization penalty coefficient, a stronger penalty is applied to the prediction error corresponding to the derived features, avoiding evaluation bias caused by insufficient correction of deep feature errors.

[0046] Furthermore, to improve the accuracy of determining the optimization coefficient term, the method provided in this application embodiment, when determining the optimization coefficient term based on derived features, may specifically include: The environment adaptation coefficient is determined based on multi-dimensional environmental data; the feature contribution corresponding to the derived feature is obtained, and the optimization weight is determined based on the feature contribution and the environment adaptation coefficient; the interactive feature group and environmental interference feature corresponding to the derived feature are identified, wherein the interactive feature group contains at least two interactive features, and the at least two interactive features are basic features and / or spatiotemporal features; the mutual information value corresponding to the interactive feature group is determined based on the preset mutual information method, and the partial correlation coefficient corresponding to the environmental interference feature is determined based on the preset partial correlation correction algorithm; the feature coupling degree corresponding to the derived feature is determined based on the mutual information value and the partial correlation coefficient; and the optimization coefficient term is determined based on the optimization weight and the feature coupling degree.

[0047] Specifically, multiple environmental parameter features can be identified from multi-dimensional environmental data based on a preset feature recognition algorithm. Different combinations of environmental parameter features correspond to different environmental assessment types. First, the target environmental assessment type corresponding to the combination of environmental parameter features in the multi-dimensional environmental data is determined according to the preset assessment type mapping relationship. Then, the environmental adaptation coefficient corresponding to the target environmental assessment type is determined according to the preset adaptation coefficient mapping relationship. The preset assessment type mapping relationship is the correspondence between different combinations of environmental parameter features and the target environmental assessment type, and the preset adaptation coefficient mapping relationship is the correspondence between the target environmental assessment type and the environmental adaptation coefficient. The specific content is not specifically limited in this embodiment of the application and can be determined by relevant personnel based on historical experimental data and uploaded to the management system in advance. For example, when the target environmental assessment type is air pollution, the environmental adaptation coefficient η is 1.3; when the target environmental assessment type is water body steady-state pollution, the environmental adaptation coefficient η is 1.0.

[0048] After determining the environment adaptation coefficient, the feature contribution degree corresponding to the derived feature can be comprehensively analyzed with the environment adaptation coefficient to determine the corresponding optimization weight. The optimization weight is calculated by a two-dimensional weighted average of the feature contribution degree and the environment adaptation coefficient. This facilitates the retention of the inherent value of the derived feature while dynamically adjusting the environment adaptation coefficient to avoid weight bias caused by single-dimensional quantification. The optimization weight corresponding to the feature contribution degree and the environment adaptation coefficient can be determined according to a preset optimization weight mapping relationship. This preset optimization weight mapping relationship is the correspondence between the parameter combination of the feature contribution degree and the environment adaptation coefficient and the optimization weight. The specific details are not specifically limited in this embodiment and can be determined by relevant personnel based on historical experimental data and uploaded to the management system.

[0049] An interactive feature group is a combination of basic features and / or spatiotemporal features that constitute a derived feature. That is, the derived feature is generated by coupling and computation of the basic features and / or spatiotemporal features contained in the interactive feature group. A preset coupling feature recognition algorithm can identify the corresponding interactive feature group based on the derived features. Each interactive feature group contains at least two interactive features, which are basic features and / or spatiotemporal features. For example, the interactive feature group can contain at least two interactive features, which could be basic feature 1 and basic feature 2, or spatiotemporal feature 1 and spatiotemporal feature 2, or a combination of basic features and spatiotemporal features. Environmental interference features are features that have no core logical connection to the derived features but may interfere with their effectiveness, such as noise or other irrelevant environmental factors. These can also be obtained from a database based on the derived features using a preset feature recognition algorithm. The database contains environmental interference features corresponding to each derived feature, which can be determined by relevant personnel based on historical experimental data and uploaded in advance. Specific content is not specifically limited in this embodiment.

[0050] Taking an interaction feature group containing two interaction features as an example, the mutual information value corresponding to the interaction feature group can be obtained based on the preset mutual information method, the two interaction features within the interaction feature group, and the preset mutual information value calculation formula. The preset mutual information value calculation formula is as follows: ; in, These are two interactive features within an interactive feature group; This represents the mutual information value between two interactive features within the interactive feature group.

[0051] Then, based on a preset partial correlation correction algorithm, the partial correlation coefficient corresponding to the environmental interference characteristics is determined. The formula for calculating the partial correlation coefficient is: Partial correlation coefficient = ; where Z represents the environmental disturbance characteristics.

[0052] Based on the feature coupling degree calculation formula, the feature coupling degree corresponding to the derived feature can be obtained by calculating the mutual information value and the partial correlation coefficient. The feature coupling degree calculation formula is as follows: ;in, This represents the feature coupling degree corresponding to the derived feature.

[0053] Finally, by weighting the optimization weights and feature coupling degrees, the optimization coefficients can be obtained. By clearly distinguishing between interactive feature groups and environmental interference features, it is easier to identify the core components of derived features, ensuring that the coupling degree calculation focuses on the correlation that truly drives the derived features, avoiding interference from irrelevant features, thereby improving the accuracy of feature coupling degrees. Finally, by optimizing the weights and feature coupling degrees, the optimization coefficients are determined, ensuring that the optimization coefficients accurately map the comprehensive contribution of derived features to environmental assessment.

[0054] Step S140: Train the preset machine learning model based on the target loss function to obtain the target machine learning model, import the multi-dimensional environmental data into the target machine learning model, and obtain the environmental assessment report corresponding to the multi-dimensional environmental data.

[0055] Step S150: Obtain the environmental protection needs provided by the user, and generate an environmental service feedback plan based on the environmental protection needs and the environmental assessment report.

[0056] Specifically, the preset machine learning model is iteratively trained based on the target loss function until the loss entropy value is lower than the preset loss threshold to obtain the target machine learning model. Multi-dimensional environmental data is then imported into the target machine learning model as model input parameters. The target machine learning model can directly complete the analysis of multi-dimensional environmental data and obtain an environmental assessment report.

[0057] After generating the environmental assessment report, it is not directly fed back to the user. Instead, the environmental protection needs provided by the user are obtained, the conditions included in the environmental protection needs are identified, and finally, relevant assessment data are extracted from the environmental assessment report based on the conditions and integrated into a personalized environmental service feedback plan. This plan is then pushed to the customer through the user interaction module. The conditions can include environmental budget, implementation difficulty, etc.

[0058] In this embodiment of the application, by analyzing multi-dimensional environmental data, it is easy to solve the problem of underestimating data value from multiple perspectives, including static attributes, dynamic changes, and intrinsic relationships. This facilitates improving the accuracy of environmental assessment results from the source. In addition, based on basic features, spatiotemporal features, and derived features, the weights of each layer of the preset machine learning model are determined in a targeted manner. This allows the model to focus on high-value features and weaken the interference of non-critical features, thereby improving the fitting ability of the preset machine learning model to complex environmental situations and avoiding the problem of weakening key information due to unreasonable weights. This facilitates improving the accuracy of the determined environmental assessment report. Finally, by deeply integrating the personalized environmental needs with the accurate environmental assessment report based on the environmental needs provided by the user, a customized service feedback solution is generated, improving the adaptability and accuracy between the environmental service feedback solution and the environmental needs provided by the user.

[0059] Furthermore, to avoid the environmental service feedback scheme becoming generalized and lacking focus, the method provided in this application, when generating an environmental service feedback scheme based on environmental needs and an environmental assessment report, includes steps S210-S250, such as... Figure 2 As shown, where: Step S210: Identify explicit demand dimension parameters from environmental protection needs, and determine implicit demand dimension parameters based on environmental protection needs, multi-dimensional environmental data, and environmental assessment reports.

[0060] Specifically, explicit demand dimension parameters can be identified from environmental protection needs based on a preset feature recognition algorithm. In this embodiment, the preset feature recognition algorithm is not specifically limited. Explicit demand dimension parameters may include: scenario dimension parameters: application scenario, such as home, enterprise, park, community, etc.; spatial scale, such as a 100-square-meter residence, a 100,000-square-meter park; target dimension parameters: core objectives, such as health protection, compliance, cost saving, ecological improvement, etc.; constraint dimension parameters: budget type, such as high budget, medium budget, low budget; execution cycle, such as urgent, short-term, long-term, etc.; operational difficulty, such as easy, medium, difficult; preference dimension parameters: service type, such as equipment procurement, technical consulting, operation and maintenance services, customized governance, etc.; cooperation mode, such as online, offline, on-site, etc.

[0061] Implicit demand parameters are those not directly stated by relevant users, but are potential demand parameters derived through logical reasoning and data correlation based on environmental protection needs, multi-dimensional environmental data, and environmental assessment reports. The specific process can be as follows: The system normalizes environmental protection needs, multi-dimensional environmental data, and environmental assessment reports. Core keywords are extracted from these data using a pre-defined feature recognition algorithm and a pre-defined keyword list. All core keywords are then integrated into a keyword dictionary. This pre-defined keyword list, determined by relevant staff based on historical experimental data, is uploaded to the management system in advance. The list includes keywords corresponding to environmental protection needs, such as scenario-specific keywords (residents, communities, industrial zones, etc.), target keywords (compliance, health protection, cost savings, etc.), constraint keywords (high budget, low budget, uninterrupted production, etc.), preference keywords (equipment procurement, technical consultation, operation and maintenance services, etc.), basic keywords (soil pH, wind speed, equipment age, etc.), spatiotemporal keywords (season, heating season, pollution source distribution, etc.), derived keywords (pollution transmission coupling index, environmental credit-subsidy correlation index, etc.), risk keywords (exceeding standards, minor pollution, work stoppage losses, etc.), causal keywords (industrial emissions, traffic pollution, agricultural non-point source pollution, etc.), and governance potential keywords (equipment upgrades, reagent optimization, process adjustments, etc.). Based on the core keywords in the keyword dictionary as nodes and logical relevance as edges, a core keyword association network is constructed, and association paths are extracted from it. For example, the demand keyword "medium budget" → the data keyword "equipment age 10 years" → the evaluation keyword "low-cost renovation plan". Closely related association paths are then clustered to obtain multiple association clusters. Finally, explicit preference extension keywords are extracted from the association clusters to clarify the user's potential service needs. For example, the associated keywords: on-site service (preference) + equipment calibration (data) + lack of maintenance team (evaluation) correspond to potential demand parameters such as service type preference (technical consultation + monthly on-site maintenance). The specific method for determining potential demand parameters is not specifically limited in this embodiment, nor is the number of potential demand parameters determined. The key point is that the method provided in this embodiment needs to consider both explicit and implicit demand dimension parameters when generating environmental service feedback plans.

[0062] Step S220: Based on the explicit demand dimension parameters and the implicit demand dimension parameters, determine the demand priority vector corresponding to the environmental protection demand and the environmental assessment report.

[0063] Specifically, a preset feature recognition algorithm can be used to identify the demand features in the explicit and implicit demand dimension parameters. Then, based on the demand features and the preset feature priority mapping relationship, the feature priority corresponding to each demand feature is determined. Finally, a demand priority vector is generated based on each feature priority. The preset feature priority mapping relationship is the correspondence between demand features and feature priorities; its specific content is not specifically limited in this embodiment and can be determined by relevant personnel based on historical experimental data and uploaded to the management system in advance. Furthermore, to enhance the adaptability of the demand priority vector to the current environmental scenario, a demand priority vector corresponding to environmental protection needs and environmental assessment reports is determined based on the explicit and implicit demand dimension parameters. This can specifically include: Each explicit demand dimension parameter is quantified into explicit indicator values, and each implicit demand dimension parameter is quantified into implicit indicator values. Environmental characteristic parameters are identified from multi-dimensional environmental data. The environmental characteristic parameters are matched with each explicit demand dimension parameter to obtain the first demand matching value corresponding to each explicit demand dimension parameter. Based on the explicit indicator values ​​and the first demand matching values ​​corresponding to each explicit demand dimension parameter, an explicit priority vector is determined. The environmental characteristic parameters are matched with each implicit demand dimension parameter to obtain the second demand matching value corresponding to each implicit demand dimension parameter. Based on the implicit indicator values ​​and the second demand matching values ​​corresponding to each implicit demand dimension parameter, an implicit priority vector is determined. Based on the explicit priority vector and the implicit priority vector, a demand priority vector corresponding to environmental protection needs and environmental assessment reports is determined.

[0064] Specifically, the demand priority vector consists of an explicit priority vector and an implicit priority vector. When determining the explicit priority vector, each explicit demand dimension parameter can first be quantified into explicit indicator values ​​based on a preset normalization algorithm. Then, an implicit feature recognition algorithm is used to identify objective data indicators strongly correlated with the explicit demand dimension parameters from the multi-dimensional environmental data. For example, when the scenario dimension of the explicit demand dimension parameters is a chemical enterprise, the corresponding objective data indicators could be regional type, pollution source distribution density, etc.; when the constraint dimension of the explicit demand dimension parameters is a medium budget, the corresponding objective data indicators could be equipment age, unit cost of renovation, production downtime losses, etc. Identification can be performed based on a preset objective data indicator form, which contains environmental feature parameters corresponding to each explicit demand dimension. The environmental characteristic parameters are matched with the explicit demand dimension parameters to obtain the first demand matching value for each explicit demand dimension parameter. Then, the explicit priority value for each explicit demand dimension parameter is calculated by summing the first demand matching value and the explicit indicator value. Finally, the explicit demand dimension parameters are sorted according to their explicit priority values ​​to obtain the explicit priority vector. The method for determining the implicit priority vector is the same as that for determining the explicit priority vector, and will not be elaborated here.

[0065] By quantifying explicit demand dimension parameters into explicit indicator values ​​and implicit demand dimension parameters into implicit indicator values, it is easy to transform vague demand requests into clear quantitative data, avoiding the problem of the importance of demands being immeasurable. In addition, by extracting environmental feature parameters from multi-dimensional environmental data and then accurately matching them with explicit and implicit demand dimension parameters respectively, the demand priority vector can be determined. This can significantly enhance the adaptability of the demand priority vector to the current environmental scenario and ensure that the priority ranking fits the actual environmental conditions.

[0066] Step S230: Determine the environmental problem label matrix based on the environmental assessment report, and determine the environmental problem structure vector based on the environmental problem label matrix. The environmental problem structure vector includes problem label, impact score and improvement potential value.

[0067] Specifically, multiple problem tags can be identified from the environmental assessment report based on a preset feature recognition algorithm. These problem tags include three types: pollution type, exceedance index, and problem severity. Primary label (pollution type), including but not limited to air pollution, water pollution, noise pollution, soil pollution, etc.; Secondary labels (exceeding standards) include, but are not limited to, PM2.5, COD, VOCs, and noise levels in decibels; Three levels of labels (problem severity), including but not limited to mild exceedance, moderate exceedance, and severe exceedance.

[0068] Different problem labels correspond to different impact coefficients. The impact coefficients for different problem labels can be determined based on a preset impact coefficient mapping relationship. This preset mapping relationship is the correspondence between problem labels and impact coefficients. For example, the impact coefficient for air pollution is 0.8, for water pollution it is 0.9, for noise pollution it is 0.5, for COD exceeding the standard it is 0.6, and for slight exceeding the standard it is 0.3. The specific content of the preset impact coefficient mapping relationship is not specifically limited in this embodiment; it can be determined by relevant personnel based on historical experimental data and uploaded to the management system in advance. After identifying the label parameter information corresponding to each problem label, the problem severity value of each label is obtained by quantifying each label parameter information. Finally, the impact score of each problem label is calculated by combining the problem severity value and the impact coefficient. The impact score calculation formula is: I = Problem Severity Value * Impact Coefficient.

[0069] Based on the preset improvement potential value mapping relationship, problem label type, and the influence score of each problem label, the improvement potential value corresponding to each problem label can be determined. The preset improvement potential value mapping relationship is the correspondence between the parameter combination of problem label type and influence score and the improvement potential value. The specific content is not specifically limited in the embodiments of this application.

[0070] Based on each problem label, impact score, and improvement potential value, a structured vector of environmental problems for each problem label can be obtained.

[0071] Step S240: Based on the demand priority vector and the environmental problem structure vector, determine the correlation and matching degree between each problem label and the demand indicator dimension.

[0072] Step S250: Determine the target issue label based on the preset matching degree threshold and the correlation matching degree corresponding to each issue label; obtain the target feedback template corresponding to the target issue label from the feedback template library based on the target issue label; and generate an environmental service feedback plan based on the target feedback template and the target issue label.

[0073] Specifically, the demand priority vector and the environmental problem structure vector are matched to obtain the correlation matching degree between each problem label and each demand indicator dimension in the demand priority vector. Based on the preset matching degree threshold, the target problem label with a correlation matching degree higher than the preset matching degree threshold is determined from multiple problem labels. The specific preset matching degree threshold is not specifically limited in this application embodiment.

[0074] Different target issue tags correspond to different target feedback templates. The target feedback template for each target issue tag can be extracted from the feedback module library. The target feedback template includes content to be filled in, such as issue diagnosis, core services, optional services, and execution paths. Specifically, the content to be filled in for issue diagnosis includes, but is not limited to, pollution level, exceedance data, and impact. This content can be filled in based on the target issue tag and environmental assessment report. The content to be filled in for core services includes, but is not limited to, adaptation technology, equipment parameters, and effects. This content can be filled in based on the target issue tag and preset treatment technology specifications. The content to be filled in is as follows: The content to be filled in for optional services includes, but is not limited to, value-added services, costs, and cycles. The content to be filled in for optional services can be based on the target issue tags and the preset service resource library. The content to be filled in for execution paths includes, but is not limited to, steps, time, and acceptance criteria. The content to be filled in for execution paths can be based on the target issue tags and the preset national standard acceptance requirements. For example, the execution path includes: Step 1, equipment procurement, 1-3 working days; Step 2, on-site installation and commissioning, 1 working day; Step 3, testing and acceptance after 7 days, PM2.5≤35μg / m³ is considered qualified.

[0075] After filling in the target feedback template, the environmental service feedback plan corresponding to the target issue label can be obtained. When there are two or more target issue labels, each target issue label can be filled into its corresponding target feedback template, and the results can be combined to obtain the environmental service feedback plan. By directly identifying explicit demand dimension parameters from environmental needs, the directness and accuracy of demand capture are ensured. At the same time, by combining environmental needs, multi-dimensional environmental data, and high-precision environmental assessment reports, implicit demand dimension parameters are derived in reverse, which helps to fill the blind spots of traditional demand mining. By determining the dual-dimensional demand parameters, the comprehensiveness of the environmental service plan can be improved. In addition, by the correlation and matching degree between the issue label and the demand indicator dimension, the target issue labels with high correlation and matching degree are selected, and the issue labels with low matching degree are weakened, so as to avoid the environmental service feedback plan being generalized and unfocused.

[0076] This application provides a management system, such as... Figure 3 As shown, Figure 3 The management system 300 shown includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the management system 300 may also include a transceiver 304. It should be noted that in practical applications, the transceiver 304 is not limited to one type, and the structure of this management system 300 does not constitute a limitation on the embodiments of this application.

[0077] Processor 301 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 can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0078] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The symbol is represented by only one line, but this does not mean that there is only one bus or one type of bus.

[0079] The memory 303 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0080] The memory 303 is used to store application code that executes the solution of this application, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the foregoing method embodiments.

[0081] The management system includes, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Tablet PCs), PMPs (Portable Multimedia Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. It can also include servers. Figure 3 The management system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0082] This application provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned method embodiments.

[0083] This application provides a computer program product including a computer program that, when executed by a processor, implements the methods described in any of the above embodiments.

[0084] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0085] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. An environmental service management method characterized by, The method comprises the following steps: Collect multi-dimensional environmental data, and determine the basic features, space-time features and derived features corresponding to the multi-dimensional environmental data; Determine the layer weights of each functional layer in the preset machine learning model based on the basic features, space-time features and derived features; Optimize the initial loss function used by the preset machine learning model in the training stage based on the layer weights corresponding to the preset machine learning model, and obtain a target loss function; Train the preset machine learning model based on the target loss function, obtain a target machine learning model, import the multi-dimensional environmental data into the target machine learning model, and obtain an environmental evaluation report corresponding to the multi-dimensional environmental data; Obtain the environmental protection demand provided by a user, and generate an environmental protection service feedback scheme based on the environmental protection demand and the environmental evaluation report.

2. The environmental service management method of claim 1, wherein The method comprises the following steps: Quantitatively evaluate the basic features, space-time features and derived features, and determine the respective feature contribution degrees; Determine the examination features corresponding to each functional layer in the preset machine learning model and the examination attention degrees corresponding to each examination feature based on a preset layer examination mapping relationship; Calculate the feature contribution degrees of the examination features corresponding to each functional layer and the examination attention degrees corresponding to each examination feature, and obtain the layer weights corresponding to each functional layer.

3. The environmental service management method of claim 1, wherein The method comprises the following steps: Calculate the square sum of each layer weight as a weight penalty term; Determine an optimization coefficient term based on the derived features; Obtain the target loss function based on the weight penalty term, the optimization coefficient term, a preset regularization penalty coefficient, the initial loss function used by the preset machine learning model in the training stage, and a preset loss function calculation formula; The preset loss function calculation formula is as follows: ； wherein, is the target loss function; an initial loss function used by the preset machine learning model in a training phase; λ is the preset regularization penalty coefficient; k is the optimization coefficient term. weight penalty term for all layer weights, is the ith layer weight.

4. The environmental service management method of claim 3, wherein, The method comprises the following steps: Determine an environmental adaptation coefficient based on the multi-dimensional environmental data; Obtain the feature contribution degree corresponding to the derived features, and determine an optimization weight based on the feature contribution degree and the environmental adaptation coefficient; Identify the interaction feature group and the environmental interference feature corresponding to the derived features, the interaction feature group contains at least two interaction features, and the at least two interaction features are basic features and / or space-time features; Determine the mutual information value corresponding to the interaction feature group based on a preset mutual information method, determine the partial correlation coefficient corresponding to the environmental interference feature based on a preset partial correlation correction algorithm, and determine the feature coupling degree corresponding to the derived features based on the mutual information value and the partial correlation coefficient; Determine the optimization coefficient term based on the optimization weight and the feature coupling degree.

5. The environmental service management method of claim 1, wherein, The method comprises the following steps: identifying a dominant demand dimension parameter from the environmental protection demand, and determining a recessive demand dimension parameter based on the environmental protection demand, the multi-dimensional environmental data, and the environmental assessment report; determining a demand priority vector corresponding to the environmental protection demand and the environmental assessment report based on the dominant demand dimension parameter and the recessive demand dimension parameter; determining an environmental problem label matrix based on the environmental assessment report, and determining an environmental problem structured vector based on the environmental problem label matrix, the environmental problem structured vector containing a problem label, an impact score, and an improvement potential value; determining a matching degree between each problem label and a demand index dimension based on the demand priority vector and the environmental problem structured vector; determining a target problem label based on a preset matching degree threshold and the matching degree corresponding to each problem label, obtaining a target feedback template corresponding to the target problem label from a feedback template library based on the target problem label, and generating an environmental protection service feedback scheme based on the target feedback template and the target problem label.

6. The environmental services management method of claim 5, wherein, The determining of the demand priority vector corresponding to the environmental protection demand and the environmental assessment report based on the dominant demand dimension parameter and the recessive demand dimension parameter comprises: quantifying each dominant demand dimension parameter into a dominant index value, and quantifying each recessive demand dimension parameter into a recessive index value; identifying an environmental feature parameter from the multi-dimensional environmental data; matching the environmental feature parameter with each dominant demand dimension parameter to obtain a first demand matching value corresponding to each dominant demand dimension parameter, and determining a dominant priority vector based on the dominant index value and the first demand matching value corresponding to each dominant demand dimension parameter; matching the environmental feature parameter with each recessive demand dimension parameter to obtain a second demand matching value corresponding to each recessive demand dimension parameter, and determining a recessive priority vector based on the recessive index value and the second demand matching value corresponding to each recessive demand dimension parameter; determining the demand priority vector corresponding to the environmental protection demand and the environmental assessment report based on the dominant priority vector and the recessive priority vector.

7. A management system characterized by comprising: The management system comprises: at least one processor; a memory; at least one application program, wherein the at least one application program is stored in the memory and configured to be executed by the at least one processor, and the at least one application program is configured to execute the environmental protection service management method in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, comprising: a computer program stored in the memory and capable of being loaded and executed by the processor to implement the environmental protection service management method in any one of claims 1-6.

9. A computer program product, characterised in that, comprising a computer program, which, when executed by the processor, implements the steps of the environmental protection service management method in any one of claims 1-6.

Images