Ecological environment health intelligent assessment and management system based on PSR and entropy weight dynamic self-adaption

Abstract

The invention discloses an ecological environment health intelligent assessment and management system based on PSR and entropy weight dynamic self-adaption, and belongs to the technical field of ecological environment monitoring and intelligent management. The system comprises a data acquisition module, a preprocessing module, a classification and dynamic empowerment module, an intelligent prediction module, a feedback optimization module and a data security module. The system uses a sensor and edge calculation to acquire and process data, and the EQR is calculated through preprocessing. The core lies in that based on PSR model classification indexes, index weights are determined by applying an entropy weight method in combination with a dynamic adjustment mechanism, and pressure, state and response factors are generated. The intelligent prediction module adopts a deep learning model, outputs a health comprehensive index (Jzzs) based on factors, and performs adaptive optimization. And the feedback optimization module compares the Jzzs with a dynamic threshold value, and generates a graded, quantized and optimized treatment scheme in combination with machine learning to realize closed-loop management. And the data security module ensures that the data cannot be tampered and can be traced by using a block chain. According to the invention, dynamic accurate evaluation and intelligent adaptive management of ecological environment health are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of ecological environment monitoring and intelligent management, and particularly relates to an ecological environment health intelligent evaluation and management system and method based on the combination of the Pressure-State-Response (PSR) model and entropy weight dynamic adaptation. Background Art

[0002] The health status of the marine ecological environment, especially in ecologically sensitive areas such as bays and estuaries, is crucial for maintaining biodiversity and sustainable development. However, these areas currently generally face ecological degradation problems caused by multiple pressures. Therefore, achieving dynamic and accurate assessment of ecological environment health and scientific management has become an urgent need in the field of environmental protection.

[0003] Existing technologies have obvious deficiencies in meeting this demand, mainly reflected in:

[0004] In terms of data acquisition and processing: Traditional monitoring means have problems such as one-sided monitoring scope, single indicators (mostly physical and chemical indicators), and poor real-time performance. Moreover, the standardization (such as the application of the ecological quality ratio EQR) and effective integration of multi-source heterogeneous data are difficult, making it difficult to comprehensively reflect the complex conditions of the ecosystem. The application of real-time processing technologies such as edge computing is insufficient.

[0005] In terms of evaluation models and methods: Existing evaluation models often adopt subjective (such as expert scoring, AHP) or static weight setting methods, which cannot objectively reflect the dynamic changes in the relative importance of indicators and mostly ignore the non-linear relationships within the ecosystem and the PSR (Pressure-State-Response) logic chain. Some complex models (such as certain machine learning models) lack interpretability.

[0006] In terms of prediction, early warning and decision support: Existing systems focus on post-event evaluation and lack the ability to conduct forward-looking prediction and early warning using advanced models (such as deep learning models like LSTM that handle time series dependencies), resulting in management often being in a passive response state. The model's ability to adapt to environmental changes (such as the application of online learning, federated learning) is insufficient.

[0007] In terms of the linkage between evaluation and governance: Evaluation results are often macroscopic and difficult to directly guide specific governance actions. There is a lack of the ability to integrate historical governance data and use means such as machine learning to optimize and recommend graded quantitative governance plans, and it is difficult to form an effective closed-loop feedback on governance effects to continuously improve management strategies.

[0008] In terms of data security and credibility: Traditional data management faces problems such as easy data tampering and difficult traceability, and there are challenges in data security and privacy protection in multi-party collaboration and model sharing. The application of technologies such as blockchain to ensure data credibility is insufficient.

[0009] In summary, the existing technologies have significant limitations in aspects such as comprehensive real-time data, dynamic objectivity of evaluation models, forward-looking prediction and warning, linkage optimization of evaluation and governance, and data security and credibility. There is an urgent need for a new paradigm of intelligent evaluation and management that integrates technologies such as the Internet of Things, edge computing, big data, artificial intelligence (deep learning, machine learning), PSR model, dynamic weighting methods (such as entropy weight method), and blockchain to overcome the bottlenecks of the existing technologies. Summary of the Invention

[0010] Object of the Invention: The present invention aims to overcome the above limitations existing in the ecological environment health evaluation and management of the existing technologies, and provides an ecological environment health intelligent evaluation and management system and method based on the PSR model and entropy weight dynamic adaptation. The system and method can realize dynamic, accurate, and intelligent evaluation of the health status of the ecological environment (especially ecologically sensitive sea areas), provide forward-looking prediction and warning, optimization, and closed-loop governance decision support, and use blockchain technology to ensure the security and credibility of data.

[0011] Technical Solution: To achieve the above object, the present invention provides an ecological environment health intelligent evaluation and management system based on PSR and entropy weight dynamic adaptation. The system is built on a hardware architecture including a processor and a memory, and instructions executable by the processor are stored in the memory to realize the collaborative work of the following modules:

[0012] Data Acquisition Module: Use the sensor network deployed in the ecologically sensitive sea area (monitoring at least indicators such as total nitrogen, total phosphorus, heavy metals, COD, dissolved inorganic nitrogen, chlorophyll a, density of planktonic animals and plants, density of benthic animals, density of fish eggs and larvae, sulfide content, etc.) to obtain water quality environment data in real time. Combine with the edge computing nodes deployed locally for preliminary data processing (such as cleaning, standardization, anomaly detection), and transmit the processed data through the Internet of Things communication interface. This ensures the real-time nature and preliminary quality of the data and reduces the burden on the central server.

[0013] Data Preprocessing Module: Receive the preliminarily processed data, perform further data cleaning (such as using multiple linear regression or time series prediction interpolation to process missing / anomaly values) and standardization processing. The core is to configure the EQR calculation logic, and calculate the EQR of each monitoring indicator according to the reference standards determined by the reference station method, etc., which distinguish cost-type (EQR = reference value / monitoring value) and benefit-type (EQR = monitoring value / reference value) indicators, and generate a dimensionless ecological quality indicator set.

[0014] Data Classification and Dynamic Weighting Module:

[0015] PSR Classification Sub-module: Based on the PSR model rules, map the EQR indicator set to land source pressure class (P), water quality status class (S), and biological response class (R) indicators.

[0016] Entropy weight calculation sub-module: Apply the entropy weight method processing logic. Based on the data distribution characteristics of each classification index within the historical and current monitoring periods (realized through standardization, calculation of proportions, and information entropy), objectively calculate the initial weights of each index. The smaller the information entropy, the greater the degree of variation of the index, the more information, and the higher the initial weight.

[0017] Weight dynamic adjustment sub-module: Set up a weight dynamic adjustment mechanism. This mechanism monitors preset trigger conditions, such as: the change in the statistical characteristics (such as variance) of the monitored data within the sliding window exceeds the threshold; a specific environmental warning event occurs (such as a red tide); or a management target change instruction is received. When the conditions are met, trigger the recalculation of weights (run the entropy weight method again) or adjust according to preset rules (such as temporarily increasing the weights of specific indicators), and output dynamic adaptive index weights that can reflect the current environmental focus and management requirements.

[0018] Factor generation sub-module: According to the dynamically adaptive index weights and the corresponding EQR values, calculate and generate the land source pressure factor (Lyyz), water quality status factor (Szyz), and biological response factor (Sxyz) respectively through a weighted algorithm.

[0019] Health status intelligent prediction module:

[0020] Health prediction model based on deep learning: Adopt a specific network structure suitable for processing time series and multi-factor coupling relationships (such as a hybrid model containing a CNN layer to extract spatial features and an LSTM layer to handle time dependencies). Use the Lyyz, Szyz, and Sxyz factors as inputs, train and fit to output the comprehensive ecological environment health index (Jzzs), and optionally predict the future short-term change trend of Jzzs. Jzzs is obtained through top-level weighted calculation: Jzzs = α × Lyyz + β × Szyz + γ × Sxyz, and the top-level weights α, β, γ are set based on expert knowledge and can be adjusted according to long-term management strategies.

[0021] Model adaptive optimization sub-module: Implement an online learning mechanism, continuously fine-tune the model using new data; or trigger periodic retraining according to the performance monitoring results. When deployed on multiple nodes, adopt a federated learning mechanism. Each node uses local data for training and uploads the parameter update values to the central node for aggregation. The central node distributes the aggregated parameters to update the models of each node, realizing collaborative optimization under privacy protection and improving the global performance and generalization ability of the model.

[0022] Feedback and optimization governance module:

[0023] Receive Jzzs, compare it with the dynamically set multi-level warning thresholds (set based on historical data analysis, seasonal patterns, and management objectives, covering levels from "excellent" to "abnormal"), determine the status level, and trigger a warning.

[0024] Integrated intelligent decision-making sub-module: Use machine learning models (trained based on the historical governance measures and their effect databases) to evaluate the potential impact of different governance measures on Jzzs. Accordingly, generate a hierarchical (corresponding to different warning levels), quantified (specific measures and parameters), and optimized recommended (considering effects and cost-benefit) governance plan. For example, recommend emergency measures for the "abnormal / poor" level, long-term improvement measures for the "general" level, and suggest maintenance for the "good / excellent" level.

[0025] Achieve closed-loop management: The governance effect is fed back to the system through subsequent monitoring, and is used to update the database, optimize the model, and make future decisions.

[0026] Data security guarantee module: Adopt blockchain technology. Perform hash calculations on key data (such as the collected environmental data, the output Jzzs, and the generated decision-making information), and record the hash values and metadata in the distributed ledger. Utilize the immutable and traceable characteristics of the blockchain to provide technical guarantees for data authenticity and integrity.

[0027] The present invention also provides a corresponding intelligent evaluation and management method for ecological environment health based on PSR and entropy weight dynamic adaptation. This method is applied to the above system and includes steps such as data collection, data preprocessing, data classification and dynamic weighting, intelligent prediction of health status, feedback and optimized governance, and data security guarantee. Its core technical features correspond one-to-one with the functions of the system modules. Among them, when generating an optimized plan in the feedback and optimized governance step, it may further include sub-steps such as maintaining the historical governance database, training the governance effect prediction model, and using the model to evaluate and recommend the optimal / efficient plan.

[0028] Beneficial effects: Compared with the prior art, the system and method provided by the present invention have the following remarkable beneficial effects:

[0029] The evaluation is more comprehensive, objective, and dynamic: Combining the classification indicators of the PSR model conforms to ecological logic; using the entropy weight method for objective weighting and introducing a dynamic weight adjustment mechanism enables the evaluation system to adapt to environmental changes and management requirements, overcomes the static and subjective defects of traditional methods, and improves the accuracy, timeliness, and scientific nature of the evaluation results.

[0030] Predictions are smarter and more forward-looking: By using deep learning models (such as CNN-LSTM hybrid models, combined with Attention mechanisms, etc.) to process time-series data and multi-factor coupling relationships, it is not only possible to accurately evaluate the current health status (Jzzs), but also predict future trends, providing technical support for proactive management and early warning. The model's adaptive optimization ability (online learning / retraining) and the application of the federated learning mechanism further improve the accuracy and adaptability of predictions, and solve the privacy protection and collaborative optimization problems in multi-node deployment.

[0031] Governance is more precise, optimized, and efficient: Based on dynamic warning thresholds and Jzzs level judgments, hierarchical responses are realized; using machine learning models to analyze historical governance effects provides data-driven decision-making support for recommending specific, quantitative, and optimized (considering effects and costs) governance solutions, significantly improving the pertinence, effectiveness, and efficiency of governance measures.

[0032] Management achieves a closed-loop and continuous improvement: By closely integrating the links of monitoring, evaluation, prediction, early warning, decision-making, and response, the governance effect is fed back to the system through monitoring data, continuously optimizing the model, thresholds, and strategies, forming a closed-loop management process of continuous improvement, and improving the overall management efficiency and scientific level.

[0033] Data is more secure and trustworthy: Introducing edge computing improves the real-time data processing and reduces the transmission pressure; using blockchain technology to perform hash evidence storage on key data (from original monitoring to final decision-making) ensures the immutability, integrity, and traceability of data, greatly enhancing the credibility of data, and providing reliable technical support for scientific research, regulatory auditing, multi-party collaboration, and liability determination.

[0034] In summary, through the deep integration and innovative integration of multiple advanced technologies such as the Internet of Things, edge computing, PSR model, dynamic entropy weight method, deep learning, machine learning, and blockchain, the present invention constructs a new paradigm for intelligent, precise, dynamic, and systematic ecological environment health assessment and management, which can significantly improve the scientific decision-making level and actual effectiveness of ecological environment protection and sustainable management. Brief Description of the Drawings

[0035] Figure 1 It is a block diagram of an ecological environment health intelligent assessment and management system based on PSR and dynamic self-adaptation of entropy weight according to the present invention. Detailed Embodiment

[0036] Next, it will be combined with the drawings (for example, Figure 1As shown in the system block diagram), the specific implementation of the ecological environment health intelligent evaluation and management system and method provided by the present invention based on PSR and entropy weight dynamic adaptation will be clearly and completely described. It should be noted that the implementation described here is only one possible implementation of the present invention, not all. Based on the core idea and technical solution disclosed by the present invention, all other implementations obtained by those of ordinary skill in the art without creative work shall fall within the scope of protection of the present invention.

[0037] Example 1: System architecture and module functions.

[0038] Refer to Figure 1 , this embodiment provides an ecological environment health intelligent evaluation and management system based on PSR and entropy weight dynamic adaptation. The system is built on a server (including a processor and a memory) and edge computing nodes deployed on-site, aiming to intelligently evaluate and manage ecologically sensitive sea areas. The system mainly includes the following functional modules:

[0039] Data acquisition module.

[0040] Core function: Continuously monitor the environmental conditions of the target sea area in real time, obtain multi-dimensional water quality environmental data, and transmit it after preliminary processing.

[0041] Sensor network: Deploy a sensor group including indicators such as total nitrogen (TN), total phosphorus (TP), heavy metals (such as Cd, Hg), COD, dissolved inorganic nitrogen (DIN), chlorophyll a (Chl-a), phytoplankton density, zooplankton density, benthic animal density, fish egg and larva density, and sulfide content. The sensor types can include optical sensors, electrochemical sensors, biosensors, etc., and are combined with automatic samplers and on-line analyzers.

[0042] Edge computing nodes: Deployed near the monitoring points to perform local preprocessing on the original data, such as:

[0043] Data cleaning: Filter out obvious noise (such as using moving average filtering).

[0044] Data standardization: Initially unify the data format.

[0045] Anomaly detection: Identify data points that exceed the threshold or mutate (such as using the 3σ principle).

[0046] Data integrity check: Ensure the complete transmission of data packets.

[0047] Data transmission: The processed data or features are transmitted to the data preprocessing module through the Internet of Things (such as NB-IoT, 4G / 5G).

[0048] Data preprocessing module.

[0049] Core functions: Deeply clean, standardize, and transform the received data, calculate the EQR, and generate a structured ecological quality indicator set.

[0050] Missing value / outlier handling: Adopt a more refined method for handling, such as:

[0051] Missing values: Use multiple linear regression interpolation (if the variables are strongly correlated) or LSTM-based time series prediction interpolation.

[0052] Outliers: Eliminate after confirmation or replace using robust statistical methods (such as the median).

[0053] EQR calculation.

[0054] Determine the reference standard: Adopt the reference station method, and select the mean or specific percentile of the long-term monitoring data in the least disturbed area as the reference standard value.

[0055] Distinguish the indicator types: Clearly define each indicator as a cost type (such as TN, TP, COD, heavy metals, sulfide) or a benefit type (such as dissolved oxygen, phytoplankton and zooplankton density, benthic animal density, fish egg and larva density, and chlorophyll a may be a benefit type within an appropriate range).

[0056] Calculate the EQR: Calculate according to the formula, for cost type EQR = reference value / monitoring value; for benefit type EQR = monitoring value / reference value.

[0057] Generate the indicator set: Aggregate the EQR values of all indicators to form a standardized ecological quality indicator set.

[0058] Data classification and dynamic weighting module.

[0059] Core functions: Classify EQR indicators based on the PSR model; calculate the initial weights using the entropy weight method and adjust them dynamically; calculate and generate pressure, state, and response factors.

[0060] PSR classification sub-module: Classify EQR indicators:

[0061] Pressure (P): Such as the EQR of the land-based TN / TP input volume, the EQR of heavy metal emissions.

[0062] State (S): Such as the EQR of water body DIN, the EQR of Chl-a, the EQR of COD, the EQR of dissolved oxygen.

[0063] Response (R): Such as the EQR of phytoplankton and zooplankton density, the EQR of benthic animal density, the EQR of fish egg and larva density, the EQR of biodiversity index.

[0064] Entropy weight calculation sub-module (calculate the initial weights).

[0065] Data standardization: Standardize the historical and current EQR data (such as min-max standardization) to ensure direction consistency (the larger the value, the better).

[0066] Construct a matrix: Construct an m×n standardized matrix (m time periods, n indicators) according to the PSR categories.

[0067] Calculate the proportion P_ijk.

[0068] Calculate the information entropy E_i = - (1 / ln(m)) * Σ (P_ijk * ln(P_ijk)).

[0069] Calculate the initial weight W_i = (1 - E_i) / Σ (1 - E_j). Calculate for the three types of indicators P, S, and R respectively.

[0070] Weight dynamic adjustment sub-module.

[0071] Trigger condition monitoring.

[0072] Statistical characteristic change: Monitor whether the change rate of the variance of a certain indicator EQR within a sliding window (such as the past 30 days) exceeds a threshold (such as 20%).

[0073] Specific event: The system detects a red tide (abnormal increase in Chl-a) or receives a report of an oil spill event.

[0074] Management instruction: Receive the instruction of "focus on controlling total phosphorus in the next month".

[0075] Weight adjustment execution:

[0076] If the variance change exceeds the threshold, trigger the recalculation of the weights of this indicator and its similar indicators (run the entropy weight method again).

[0077] If a red tide occurs, temporarily increase the weight of the Chl-a indicator (multiply by an adjustment factor such as 1.5), and normalize the weights of other similar indicators.

[0078] If the total phosphorus control instruction is received, increase the weights of TP-related indicators in proportion (such as increasing by 30%) and normalize them.

[0079] Factor generation sub-module: Calculate the comprehensive factor:

[0080] Lyyz = Σ (W_pi * EQR_pi); Szyz = Σ (W_sj * EQR_sj); Sxyz = Σ (W_rk * EQR_rk) (W is the dynamic adaptive weight).

[0081] Health status intelligent prediction module.

[0082] Core function: Utilize a deep learning model to evaluate / predict Jzzs based on factors and possess the ability of adaptive optimization.

[0083] Deep learning model: Preferably a model that combines LSTM and the Attention mechanism.

[0084] Input: Time series of Lyyz, Szyz, and Sxyz in the past N time steps (e.g., N = 30 days).

[0085] Structure: Multiple layers of LSTM capture temporal dependencies, the Attention layer learns the dynamic attention to historical information and factors, and the fully connected layer outputs the predicted value of Jzzs.

[0086] Calculation of Jzzs: The model directly outputs Jzzs, or calculates Jzzs = α×Lyyznorm + β×Szyznorm + γ×Sxyznorm after outputting the normalization factors. The top - layer weights α, β, γ are set by experts (e.g., α = 0.3, β = 0.4, γ = 0.3) and can be adjusted according to management strategies (e.g., increasing α during the pollution control phase).

[0087] Model adaptive optimization.

[0088] Online learning: Use new data points for mini - batch gradient descent to fine - tune the model.

[0089] Periodic retraining: Retrain the model monthly or when the prediction error (e.g., MAE) exceeds the threshold using the cumulative data.

[0090] Federated learning (when deployed on multiple nodes):

[0091] The central server initializes the global model W_global and distributes it.

[0092] Each node k trains with local data D_k and calculates the update amount ΔW_k.

[0093] Each node uploads the encrypted ΔW_k.

[0094] The central server securely aggregates ΔW_aggregated = Σ (n_k / N) * ΔW_k.

[0095] The central server updates W_global_new = W_global + η * ΔW_aggregated.

[0096] Repeat steps 2 - 5 until convergence.

[0097] Feedback and optimization governance module.

[0098] Core functions: Determine the health level, give early warnings, generate optimized treatment plans, and achieve closed-loop control.

[0099] Dynamic early warning thresholds: Based on the percentiles (such as 90%, 75%, 50%, 25%) of historical Jzzs (e.g., in the recent 3 years), combined with seasonal adjustment factors and management objectives (such as water quality standards), set the dynamic threshold ranges for the five levels of "excellent", "good", "average", "poor", and "abnormal".

[0100] Level determination and early warning: Compare real-time / predicted Jzzs with the thresholds, determine the level, and trigger corresponding early warnings.

[0101] Intelligent decision-making sub-module.

[0102] Maintain the historical database: Record treatment measures (such as emission reduction ratios, types of restoration projects, types and dosages of chemicals added), environmental factors at the time of implementation (Lyyz, Szyz, Sxyz, Jzzs), and subsequent changes in Jzzs.

[0103] Train the effect prediction model: Use this database to train machine learning models (such as random forest regression or reinforcement learning Q-learning) to predict the effects of measures (such as ΔJzzs).

[0104] Generate optimized treatment plans.

[0105] Evaluate candidate measures: For possible measures (such as different levels of emission reduction, different restoration plans), use the model to predict their effects and estimated costs.

[0106] Optimized recommendation: Recommend the plan with the best expected effect or the highest cost-effectiveness.

[0107] Hierarchical quantification.

[0108] Abnormal / Poor: Recommend emergency measures, such as "Immediately activate the emergency plan for reducing total nitrogen by 50% at Outfall A, and at the same time, add Chemical X for phosphorus removal at Y kg per square kilometer in Area B".

[0109] Average: Recommend long-term measures, such as "It is recommended to plan and construct an artificial wetland in Area C, which is expected to improve Jzzs by 0.1 within 3 years; adjust the aquaculture density in Area D to reduce it by 20%".

[0110] Good / Excellent: Suggest "Maintain the current management measures and strengthen the monitoring of Risk Point E".

[0111] Closed-loop feedback: After the treatment is implemented, the monitoring data is continuously input into the system to update the database, which is used for model verification and optimization of future decisions.

[0112] Data security guarantee module.

[0113] Core function: Utilize blockchain to ensure the immutability and traceability of key data.

[0114] Implementation: Adopt Hyperledger Fabric or a similar consortium blockchain platform.

[0115] Data on the chain: Hash of key original monitoring data packets, hash of EQR and factor values, hash of Jzzs evaluation / prediction results, hash of early warning and governance decision records.

[0116] Process: Calculate the SHA-256 hash value for the selected data, and write it into the blockchain ledger through a smart contract together with metadata such as timestamps and source IDs.

[0117] Effect: Ensure the integrity, authenticity, and traceability of the entire process from data collection to decision-making, and enhance the credibility.

[0118] Example 2: Management method process.

[0119] The present invention also provides an ecological environment health intelligent evaluation and management method based on PSR and entropy weight dynamic adaptation corresponding to the above system. The execution process of this method corresponds to the functions of each module of the system, and mainly includes the following steps.

[0120] Data collection step: Through the sensor network and edge computing, real-time monitor and preliminarily process multi-dimensional data such as the water quality environment of the target ecological area.

[0121] Data preprocessing step: Clean, standardize the data, and calculate the EQR values of each index to generate an ecological quality index set.

[0122] Data classification and dynamic weighting step: Apply the PSR model to classify indicators; use the entropy weight method to calculate the initial weights, and update the weights through a dynamic adjustment mechanism; combine the dynamic weights and EQR to calculate the Lyyz, Szyz, and Sxyz factors.

[0123] Health status intelligent prediction step: Utilize a deep learning model with adaptive optimization ability to fit and output Jzzs based on the Lyyz, Szyz, and Sxyz factors, and optionally predict its trend.

[0124] Feedback and optimized governance step: Compare Jzzs with the dynamic early warning threshold for grading; combine historical data and machine learning models to predict the governance effect; generate a graded, quantified, and optimized governance plan.

[0125] Among them, when generating an optimized plan in the feedback and optimized governance step, it can be further refined into: maintaining a historical governance database; training a governance effect prediction model; using the model to evaluate candidate measures and recommend the optimal / effective plan.

[0126] Example 3: Detailed steps for calculating initial weights using the entropy weight method.

[0127] In the data classification and dynamic weighting module described in Example 1, for calculating the weight values of each index involved in the formulas of the land source pressure factor (Lyyz), water quality status factor (Szyz), and biological response factor (Sxyz) (for example, the weight value W_pi of the i-th land source pressure type index, the weight value W_sj of the j-th water quality status type index, and the weight value W_rk of the g-th biological response type index), the determination of its initial or basic value is preferably based on the Entropy Weight Method. The entropy weight method is an objective weighting method that determines weights according to the degree of variation of index data (i.e., the magnitude of information entropy). The index with a greater degree of variation (smaller information entropy) is considered to contain more information and should be given a higher weight. The specific calculation steps are as follows.

[0128] Data standardization preprocessing: Select the EQR index data set that has been processed by the data preprocessing module in Example 1 during the historical and current monitoring periods for calculating weights.

[0129] Standardize the index data (EQR values) belonging to each PSR classification (land source pressure type, water quality status type, biological response type) to eliminate the influence of dimension and unify the index direction.

[0130] This standardization process needs to be combined with the cost / benefit type attribute of the index (this attribute has been determined in the data preprocessing module of Example 1):

[0131] For cost-type indicators (the larger the original value or EQR value, the worse the state), use the positive standardization method to make the converted value larger, indicating a better "optimal" state (for example, through the formula Y = (X_max - X) / (X_max - X_min) or a similar method).

[0132] For benefit-type indicators (the larger the original value or EQR value, the better the state), use the reverse or original direction retention standardization method to make the converted value also larger, indicating a better "optimal" state (for example, through the formula Y = (X - X_min) / (X_max - X_min) or a similar method).

[0133] The goal of standardization is to ensure that after all indicators are processed, the meaning of the numerical magnitude is unified (for example, all are unified to the larger the value, the closer to the ideal state), which is convenient for subsequent calculation of information entropy and weights. Let Y_ik represent the value of the i-th index after the above standardization processing in the k-th monitoring period (k = 1, 2,..., K, where K is the total number of periods).

[0134] Calculate the proportion of index values: For each index \(i\), calculate the proportion (weight) \(P_{ik}\) of the standardized value \(Y_{ik}\) of the \(k\)-th period in all \(K\) monitoring periods to the total sum of the standardized values of this index in all periods: \(P_{ik}=Y_{ik} / \sum_{k = 1}^{K}Y_{ik}\). Here, it is necessary to ensure that the denominator \(\sum_{k = 1}^{K}Y_{ik}\) is not zero.

[0135] Calculate the information entropy: According to the definition of information entropy, calculate the information entropy \(E_i\) of the \(i\)-th index: \(E_i=-k\times\sum_{k = 1}^{K}(P_{ik}\times\ln(P_{ik}))\).

[0136] Among them, \(k\) is a constant, usually taking \(k = 1 / \ln(K)\) to ensure that the value of the information entropy \(E_i\) falls within the interval \([0, 1]\). \(K\) is the total number of historical monitoring periods (or samples) used to calculate the weights.

[0137] It is stipulated that when \(P_{ik}=0\), \(P_{ik}\times\ln(P_{ik}) = 0\) to avoid meaningless logarithmic calculations.

[0138] The information entropy \(E_i\) measures the degree of chaos or uncertainty in the data distribution of the \(i\)-th index. The smaller \(E_i\) is, the greater the difference in the data values of this index in different periods and the more information it provides.

[0139] Calculate the index weights: Calculate the entropy weight (i.e., weight value) \(W_i\) of the \(i\)-th index: \(W_i=(1 - E_i) / \sum_{j = 1}^{n}(1 - E_j)\). Among them, \((1 - E_i)\) represents the difference or information utility value of the \(i\)-th index. The smaller the information entropy \(E_i\), the larger this value. \(n\) is the total number of indexes in this PSR category. The denominator is the total sum of the information utility values of all indexes, which is used for normalization to ensure that the sum of the weights of all indexes is 1 (\(\sum_{i = 1}^{n}W_i = 1\)).

[0140] The larger the weight \(W_i\) is, the more important the \(i\)-th index is in the comprehensive evaluation.

[0141] Apply to each PSR category respectively: The above entropy weight method calculation process needs to be independently applied to the three index sets of land-based pressure category, water quality status category, and biological response category respectively, so as to obtain the weight values of each index within their respective categories (i.e., \(W_{pi}\), \(W_{sj}\), \(W_{rk}\)).

[0142] It should be emphasized that the weight values (W_pi, W_sj, W_rk) calculated by the entropy weight method are the initial weights or basic weights objectively determined based on historical and current data distributions. In the system described in Embodiment 1, these weights are the basis for the dynamic adaptive weights. They will be further affected by the weight dynamic adjustment mechanism (triggered based on changes in data statistical characteristics, specific environmental events, or changes in management objectives), and are adjusted or recalculated in real time to finally obtain the dynamic adaptive weights used to calculate the factors Lyyz, Szyz, and Sxyz. This way of combining objective weight assignment with dynamic adjustment makes the weight assignment have data basis and can flexibly adapt to environmental changes and management requirements.

[0143] This embodiment details a specific implementation manner of the system and method of the present invention. Those skilled in the art can understand that, without departing from the core principle and spirit of the present invention, various changes, modifications, substitutions, and variations can be made to the specific details in the above embodiments, such as using different sensor types, interpolation methods, deep learning model structures, machine learning algorithms, blockchain platforms, etc. These changes should all fall within the scope of protection required by the present invention.

Claims

1. An intelligent ecological environment health assessment and management system based on PSR and entropy weight dynamic adaptation, characterized by: The system is built on a hardware architecture including a processor and a memory, wherein the memory stores instructions executable by the processor to achieve the collaborative work of the following modules: Data collection module, used to obtain ecological environment data; A data preprocessing module, used to preprocess the ecological environment data, calculate the ecological quality ratio index EQR, and generate an ecological quality indicator set; A data classification and dynamic weighting module is used to classify the ecological quality indicator set based on the PSR model, apply the entropy weight method and combine it with the dynamic adjustment mechanism to calculate the dynamic adaptive indicator weight, and calculate and generate the pressure factor, state factor and response factor based on the dynamic adaptive indicator weight and the corresponding EQR value; A health status intelligent prediction module, which is used to output the comprehensive ecological environment health index Jzzs based on the pressure factor, state factor and response factor using a prediction model with adaptive optimization capability; A feedback and optimization management module is used to generate an optimization management plan based on the comparison result between the comprehensive health index Jzzs and the warning threshold to achieve closed-loop management; The data security assurance module is used to ensure that data cannot be tampered with and is traceable.

2. According to claim 1, an ecological environment health intelligent assessment and management system based on PSR and entropy weight dynamic self-adaptation is characterized by: The data acquisition module is configured with a sensor network deployed in ecologically sensitive sea areas, and the sensor network includes sensors for monitoring at least one indicator of total nitrogen, total phosphorus, heavy metals, chemical oxygen demand, dissolved inorganic nitrogen, chlorophyll a, phytoplankton density, benthic animal density, fish egg and larvae density, and sulfide content to obtain real-time water quality environment data; the data acquisition module also includes an edge computing node deployed locally, the edge computing node performs preliminary data processing operations, and transmits the processed data through the Internet of Things communication interface. The edge computing node of the data acquisition module performs local preliminary processing on the original water quality environment data collected by the sensor, including at least one of data cleaning, standardization and anomaly detection, and sends the preprocessed data or extracted feature data to the data preprocessing module.

3. According to claim 1, an ecological environment health intelligent assessment and management system based on PSR and entropy weight dynamic self-adaptation is characterized by: The data preprocessing module is configured to receive the data after the preliminary processing, perform data cleaning and standardization processing, and is configured with an ecological quality ratio index EQR calculation logic, calculate the EQR of each monitoring indicator according to a preset reference standard that distinguishes between cost-type and benefit-type indicators, and generate a structured ecological quality indicator set. The data preprocessing module includes: Missing values ​​and outlier processing logic are filled using multivariate linear regression interpolation or interpolation methods based on time series prediction; The EQR calculation logic uses the reference position method to determine the reference standard, distinguishes between cost-based indicators and benefit-based indicators, and uses the following formula to calculate EQR: Cost-based indicator EQR = reference standard value / actual monitoring value, Benefit indicator EQR = actual monitoring value / reference standard value.

4. According to claim 1, an ecological environment health intelligent assessment and management system based on PSR and entropy weight dynamic self-adaptation is characterized by: Data classification and dynamic weighting module, including: A PSR classification submodule is configured to map the ecological quality indicator set into terrestrial pressure indicators, water quality state indicators, and biological response indicators based on a preset pressure-state-response PSR model rule; The entropy weight calculation submodule is configured to apply the entropy weight method processing logic to standardize the PSR classification indicator data in the historical and current monitoring periods, construct a standardized indicator matrix, calculate the proportion of each indicator and information entropy, and thus calculate the initial weight based on the data distribution characteristics of each classification indicator in the historical and current monitoring periods; The weight dynamic adjustment submodule is configured with a weight dynamic adjustment mechanism, which monitors preset trigger conditions, which at least include one or more of the changes in the statistical characteristics of the monitoring data reaching a threshold, the occurrence of a specific environmental warning event, or the change of the management target, and triggers the entropy weight calculation submodule to recalculate or adjust the weight when the conditions are met, and outputs a dynamically adaptive indicator weight; The factor generation submodule is configured to calculate and generate terrestrial pressure factors, water quality status factors and biological response factors respectively according to the dynamically adaptive indicator weights and corresponding EQR values ​​through a preset weighting algorithm. In the data classification and dynamic weighting module, the specific steps of applying the entropy weight method to calculate the initial weights include: standardizing the PSR classification indicator data in the historical and current monitoring periods, distinguishing between the positive standardization of cost-type indicators and the reverse standardization of benefit-type indicators; constructing a standardized indicator matrix; calculating the proportion of each indicator in different time periods; calculating the information entropy of each indicator; calculating the initial weight based on the information entropy; the specific triggering conditions of the dynamic weight adjustment mechanism include: when the variance change of the indicator data in the sliding window of the monitoring data exceeds the preset proportion, the weight recalculation is triggered; or when an external management instruction is received, the corresponding indicator weight is temporarily increased.

5. According to claim 1, the ecological environment health intelligent assessment and management system based on PSR and entropy weight dynamic self-adaptation is characterized by: The health status intelligent prediction module comprises: a health prediction model based on deep learning, the model adopts a specific network structure suitable for processing time series data and multi-factor coupling relationship, takes the terrestrial pressure factor, water quality state factor and biological response factor as input, trains and fits to output the comprehensive index of ecological environment health Jzzs, and optionally predicts the change trend of Jzzs in the short term in the future; the Jzzs is obtained by performing top-level weighted calculation on each factor: Jzzs=α×terrestrial pressure factor+β×water quality state factor+γ×biological response factor, wherein the top-level weights \alpha, \beta, \gamma are set based on expert knowledge and can be adjusted according to management strategies, and the deep learning model adopted by the health status intelligent prediction module is a hybrid model, which comprises a convolutional neural network CNN layer for extracting spatial correlation characteristics of water quality environment data, and a long short-term memory network LSTM layer for processing time series dependency; The model adaptive optimization submodule is configured to implement an online learning mechanism to continuously fine-tune model parameters using new monitoring data, or to trigger periodic model retraining based on the results of predictive performance monitoring. When the system is deployed on multiple monitoring nodes, a federated learning mechanism can be adopted. Each node independently trains a submodel and uploads parameter update values ​​to the central node for aggregation. The central node sends aggregated parameters to update the models of each node, thereby improving the global performance of the model while protecting data privacy.

6. The method of claim 1, wherein the method is characterized in that: The health status intelligent prediction module adopts a federated learning mechanism to train its deep learning prediction model parameters; each monitoring node independently uses local data to train a sub-model and extracts the model parameter update value and uploads it to the central node for aggregation. The central node sends the aggregated model parameters back to each monitoring node for updating, and iterates to obtain a globally optimized prediction model.

7. The method of claim 1, wherein the method is characterized in that: The feedback and optimization governance module is configured to receive the comprehensive health index Jzzs, compare it with the multi-level warning thresholds dynamically set based on the time series analysis of historical Jzzs data, seasonal variation patterns and management objectives, and determine the level of ecological environment status; the module also integrates an intelligent decision-making sub-module, which uses a machine learning model based on historical governance measures and their effect data to evaluate the impact of different governance measures on Jzzs, and based on this, generates a hierarchical, quantified and optimized recommended governance plan containing specific measures to achieve closed-loop management. In the feedback and optimization governance module, the method for dynamically setting multi-level warning thresholds includes: based on the time series analysis of historical Jzzs data, combined with seasonal variation patterns and management objectives, setting a dynamic threshold range covering "excellent", "good", "general", "poor" and "abnormal" levels.

8. The method of claim 1, wherein the method is characterized by: The intelligent decision-making submodule of the feedback and optimization governance module maintains a database of historical governance measures and their effects, and uses a machine learning model to learn the expected effects of different governance measures under specific environmental conditions, so as to evaluate the potential impact of candidate measures and recommend the best expected improvement effect or the most cost-effective solution when generating governance solutions. The hierarchical, quantitative, and optimized governance solutions include: When Jzzs is at the "abnormal" or "poor" level, a combination of emergency management measures and parameters are recommended; When Jzzs is at the "general" level, targeted long-term improvement measures are recommended, such as optimizing the layout of sewage outlets, planning ecological restoration projects, and adjusting breeding strategies; When Jzzs is at a "good" or "excellent" level, output recommendations for maintaining current management measures.

9. An intelligent ecological environment health assessment and management system based on PSR and entropy weight dynamic adaptation, characterized in that: The following steps are involved: Data collection steps: Use sensor networks to monitor water quality environmental data of the target ecological area in real time, and transmit it after preliminary processing through edge computing nodes; Data preprocessing steps: clean and standardize the received water quality environmental data, calculate the ecological quality ratio index (EQR) according to the preset reference standard, and generate an ecological quality indicator set; Data classification and dynamic weighting steps: Classify the ecological quality indicator set based on the PSR model; apply the entropy weight method to calculate the initial indicator weights, and update the weights according to the preset dynamic adjustment mechanism to obtain dynamically adaptive indicator weights; combine the dynamic adaptive weights with the corresponding EQR values ​​to calculate and generate terrestrial pressure factors, water quality status factors and biological response factors; Health status intelligent prediction step: using a deep learning prediction model with adaptive optimization capability, based on the terrestrial pressure factors, water quality status factors and biological response factors, the comprehensive index of ecological environment health Jzzs is output, and its short-term change trend is optionally predicted; Feedback and optimization governance steps: compare the comprehensive health index Jzzs with the dynamically set multi-level warning thresholds to determine the ecological status level; based on the level and optionally combined with the analysis and prediction results of historical governance measures effect data using machine learning models, generate hierarchical, quantified, and optimized governance plans to support management decisions and form closed-loop management; Data security assurance steps: Use blockchain technology to perform hash calculations on key environmental data, comprehensive health indices, and governance decision-making information and record them in a distributed ledger to ensure that the data cannot be tampered with and is traceable.

10. The steps of the ecological environment health intelligent assessment and management system based on PSR and entropy weight dynamic self-adaptation according to claim 9 are characterized by: In the feedback and optimization management step, generating an optimization management plan further includes: Maintain a database containing historical governance measures, corresponding environmental conditions, and subsequent Jzzs changes; Train a machine learning model to learn the expected effects of different combinations of governance measures under specific environmental conditions; When generating governance recommendations, the model is used to evaluate the potential impact of candidate measures and recommend the options that are expected to be most effective or efficient in raising Jzzs to the target level.

Images