System for evaluating and dynamically simulating treatment efficiency of sewage treatment plant based on 3e1s theory

Abstract

The present application relates to the technical field of intelligent water treatment, and particularly relates to a sewage treatment plant governance efficiency evaluation and dynamic simulation system based on 3E1S theory, a core theory layer is used to define the interaction rules among the environment system, the energy system, the economic system and the social system based on the 3E1S theory framework; a core technology layer is used to load multiple algorithm models and operating environments; a data layer is used to collect multi-source heterogeneous data in four dimensions of environment, energy, economy and society; a calculation layer is used to call relevant algorithm models to quantify resource waste and efficiency values based on the collected multi-source heterogeneous data, and dynamically simulate the governance path combined with a preset strategy library and a self-defined strategy, and iteratively optimize to obtain an optimal strategy package; a decision layer is used to present the coordination state among the four systems of the environment system, the energy system, the economic system and the social system through a visual early warning mode. The present application can realize the collaborative governance and dynamic optimization of the whole chain and multiple systems of the sewage treatment plant.

Description

Technical Field

[0001] This invention relates to the field of intelligent water treatment technology, and more specifically to a wastewater treatment plant treatment efficiency evaluation and dynamic simulation system based on the 3E1S theory. Background Technology

[0002] Currently, the treatment objectives of wastewater treatment plants have evolved from simply meeting water quality standards to multi-dimensional synergistic optimization encompassing water quality, energy consumption, cost, and social impact. Against this backdrop, traditional assessment techniques, lacking systematic theoretical support, struggle to analyze the dynamic coupling relationships between multiple systems, such as the cascading effects of energy consumption adjustments on water quality and cost.

[0003] The 3E1S theory integrates the environmental, energy, economic, and social systems, corresponding to water treatment effectiveness, energy efficiency, treatment costs, and social impact, respectively. Through the synergistic logic of these four systems (3E+1S), abstract multi-objective governance needs are transformed into quantifiable interaction analyses, laying a theoretical foundation for multi-dimensional assessment.

[0004] Existing technologies lack a systematic framework supported by the 3E1S theory, making it difficult to meet the needs of smart water management and refined governance. The specific technical bottlenecks are as follows:

[0005] 1. Existing models either focus on the total watershed allocation and water quality simulation, ignoring the relationship between energy and social systems; or they only build the assessment logic around a single energy efficiency dimension, failing to analyze the dynamic interaction between the four systems of environment, energy, economy and society.

[0006] 2. The existing technical indicator system suffers from a structural flaw: it emphasizes water quality but neglects synergistic effects. Key dimensions such as energy efficiency, economic costs, and social impacts are not adequately covered, and assessments of undesirable outputs like sludge moisture content and carbon emissions are also lacking. Furthermore, the indicator update cycle is too long, making it difficult to capture the real-time impact of seasonal water quality fluctuations on reagent dosage. This results in sludge moisture content at wastewater treatment plants consistently exceeding reasonable thresholds without timely intervention, significantly increasing disposal costs.

[0007] 3. Existing decision support systems are mostly limited to optimal solution selection and do not achieve dynamic simulation and closed-loop optimization; they rely to some extent on analogy analysis of benchmark objects and lack the ability to predict future scenarios.

[0008] 4. Traditional algorithms have significant drawbacks when dealing with the complex systems of wastewater treatment plants. They have a high quantification error rate and are difficult to adapt to the differences in water quality standards and processes in different regions.

[0009] Therefore, how to break through the limitations of traditional single-dimensional technologies and achieve collaborative governance and dynamic optimization of the entire chain and multiple systems of sewage treatment plants is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0010] In view of this, the present invention provides a wastewater treatment plant treatment efficiency assessment and dynamic simulation system based on the 3E1S theory, which realizes the collaborative treatment and dynamic optimization of the entire chain and multiple systems of the wastewater treatment plant.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] A wastewater treatment plant treatment efficiency assessment and dynamic simulation system based on the 3E1S theory is characterized by comprising: a core theory layer, a core technology layer, and a functional module layer; wherein the functional module layer includes a data layer, a calculation layer, and a decision layer.

[0013] The core theoretical layer is used to define the interaction rules between environmental systems, energy systems, economic systems and social systems based on the 3E1S theoretical framework.

[0014] The core technology layer is used to load various algorithm models and runtime environments;

[0015] The data layer is used to collect multi-source heterogeneous data in four dimensions: environment, energy, economy, and society.

[0016] The computing layer is used to retrieve relevant algorithm models based on the collected multi-source heterogeneous data to quantify resource waste and efficiency values, and to dynamically simulate governance paths by combining preset strategy libraries and custom strategies to perform iterative optimization and obtain the optimal strategy package.

[0017] The decision-making layer is used to present the collaborative status among the four systems—environmental, energy, economic, and social—through a visual early warning system.

[0018] Furthermore, taking the environmental system as the foundation, the energy system as the driving force, the economic system as the constraint, and the social system as the feedback, a directed weighted graph is used to represent the interaction between the four systems, and the set of nodes is represented as V={E,N,C,S}.

[0019] Among them, E represents the environmental system, which characterizes the water treatment effect, with the core indicators being COD removal rate and sludge moisture content; N represents the energy system, which characterizes energy consumption efficiency, with the core indicators being electricity consumption per unit of water treatment and the proportion of photovoltaic power supply; C represents the economic system, which characterizes the treatment cost, with the core indicators being cost per ton of water treated and carbon trading revenue; and S represents the social system, which characterizes the social impact, with the core indicators being the complaint rate of surrounding residents and the degree of implementation of environmental protection policies.

[0020] Integrating expert knowledge and machine learning to calculate edge weights w ij :

[0021] w ij =α·W e (i,j)+(1-α)·W m (i,j)

[0022] Among them, W e (i,j) represents the expert's score for the impact on system i to j; W m (i,j) represents the training result of the random forest algorithm on historical data; α is the weight fusion coefficient.

[0023] Furthermore, the core theoretical layer also constructs a three-level tree-like architecture of a full-element indicator system, consisting of a target layer, a criterion layer, and an indicator layer. The target layer is for the evaluation of the 3E1S collaborative governance efficiency of wastewater treatment plants; the criterion layer corresponds to the environmental system, energy system, economic system, and social system, respectively; and the indicator layer is divided into positive indicators, negative indicators, and constraint indicators for each system.

[0024] Furthermore, the algorithm models carried by the core technology layer include the SBM-DEA model, the LSTM model, the Monte Carlo algorithm, and the TreeSHAP algorithm;

[0025] The computation layer includes a computation module and a scenario simulation module. The computation module is used to retrieve the SBM-DEA model to quantify resource waste and efficiency. The scenario simulation module is used to generate a wastewater treatment plant treatment strategy, retrieve the LSTM model and Monte Carlo model to simulate and extrapolate the current strategy, and iteratively optimize the parameter values ​​of relevant indicators under the current strategy until the optimal strategy package is output.

[0026] The decision-making layer includes an analysis and decision-making module and a visualization module; the analysis and decision-making module is used to retrieve...

[0027] The TreeSHAP algorithm calculates the correlation strength between different indicators and the four systems; the visualization module is used to visualize the calculation results of the analysis and decision-making module and provide a user interface.

[0028] Furthermore, the process by which the computing module quantifies resource waste and efficiency loss using the SBM-DEA model includes:

[0029] Preprocess the collected indicators;

[0030] For any given indicator, the information entropy H of that indicator is calculated using the entropy weight method. i and through Obtain objective weights By introducing a subjective assignment method, the subjective weights of the indicators are obtained. Through the fusion formula Obtain the dynamic weight w of this indicator. i ;

[0031] Construct a relaxation matrix to quantify resource usage from three dimensions: input, expected output, and unexpected output;

[0032] The efficiency value is calculated using the non-radial SBM-DEA model, and the formula is as follows:

[0033]

[0034] Where ρ represents the efficiency value; m represents the number of input indicators: This represents the dynamic weight of the i-th input indicator. This represents the dynamic weight of the r-th expected output indicator. x represents the dynamic weight of the t-th undesirable output indicator; ij0 s1 represents the actual value of the i-th input indicator of the j-th wastewater treatment plant; s2 represents the expected output: y rj0 s1 represents the actual value of the r-th expected output indicator of the j-th wastewater treatment plant; s2 represents the number of unexpected outputs: b tj0 This represents the actual value of the t-th undesired output indicator of the j-th wastewater treatment plant; Indicates the input slack variable; P represents the expected output slack variable; t This represents the penalty coefficient for undesirable output; This represents the maximum value of the undesired output;

[0035] The computational task of the SBM-DEA model is decomposed into multiple parallel tasks using the Spark distributed computing framework. The slack variables and efficiency values ​​are solved iteratively. When the difference in the efficiency value between two iterations is less than 0.01, the model is considered to have converged, and the final resource waste quantification result and efficiency value are output.

[0036] Furthermore, the steps of the scenario simulation module to deduce the treatment strategy of the wastewater treatment plant include:

[0037] You can call a wastewater treatment plant treatment strategy from the preset strategy library or input a custom wastewater treatment plant treatment strategy. Each wastewater treatment plant treatment strategy contains parameter values ​​for multiple indicators.

[0038] The LSTM model is used to capture the long-term and short-term dependencies of seasonal and lagging indicators in wastewater treatment plant governance strategies, and to obtain the future time series data of the corresponding indicators.

[0039] The Monte Carlo algorithm is used to randomly sample the parameter values ​​of indicators that are susceptible to uncertainty in the current wastewater treatment plant treatment strategy. The output is the probability distribution of the predicted effect of the current strategy, and the potential benefits and risk boundaries of the strategy are presented.

[0040] The four systems are set to have a synergy index (SCI) ≥ 0.6, and the simulation error is > 0.1, as a dual-condition iteration mechanism. If the simulation result does not meet this dual-condition iteration mechanism, the relevant indicator parameters in the current strategy are adjusted, and the simulation is repeated until the dual-condition iteration mechanism is met, outputting the optimal strategy package. The optimal strategy package includes: environmental benefits, economic benefits, and implementation path. Among them, environmental benefits include the increase in COD removal rate and carbon emission reduction; economic benefits include the percentage decrease in cost per ton of water and investment payback period; and the implementation path includes a phased transformation plan.

[0041] Furthermore, the synergy index (SCI) of the four systems is calculated as follows:

[0042]

[0043] Among them, S i Standardized scores are given for environmental, energy, economic, and social systems, ranging from 0 to 1.

[0044] w i The weights for the four systems are: environmental system w1 = 0.3, energy system w2 = 0.25, economic system w3 = 0.25, and social system w4 = 0.2.

[0045] Furthermore, the specific process by which the analysis and decision-making module uses the TreeSHAP algorithm to calculate the correlation strength between different indicators and the four systems includes:

[0046] Multiple regression trees are generated iteratively through a pre-built gradient boosting tree model to capture the nonlinear relationship between various indicators and the four system objectives;

[0047] The TreeSHAP algorithm is used to calculate the contribution of each indicator to a given prediction result, i.e., the SHAP value. The calculation formula is as follows:

[0048] φ i =E x'～D [F(x)-F(x' -i )∣x' i =x i ]

[0049] Where, φ i Let x' be the SHAP value of the i-th indicator. The larger the absolute value of the SHAP value, the more significant the impact of the indicator on the four system objectives. iLet x' represent a sample drawn from data distribution D. -i E represents the sample after removing the i-th indicator; x'～D This represents the expectation of a sample x' drawn from the data distribution D, where x is the expected value. i This represents the specific value of the i-th indicator; F(x) represents the target value of the four systems.

[0050] The visualization module outputs three types of core icons based on the SHAP value of each indicator: an indicator weight ranking chart, an indicator interaction relationship chart, and a feature dependency chart. The indicator weight ranking chart sorts the key indicators from largest to smallest according to their absolute SHAP values, displaying the top n key indicators and their SHAP values. The indicator interaction relationship chart displays the synergistic or restrictive relationships between indicators in the form of a heatmap, with color depth representing the intensity of the interaction. The feature dependency chart plots a nonlinear relationship curve between a single indicator value and the four system objectives.

[0051] Furthermore, the visualization module is also used to automatically trigger a red alert and simultaneously highlight related indicators when the four-system synergy index (SCI) < 0.6; when the four-system synergy index (SCI) ≥ 0.6, it displays a normal state, generates an efficiency heatmap of the wastewater treatment unit using 3D rendering technology, intuitively displays the operating status of each area with varying color depths, and plots the dynamic changes of key indicators; it also supports interactive queries, allowing users to click on any indicator to display the calculation results output by the analysis and decision-making module.

[0052] Furthermore, the system also includes an application adaptation layer that supports two types of scenarios: watershed level and plant level. For watershed level, the built-in models of terminals in different regions are trained collaboratively through a federated learning architecture. For plant level, the system outputs the resource waste and efficiency values ​​of a single plant.

[0053] As can be seen from the above technical solution, compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. This invention possesses global analytical capabilities for multi-system collaborative governance:

[0055] This invention overcomes the limitations of traditional single-dimensional optimization by constructing a topological assessment framework that couples four systems: environment, energy, economy, and society, enabling the quantification of dynamic interaction effects between these systems. It can accurately analyze the cascading effects of aeration energy consumption adjustments on water quality compliance rates, or the synergistic relationship between increasing photovoltaic power generation and the cost per ton of water treated. This avoids problems such as soaring energy consumption and uncontrolled costs caused by optimizing a single indicator, providing a holistic perspective for multi-objective balanced governance decisions.

[0056] 2. This invention possesses precise diagnostic capabilities for comprehensive assessment:

[0057] This invention establishes a three-level tree-structured indicator system covering the entire process, incorporating not only conventional indicators such as effluent quality, energy consumption, and cost, but also introducing undesirable output constraints and social impact dimensions. Through objective weighting using the Gini coefficient-entropy weighting method, it accurately identifies governance shortcomings, achieving a leap from fuzzy assessment to precise diagnosis.

[0058] 3. This invention possesses dynamic scenario simulation and intelligent decision support functions:

[0059] This invention combines a preset strategy library and custom strategies to dynamically simulate governance paths. The scenario-based intelligent inference module has a built-in typical strategy library. By combining Monte Carlo simulation and LSTM neural network, it can realize real-time inference and effect prediction of governance strategies and generate the optimal strategy.

[0060] 4. This invention possesses cross-regional adaptation and dynamic expansion capabilities:

[0061] This invention can accurately address the differences in water quality standards across different river basins. At the regional level, it achieves adaptive optimization of model parameters for wastewater treatment plants in different regions through federated learning. It dynamically adjusts the weighting of indicators to address the differences in water quality standards across different river basins, ensuring accurate adaptation for cross-regional assessments. At the plant level, it adopts a modular design with reserved interfaces for new indicators, enabling continuous iteration and meeting long-term governance needs.

[0062] 5. This invention possesses visual risk warning and intelligent operation and maintenance functions:

[0063] This invention can present the collaborative status of four systems in real time. When key indicators are abnormal, it can automatically trigger an early warning and analyze the root cause to assist staff in quickly locating the anomaly. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0065] Figure 1 This is a schematic diagram of the wastewater treatment plant treatment efficiency assessment and dynamic simulation system based on the 3E1S theory provided by the present invention.

[0066] Figure 2 This is a schematic diagram of the interaction relationship of the 3E1S four systems provided by the present invention.

[0067] Figure 3 The three-level tree structure diagram of the full-element indicator system provided by this invention.

[0068] Figure 4 The flowchart provided by this invention quantifies resource waste and efficiency loss using the SBM-DEA model.

[0069] Figure 5 The flowchart shows how the scenario simulation module provided by this invention simulates the treatment strategies of a wastewater treatment plant.

[0070] Figure 6 The flowchart shows the calculation of the correlation strength between different indicators and four systems provided by this invention.

[0071] Figure 7 This is a schematic diagram illustrating the dynamic display of the four-system synergy index provided by the present invention.

[0072] Figure 8 This invention provides a flowchart for federated learning of different client-built-in models in a watershed-level scenario. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] like Figure 1 As shown, this invention discloses a wastewater treatment plant treatment efficiency evaluation and dynamic simulation system based on the 3E1S theory, comprising: a core theory layer, a core technology layer, and a functional module layer; wherein, the functional module layer includes a data layer, a calculation layer, a decision layer, and an application adaptation layer;

[0075] The core theoretical layer is used to define the interaction rules between environmental systems, energy systems, economic systems and social systems based on the 3E1S theoretical framework;

[0076] The core technology layer is used to load various algorithm models and runtime environments;

[0077] The data layer is used to collect multi-source heterogeneous data across four dimensions: environment, energy, economy, and society.

[0078] The computing layer is used to retrieve relevant algorithm models based on the collected multi-source heterogeneous data to quantify resource waste and efficiency values, and to dynamically simulate governance paths by combining preset strategy libraries and custom strategies to perform iterative optimization and obtain the optimal strategy package;

[0079] The decision-making level uses a visual early warning system to present the coordinated status among the four systems: environmental, energy, economic, and social.

[0080] The application adaptation layer supports two types of scenarios: watershed level and plant level. For watershed level, it uses a federated learning architecture to collaboratively train the built-in models of wastewater treatment plant terminals in different regions. For plant level, it outputs the resource waste and efficiency values ​​of a single plant.

[0081] The functions of each layer of the system architecture of this invention will be further explained below.

[0082] 1. Core Theoretical Layer: The core theoretical layer is based on the 3E1S theoretical framework, which clearly defines the interaction rules between the four systems of environment, energy, economy and society. It clarifies the logical relationship of the environmental system as the foundation, the energy system as the driver, the economic system as the constraint, and the social system as the feedback, thus anchoring the direction for subsequent technological applications.

[0083] (1) Construction of bidirectional coupling topology of four systems:

[0084] like Figure 2 As shown, for wastewater treatment plants, this paper breaks through the limitations of traditional single-dimensional assessment and transforms the 3E1S theory into a quantifiable indicator network, which is visualized as a two-way coupled model of four systems: water quality, energy consumption, cost, and social impact. Water treatment effect (environmental system), energy efficiency (energy system), treatment cost (economic system), and surrounding resident complaint rate (social system) are used as topological nodes. The environmental system serves as the foundation, the energy system as the driver, the economic system as the constraint, and the social system as the feedback. A directed weighted graph is used to represent the interaction between the four systems, and the node set is represented as V = {E, N, C, S}.

[0085] Among them, E represents the environmental system, which characterizes the water treatment effect, with the core indicators being COD removal rate and sludge moisture content; N represents the energy system, which characterizes energy consumption efficiency, with the core indicators being electricity consumption per unit of water treatment and the proportion of photovoltaic power supply; C represents the economic system, which characterizes the treatment cost, with the core indicators being cost per ton of water treated and carbon trading revenue; and S represents the social system, which characterizes the social impact, with the core indicators being the complaint rate of surrounding residents and the degree of implementation of environmental protection policies.

[0086] Integrating expert knowledge and machine learning to calculate edge weights w ij :

[0087] w ij =α·W e (i,j)+(1-α)·W m (i,j)

[0088] Among them, W e (i,j) represents the expert's score for the impact on systems i to j, with a value ranging from 0.5 to 0.8; W m (i,j) represents the training result of the random forest algorithm on historical data, such as W obtained based on 5 years of running data. m(NE) = 0.6; α is the weighted fusion coefficient, and the final w NE =0.4×0.7+0.6×0.6=0.64, quantifying the effect of photovoltaic power supply on carbon emission suppression.

[0089] The weights W are calculated using the random forest algorithm. m The steps for (i,j) are as follows: Construct a model based on the random forest algorithm, collect historical operating data of the wastewater treatment plant, and take energy system N and environmental system E as examples to organize the indicators of energy system N, such as daily unit water treatment power consumption and photovoltaic power supply ratio over 5 years, as well as the corresponding indicators of environmental system E, such as COD removal rate and sludge moisture content.

[0090] A random forest model is trained using historical data to learn how changes in the indicators of the energy system N affect the indicators of the environmental system E.

[0091] After the random forest model is trained, it outputs the feature importance of each energy system feature to the environmental system indicator, that is, the contribution of a 1-unit change in a feature to the prediction result of the environmental system indicator. The importance of all features within the energy system is aggregated to obtain the overall influence weight W of the energy system N on the environmental system E. m (NE).

[0092] (2) Quantification of the interaction relationship among the four systems:

[0093] 1) Construct a positive impact dynamic model based on LSTM neural networks to capture the lag and nonlinear effects between systems. This model is universal and applicable to the interaction scenarios of indicators between any two systems within four systems (environmental system, energy system, economic system, and social system). Its core is to quantify the dynamic impact of indicators in system A on indicators in system B by mining the correlation patterns of indicator changes in time-series data. The positive impact dynamic model is expressed as:

[0094] E imp (t)=β0+β1·X A (t)+β2·X A (tk)+∈(t)

[0095] In the formula, E imp (t) represents the magnitude of the positive impact of the index of system A on the index of system B at time t, such as the improvement rate or amount of improvement of the index of system B; X A (t) represents the target interaction index of system A at time t (which can be replaced by any core index of any system); X A(tk) represents the system A index lagged by k time units; β0 is a constant term, β1 is the influence coefficient of the current system A index on the system B index, β2 is the influence coefficient of the system A index lagged by k time units on the system B index; ∈(t) is the random error term, reflecting the small disturbance factors that the model did not capture.

[0096] Taking the impact of aeration on water quality as an example: By analyzing seasonal operating data of wastewater treatment plants, it is possible to identify chain reactions with lag and nonlinearity, such as a decrease in water temperature in winter → a 15% increase in aeration energy consumption → fluctuations in effluent ammonia nitrogen compliance rate. This allows for the automatic generation of seasonal operating strategies. At this point, X in the formula... A E represents the aeration time at time t. imp (t) represents the positive impact of aeration on water quality at time t; β1 = 0.28, indicating that for every 1 hour increase in aeration time at the current time, the daily COD removal rate can be increased by 2.8%; β2 = 0.15, indicating that for every 1 hour increase in aeration time the previous day, the daily COD removal rate can still be increased by 1.5%; k = 1, determined by the statistical analysis of the time difference between aeration adjustment and water quality response in historical data; ∈(t) ranges from ±0.02, mainly affected by factors such as water temperature and fluctuations in influent pollutant concentration.

[0097] The model replaces X A With E imp The specific metrics of (t) can be quickly adapted to the metric interaction scenarios of other systems.

[0098] 2) For the sludge treatment process, establish an early warning threshold that the treatment cost will increase non-linearly by 20% when the sludge moisture content is >80%, and trigger the recommendation to add conditioning agents in advance to avoid cost out-of-control.

[0099] The constructed inverse constraint threshold model is as follows:

[0100]

[0101] In the formula, C sludge (t) represents the sludge disposal cost at time t (yuan / ton); C0 = 200 yuan / ton (base cost); when the moisture content is 85%, C sludge =200×(1+0.15×e) (5％) / 5％ )=200×(1+0.15×e 1 The price is approximately 303 yuan / ton, reflecting a non-linear penalty.

[0102] 3) Construction of a comprehensive indicator system:

[0103] The core theoretical layer also constructs a three-level tree-like architecture for a comprehensive indicator system, consisting of a target layer, a criterion layer, and an indicator layer. Figure 3As shown; the target layer is the 3E1S collaborative governance efficiency assessment of wastewater treatment plants; the criteria layer corresponds to the environmental system (water quality dimension), energy system (energy consumption dimension), economic system (cost dimension), and social system (social impact dimension). The indicator layer is divided into positive indicators, negative indicators, and constraint indicators for each system.

[0104] The environmental system focuses on effluent quality compliance, COD emissions exceeding standards, and total phosphorus emissions exceeding standards. It combines COD removal rate, total phosphorus removal rate, and effluent compliance rate to cover the environmental impact of the entire wastewater treatment process. Through water quality compliance constraints and pollutant reduction targets, it quantifies the positive contributions and potential negative impacts of treatment processes on the environment. Positive indicators such as COD removal rate are used to measure treatment effectiveness, negative indicators such as total phosphorus emissions exceeding standards monitor shortcomings, and constraint indicators such as the lower limit of effluent compliance rate define compliance red lines.

[0105] The energy system incorporates unit water treatment power consumption, photovoltaic power generation ratio, and energy recovery rate, combined with carbon emission intensity, aeration equipment idle rate, and energy loss rate. Through total energy consumption constraints and carbon emission limits, it achieves a refined assessment of energy utilization efficiency, capturing the impact of clean energy substitution and equipment energy efficiency optimization on energy consumption. Among these, positive indicators such as the photovoltaic power generation ratio reflect energy-saving and efficiency-enhancing results, negative indicators such as aeration energy waste rate expose energy waste problems, and constraint indicators such as the total energy consumption ceiling control the scale of energy consumption.

[0106] The economic system focuses on the cost per ton of water treated, the rate of return on sludge resource utilization, and carbon trading revenue. It combines annual operating cost budget constraints, energy procurement cost overruns, and additional costs associated with equipment failures. Utilizing Life Cycle Cost Analysis (LCCA), and taking into account constraints such as the lower limit of return on investment, it balances short-term operating expenditures with long-term upgrade benefits, covering both cost control and revenue management dimensions of the economic system. Positive indicators such as carbon trading revenue reflect economic efficiency, negative indicators such as reagent waste costs warn of cost loopholes, and constraints such as the lower limit of return on investment safeguard investment value.

[0107] The social system relies on resident satisfaction, environmental policy implementation, and public awareness of environmental protection, combined with resident complaint rates, negative media exposure, and factors contributing to social instability. Through constraints such as social impact assessment compliance requirements and minimum social responsibility fulfillment standards, it quantifies the impact of wastewater treatment plant operations on the surrounding social environment, filling gaps in social collaborative governance assessments. Positive indicators like resident satisfaction demonstrate social acceptance, negative indicators like complaint numbers reflect public demands, and constraint indicators such as minimum policy implementation standards ensure compliance with governance guidelines.

[0108] By constructing an influence coefficient matrix through edge weights among the four systems, the interaction relationships between them can be quantified. For example, the increased proportion of photovoltaic power supply in the energy system can alleviate the carbon emission intensity in the environmental system, and the tightening of environmental protection policies in the social system can constrain the transformation costs in the economic system. The coefficients are calibrated by fusion of expert scoring and machine learning to achieve accurate quantification of the synergistic effect of multiple systems. Through such multi-level, multi-dimensional and clearly categorized indicator settings, the core control elements of the four systems are fully covered, providing accurate and systematic quantitative basis for efficiency assessment and optimization decisions of the 3E1S collaborative governance of sewage treatment plants.

[0109] 2. The core technology layer comprises two main pillars: the algorithm engine and the computational support runtime environment. The algorithm models used in the core technology layer include the SBM-DEA model, the LSTM model, the Monte Carlo algorithm, and the TreeSHAP algorithm. The computational support runtime environment includes: the Spark distributed computing architecture, federated learning technology, and data fusion technology.

[0110] The SBM-DEA model is used to accurately assess efficiency and uncover resource slack and efficiency loss issues; the LSTM model focuses on time series prediction and captures the dynamic trends of data such as water quality and energy consumption; the TreeSHAP algorithm deeply analyzes the correlation between indicators and clarifies the impact of each factor on the four systems.

[0111] In terms of computing support, Spark's distributed computing architecture accelerates big data processing and significantly improves computational efficiency through its parallel computing capabilities. Federated learning architecture enables cross-regional data collaboration while ensuring data privacy, breaking down data barriers between factories. Multi-source data fusion technology effectively solves the problem of data heterogeneity, integrating multi-source data such as PLC sensors, policy documents, and resident complaints. The specific fusion process is as follows:

[0112] ① Multi-source data preprocessing:

[0113] For real-time data, extreme values ​​are first filtered using the 3σ principle, and then the Isolation Forest algorithm is used to detect abnormal data. Abnormal data is filled with the mean of the previous 3 normal data.

[0114] For text data, meaningless characters are removed, and invalid information, such as traffic congestion content unrelated to sewage treatment, is eliminated through keyword matching.

[0115] ② Standardize and unify data to eliminate differences in data units:

[0116] For numerical data, Z-Score standardization is used to transform the data into a standard distribution with a mean of 0 and a standard deviation of 1. For skewed distributions or data with a defined range, Min-Max standardization is used to map the data to the interval [0,1].

[0117] For categorized data, such as policy implementation rate and complaint type, the text classification information is converted into numerical codes. For example, the policy implementation rate is divided into excellent, good, qualified, and unqualified, with corresponding codes of 4 / 3 / 2 / 1, to ensure that the categorized data can participate in the algorithm calculation.

[0118] For time-based data, it is uniformly converted to YYYY-MM-DDHH:MM format, and time features are extracted to provide adapted input for LSTM models to capture seasonal time series dependencies.

[0119] The merged data is stored uniformly in a distributed database and classified according to four systems and indicator types, such as the environmental system - COD concentration - 2024-05-01 - 33.8 mg / L.

[0120] 3. The functional modules are further subdivided into a data layer, a computation layer, and a decision layer. The data layer is responsible for data collection, cleaning, and standardization, providing high-quality data for upper-level computation. The computation layer relies on multi-model computation and combines scenario simulation with a pre-set strategy library and custom strategies to explore diverse governance paths. The decision layer conducts in-depth analysis of the computation results, intuitively presenting the collaborative status of the four systems through visual early warnings, while also supporting user interaction to facilitate user participation in strategy adjustments.

[0121] (1) The data layer extensively collects heterogeneous data from multiple sources, captures equipment operation data of the PLC system in real time through the OPCUA interface, and captures real-time production data such as water quality and energy consumption using PLC sensors; it connects to the provincial environmental protection database to obtain watershed water quality standards and supports the custom import of local characteristic indicators; it obtains environmental management requirements from policy documents and collects social feedback information through resident complaint channels, comprehensively gathering data from environmental, energy, economic, and social dimensions. It integrates the 3σ principle and the isolated forest algorithm to clean the data, automatically filters outliers, and uses Kalman filtering to fuse multi-source data to ensure the consistency and reliability of data such as water quality and energy consumption.

[0122] (2) The computation layer includes a computation module and a scenario simulation module. The computation module is used to retrieve the SBM-DEA model to quantify resource waste and efficiency. The scenario simulation module is used to generate wastewater treatment plant treatment strategies, retrieve the LSTM model and Monte Carlo model to simulate and deduce the current strategy, and iteratively optimize the parameter values ​​of relevant indicators under the current strategy until the optimal strategy package is output.

[0123] 1) In the system of this invention, the SBM-DEA model is a key technical support for achieving efficient resource utilization and precise cost control. Its core objective is to quantify resource slack and efficiency loss, providing a clear and operable optimization direction for energy saving and cost reduction in the sewage treatment process. For example, it can accurately identify hidden waste problems such as idle aeration equipment and excessive addition of chemicals, thereby guiding the adjustment of operation strategies.

[0124] The process of quantifying resource waste and efficiency loss using the SBM-DEA model is as follows: Figure 4 As shown, it specifically includes:

[0125] ① Preprocessing of collected indicators: Data quality directly determines the reliability of model output, so the first step is a rigorous data preprocessing process, employing the 3σ principle and the Isolation Forest algorithm to clean outliers. The 3σ principle, based on the normal distribution characteristics of the data, identifies and filters extreme data that deviate from the mean by three times the standard deviation, effectively handling occasional abrupt anomalies such as those occurring in water quality monitoring; the Isolation Forest algorithm, by constructing isolated trees, quickly detects isolated points in the dataset, specifically cleaning outliers caused by transient equipment failures, etc.

[0126] After outlier handling, the Z-Score or Min-Max standardization method should be flexibly selected based on the distribution characteristics of the indicator data: for indicators that are approximately normally distributed, Z-Score standardization is used, and the formula is as follows: Transform the data into a standard distribution with a mean of 0 and a standard deviation of 1; for skewed distributions or indicators with a defined range (such as the number of complaints), use Min-Max standardization, as shown in the formula. Mapping the data to the [0,1] interval ensures that different types of indicators are comparable in subsequent calculations, laying a data foundation for accurate model calculations.

[0127] ② Dynamic Weight Setting: Appropriate weight setting for indicators is crucial for accurate efficiency evaluation. The adaptive SBM-DEA model employs a dual-dimensional weighting method combining objective and subjective approaches. For any indicator i, the entropy H of that indicator is calculated using the entropy weighting method. i and through Obtain objective weights This method fully respects the dispersion and contribution of the data itself, reflecting the objective influence of the indicators in the overall data; it introduces a subjective assignment method to obtain the subjective weights of the indicators. Through the fusion formula (Typically β = 0.6, balancing objective data and expert experience), ultimately yielding the dynamic weight w of this indicator. i The weight of this indicator not only aligns with the actual distribution of data but also conforms to the logic of industry governance, ensuring that the indicator assessment more accurately reflects the true efficiency requirements of wastewater treatment.

[0128] In the iterative solution of the SBM-DEA model, the adjustment of the index weights is a dynamic adaptive process. The adjustment is mainly based on the deviation between the efficiency value calculation results of each iteration and the feedback from the actual system operation. The specific steps are as follows:

[0129] When the efficiency value obtained from a certain iteration deviates significantly from the actual operating performance of the wastewater treatment system, the indicator weights will be adjusted. For example, if the efficiency value calculated by the model shows that a wastewater treatment plant has high efficiency, but the actual wastewater treatment plant has serious problems of idle aeration equipment and wasted chemicals, which does not match the calculation results, the indicator weights need to be adjusted.

[0130] The adjustments are based on efficiency value deviation analysis and changes in indicator contribution. Efficiency value deviation analysis compares the differences between the iteratively calculated efficiency values ​​and the actual values ​​to identify which indicator weights were improperly set, leading to the deviation. Changes in indicator contribution reflect the shifts in the contribution of each indicator to resource waste and efficiency as the wastewater treatment system operates. During the iteration process, changes in data characteristics such as the dispersion of each indicator and its correlation with other indicators are continuously monitored. If the information entropy of an indicator changes significantly, indicating a change in its objective influence on the overall data, the objective weight of that indicator is recalculated based on the new information entropy, and the dynamic weights are adjusted in conjunction with expert feedback on subjective weights.

[0131] The adjustment process involves re-executing the objective and subjective weighting process to calculate the adjusted dynamic weights, which are then used in the next iteration to improve the model's accuracy in evaluating efficiency.

[0132] ③ Relaxation variation dynamic processing mechanism:

[0133] To accurately pinpoint resource waste and efficiency losses, a relaxation matrix is ​​constructed to quantify resource usage across three dimensions: input, expected output, and unexpected output. The construction process is as follows:

[0134] Input dimension: By statistically analyzing the differences between the actual values ​​and optimal input values ​​of various input indicators, such as pesticide dosage, aeration time, and labor costs, input slack variables are obtained.

[0135] Expected output dimension: Calculate the difference between the actual values ​​and the maximum expected output values ​​of various expected output indicators, such as effluent compliance rate and sludge resource utilization rate, to obtain the expected output slack variable.

[0136] Undesirable Output Dimension: This dimension measures the deviation between the actual and standard values ​​of various undesirable output indicators, such as carbon emission intensity and sludge moisture content. A penalty value is calculated using a nonlinear penalty function and incorporated into the relaxation matrix. For example, if the actual moisture content b exceeds the standard, the penalty calculated using the nonlinear penalty function reflects the negative impact of undesirable output on efficiency, and is thus reflected in the relaxation quantification related to undesirable output.

[0137] The relaxation amounts of these three dimensions are categorized and organized according to indicators to form a matrix, which is the relaxation matrix. This matrix comprehensively presents the waste or insufficiency of resources in the input and output stages.

[0138] After constructing the relaxation matrix, the next step is to calculate the negative constraints on undesirable outputs and the nonlinear penalty function. For undesirable outputs in wastewater treatment, such as excessive sludge moisture content and excessive carbon emissions, a nonlinear penalty function needs to be introduced to reflect their negative impact on efficiency. Taking excessive sludge moisture content as an example, the constructed nonlinear penalty function can be expressed as:

[0139]

[0140] Where b is the actual sludge moisture content, b0 is the standard value of sludge moisture content; k is the penalty coefficient, which can be determined based on factors such as the additional cost of sludge disposal. For example, it is calculated that when the sludge moisture content exceeds the standard value by 1%, it will bring an additional expenditure equivalent to 2% of the unit water treatment cost. In this case, k = 0.02 can be set; δ is the adjustment parameter, which is used to control the growth rate of the penalty function. If it is desired that the penalty after the moisture content exceeds the standard will increase rapidly, δ can be set to a smaller value, such as δ = 0.5.

[0141] When the sludge moisture content is within the standard (b≤b0), the penalty is 0; when it exceeds the standard (b>b0), the penalty increases exponentially with the excess moisture content, thus accurately reflecting the severe impact of undesirable output exceeding the standard on wastewater treatment efficiency. Integrating this type of nonlinear penalty function into the objective function or constraints of the SBM-DEA model can more realistically quantify the efficiency loss caused by undesirable output, making the model's assessment of resource waste and efficiency more accurate and improving the entire quantification process.

[0142] The efficiency value is calculated using the non-radial SBM-DEA model, and the formula is as follows:

[0143]

[0144] Where ρ represents the efficiency value (range 0-1, the closer to 1 the higher the efficiency); This represents the dynamic weight of the i-th input indicator. This represents the dynamic weight of the r-th expected output indicator. The dynamic weight of the t-th undesirable output indicator; m represents the number of input indicators, x ij0 This represents the actual value of the i-th input indicator for the j-th wastewater treatment plant (decision-making unit); for example, m = 3, x1 = reagent dosage (kg / day), x2 = aeration time (hours / day), x3 = labor cost (yuan / day).

[0145] s1 represents the expected output: y rj0This represents the actual value of the r-th expected output indicator of the j-th wastewater treatment plant; for example: s1 = 2, y1 = effluent compliance rate (tons / day), y2 = sludge resource utilization rate (tons / day).

[0146] s2 represents the number of unexpected outputs: b tj0 This represents the actual value of the t-th undesired output indicator of the j-th wastewater treatment plant; for example: s2 = 2, b1 = carbon emission intensity (kgCO2 / ton water), b2 = sludge moisture content (%);

[0147] Indicates the input slack variables (e.g.) (Hours indicates that the aeration equipment has been idle for 2 hours); This represents the expected output slack variable, for example The figure of tons indicates that the amount of effluent meeting the standard did not reach the expected 50 tons; P t This represents the penalty coefficient for undesirable output; This represents the maximum value of the undesired output;

[0148] ④ The computational task of the SBM-DEA model is decomposed into multiple parallel tasks using the Spark distributed computing framework. The slack variables and efficiency values ​​are iteratively solved. If the difference in efficiency value between two iterations is less than 0.01, the model is considered converged, and the final resource waste quantification result and efficiency value are output. The specific iterative process is as follows:

[0149] a. Initialization Randomly generate a weight vector λ(0);

[0150] b. The kth iteration:

[0151]

[0152] c. The process terminates when |ρ(k)-ρ(k-1)|<0.01, indicating model convergence. The output relaxor value (such as the specific quantitative results of resource waste such as aeration time redundancy and reagent waste) and efficiency value ρ (the closer ρ is to 1, the higher the resource utilization efficiency of the wastewater treatment process) are provided to the operation manager. This provides clear and accurate decision-making basis and guides the formulation and implementation of subsequent energy-saving and cost-reduction strategies.

[0153] This invention overcomes the limitations of traditional DEA models, which focus on the overall picture while neglecting local aspects, by introducing a non-radial SBM model, and accurately identifies resource waste and efficiency loss. For scenarios such as excessive idle rates of aeration equipment and redundant chemical dosing, a nonlinear optimization algorithm iteratively solves for input relaxation values, quantifying local efficiency losses. When aeration energy consumption exceeds 30%, it is marked as an energy-inefficient unit; when chemical waste exceeds 8%, a calibration recommendation is triggered, achieving a closed-loop diagnosis of resource efficiency across multiple systems.

[0154] 2) The core objective of the scenario simulation module is to accurately predict the synergistic effects of environmental, energy, economic, and social systems by simulating diversified governance strategies, providing wastewater treatment plants with an optimal strategy combination that is both forward-looking and feasible. This module achieves a complete transformation from abstract strategies to quantitative results through a closed-loop process of strategy input—multi-model simulation—collaborative analysis—iterative optimization. The process is as follows: Figure 5 As shown, the specific steps include:

[0155] ① Strategy Input: To adapt to the different scenarios of different wastewater treatment plants, a dual-path strategy input mechanism is designed.

[0156] Preset strategy library: Based on industry practice and technological maturity, it has a variety of typical solutions built in, covering combinations such as 3MW photovoltaic installation + 1000kWh energy storage, sludge anaerobic digestion + cogeneration, aeration frequency conversion retrofit + precise dosing of chemicals, etc. Each strategy includes default values ​​for core parameters (such as photovoltaic panel conversion efficiency and anaerobic digester temperature control range). You can call a wastewater treatment plant treatment strategy from the preset strategy library for rapid simulation.

[0157] Custom strategy input: Users can input personalized parameters based on plant characteristics (such as land area and existing equipment), such as "water reuse rate increased to 40% + MBR membrane process upgrade". The system automatically analyzes the correlation between parameters and generates a custom simulation plan.

[0158] This input mechanism satisfies the need for standardized and rapid evaluation while also allowing for in-depth exploration of personalized scenarios, laying a flexible strategic foundation for subsequent simulations.

[0159] ② Multi-model collaborative simulation: To overcome the limitations of a single model, a dual-model collaborative architecture of LSTM neural network and Monte Carlo simulation is adopted to simulate time series data and non-time series data respectively, so as to achieve accurate prediction of policy effects and risk quantification.

[0160] The LSTM model is used to capture the long-term and short-term dependencies of seasonal and lagging indicators in the treatment strategies of wastewater treatment plants. For example, the drop in water temperature in winter causes a two-day lag in aeration effect, and the future time series data of the corresponding indicators are obtained.

[0161] The model is input with historical data, such as daily aeration time and COD concentration over the past three years, and trained to obtain a time-series prediction formula:

[0162]

[0163] Among them, h t For the hidden layer state, W yThe output weights are σ, which is the activation function. It can accurately predict water quality fluctuations (MAE≤0.05mg / L) and energy consumption peaks (error≤5%) within 72 hours after the strategy is implemented, providing a basis for the time-series adaptation of the strategy.

[0164] The Monte Carlo algorithm is used to randomly sample the parameter values ​​of indicators that are susceptible to uncertainties in the current wastewater treatment plant's treatment strategy multiple times, and output the probability distribution of the predicted effect of the current strategy.

[0165] For example, considering the uncertainties of strategy parameters, such as the impact of weather on the actual power generation of photovoltaic panels and fluctuations in carbon trading prices, 10,000 random samples are taken from core parameters, such as photovoltaic installed capacity and anaerobic digestion gas production, to output the probability distribution of the strategy's effect. For instance, when simulating a photovoltaic + energy storage strategy, quantitative results such as a carbon emission reduction of 20%-30% (95% confidence interval) and an investment payback period of 5-7 years (80% probability) can be obtained, intuitively presenting the potential returns and risk boundaries of the strategy.

[0166] The dual-model collaboration not only ensures the accuracy of short-term time series forecasts but also enables probabilistic evaluation of long-term effects, providing comprehensive support for strategy feasibility analysis.

[0167] ③ Four-System Synergistic Analysis: The effectiveness of a strategy depends not only on the optimization of individual systems but also on the synergistic improvement of all four systems. The module achieves global performance evaluation by calculating the Synergy Index (SCI). The SCI is calculated as follows:

[0168]

[0169] Among them, S i Standardized scores for environmental, energy, economic, and social systems, ranging from 0 to 1, are generated using Min-Max standardization. i The weights for the four systems are: environmental system w1 = 0.3, energy system w2 = 0.25, economic system w3 = 0.25, and social system w4 = 0.2.

[0170] Among them, S i The standardized scores of each system are divided into time-series and non-time-series indicators. For time-series indicators, such as water quality COD concentration, aeration energy consumption, and daily water treatment volume, which change dynamically over time, future time-series data are predicted using an LSTM model, and the statistical values ​​of typical time slices are taken as representative values ​​of the indicators.

[0171] For non-time-series indicators, such as photovoltaic installed capacity, investment cost of sludge digesters, and number of resident complaints, which do not fluctuate frequently over time, the Monte Carlo algorithm is used to randomly sample the parameters and output the statistical characteristic values ​​of the probability distribution, such as the mean, median, or representative value of the 95% confidence interval, as the representative value of the indicator.

[0172] Then, through Min-Max standardization, the representative values ​​of each indicator are mapped to the [0,1] interval to obtain the standardized score Si of each system:

[0173]

[0174] Where: X represents the representative value of a time-series or non-time-series indicator; X min This represents the minimum value of the indicator, a theoretical minimum value determined based on historical data or industry standards; X max This represents the maximum value of the indicator, which is the theoretical maximum value determined based on historical best data.

[0175] If the SCI ≥ 0.6, it indicates synergistic compliance, meaning the strategy can balance the needs of the four systems; if the SCI < 0.6, it indicates a significant weakness. For example, the simulation results of a certain sludge anaerobic digestion strategy show an environmental system score of 0.8, an energy system score of 0.7, an economic system score of 0.5, and a social system score of 0.6. The calculated SCI = 0.3 × 0.8 + 0.25 × 0.7 + 0.25 × 0.5 + 0.2 × 0.6 = 0.65, which is judged as synergistic compliance.

[0176] ④ Iterative Optimization: To ensure the accuracy and feasibility of the strategy, a dual-condition iterative mechanism is set up. This requires a synergy index (SCI) ≥ 0.6 for the four systems and an effect error > 0.1 in the strategy simulation. The effect error refers to the deviation between the dual-model simulation results and the historical best value, which refers to the optimal effect achieved by the wastewater treatment plant in similar scenarios using a similar strategy. If the simulation results do not meet this dual-condition iterative mechanism, an automatic parameter adjustment process is triggered to adjust the relevant parameters in the current strategy. For environmental shortcomings, the aeration frequency is increased and the reagent dosage is optimized; for energy inefficiency, the photovoltaic installed capacity is adjusted and the energy storage duration is optimized; for economic imbalances, the initial investment and operation and maintenance costs are balanced.

[0177] After each adjustment, the simulation is repeated until the dual-condition iteration mechanism is met, and the optimal strategy package is output. The optimal strategy package includes: environmental benefits, economic benefits, and implementation path. Among them, environmental benefits include the increase in COD removal rate and carbon emission reduction; economic benefits include the percentage decrease in cost per ton of water and investment payback period; implementation path includes phased transformation plan, such as photovoltaic grid connection in the first quarter and commissioning of sludge digestion tank in the second quarter.

[0178] The time-series projection results of the LSTM model provide a basis for the dynamic details of environmental and economic benefits, ensuring the time-series adaptability of the strategy in the short term.

[0179] The probabilistic projections from Monte Carlo provide a basis for the risk boundaries of environmental and economic benefits, ensuring the feasibility and resilience of the strategy in long-term implementation.

[0180] The results from both models jointly support the S-system collaborative analysis. i The SCI is calculated, and then through iterative optimization, a combination of policy parameters that satisfies SCI≥0.6 and error<0.1 is selected, and finally the optimal policy package is generated.

[0181] This closed-loop process transforms abstract governance goals into quantifiable and implementable strategies, providing scientific and precise decision-making support for the low-carbon and refined governance of wastewater treatment plants.

[0182] (3) The decision-making layer includes an analysis and decision-making module and a visualization module. The analysis and decision-making module is used to retrieve the TreeSHAP algorithm to calculate the correlation strength between different indicators and the four systems; the visualization module is used to visualize the calculation results of the analysis and decision-making module and provide a user interface. The specific implementation process is as follows: Figure 6 As shown.

[0183] 1) The analysis and decision-making module uses the TreeSHAP algorithm to calculate the correlation strength between different indicators and the four systems. The TreeSHAP algorithm plays a crucial role in the indicator influence decoder, focusing on identifying key influencing indicators and deeply exploring the influence strength and patterns of factors such as aeration time and photovoltaic ratio on the four systems of environment, energy, economy, and society. This provides a clear direction for strategy optimization and makes complex indicator correlations quantifiable and visualized. The specific process includes:

[0184] ① Multiple regression trees are iteratively generated using a pre-built gradient boosting tree model to capture the nonlinear relationship between various indicators and the four system objectives; the learning process follows the formula below:

[0185]

[0186] Where K is the number of trees, It is a tree-structured space.

[0187] Each tree focuses on learning local data patterns, and multiple trees collaboratively construct a non-linear mapping relationship from indicators to the four-system objectives. This effectively characterizes complex chain effects such as extended aeration time → improved water quality → increased social satisfaction. During training, cross-validation is used to optimize model parameters, ensuring that the model has good generalization ability while fitting the training data, and can accurately predict the trend of the indicators' impact on the four-system objectives under unknown data.

[0188] ② SHAP value calculation: The TreeSHAP algorithm is used to calculate the contribution of each indicator to a certain prediction result, i.e., the SHAP value. The calculation formula is as follows:

[0189] φ i =E x'～D [F(x)-F(x' -i )∣x' i =x i ]

[0190] Where, φ i Let x' be the SHAP value of the i-th indicator. The larger the absolute value of the SHAP value, the more significant the impact of the indicator on the four system objectives. i Let x' represent a sample drawn from data distribution D. -i E represents the sample after removing the i-th indicator; x'～D This represents the expectation of a sample x' drawn from a data distribution D. It is obtained by averaging a large number of samples conforming to the data distribution D, and is used to quantify the average impact at the data distribution level; x i The i-th indicator represents the specific value of the i-th indicator, which is a specific indicator value used in the four-system target assessment of a wastewater treatment plant, such as the aeration energy consumption value or COD removal rate value of a wastewater treatment plant; the data distribution D refers to the probability distribution followed by the sample data used for analysis; F(x) represents the target value of the four systems.

[0191] Taking the aeration time index as an example, if the calculated φ = 0.28, it means that under the current data distribution, for every unit change in aeration time, the predicted value of the four system targets changes by an average of 0.28 units. This can be used as a key indicator. Meanwhile, the sludge dewatering agent type φ = 0.05 indicates that its impact on the four system targets is relatively weak and can be considered as a secondary factor in strategy optimization.

[0192] 2) The visualization module outputs three types of core icons based on the SHAP value of each indicator: indicator weight ranking chart, indicator interaction relationship chart, and feature dependency chart.

[0193] The indicator weight ranking chart is sorted from largest to smallest by absolute value of SHAP, showing the top n key indicators and their SHAP values; for example, aeration time φ = 0.28, photovoltaic ratio φ = 0.22, sludge moisture content φ = 0.18, etc., clearly showing which indicators play a leading role in the synergistic effect of the four systems.

[0194] In the indicator interaction diagram, an influence factor matrix is ​​generated, and the synergistic or restrictive relationships between indicators are displayed in the form of a heatmap, with the color intensity representing the strength of the interaction. For example, the proportion of photovoltaic power generation is negatively correlated with carbon emission intensity; the bluer the color, the stronger the negative correlation, intuitively reflecting the synergistic effect of increasing the proportion of photovoltaic power generation leading to a decrease in carbon emission intensity. Using all the indicators to be analyzed as rows and columns, an n×n influence factor matrix is ​​formed, where n is the number of indicators. The element in the i-th row and j-th column of the matrix is ​​the quantified result of the interaction SHAP value between indicator i and indicator j. This is usually the absolute value of the interaction SHAP value, or a normalized value, representing the strength of the interaction.

[0195] The feature dependency graph plots the nonlinear relationship curves between individual index values ​​and the four system objectives. For example, the impact of sludge moisture content on cost shows that when the moisture content is ≤80%, the cost increases slowly with increasing moisture content; when the moisture content is >80%, the cost increases exponentially, clearly revealing the nonlinear impact of index changes on system objectives.

[0196] These three types of visualizations allow operations managers to quickly grasp the key impacts and interaction patterns of metrics without needing to deeply understand complex algorithms, providing a clear and intuitive basis for strategy optimization.

[0197] In addition, the four-system synergy index dashboard serves as a real-time monitoring window for the 3E1S synergistic governance effect of wastewater treatment plants. It transforms complex data from the four systems of environment, energy, economy, and society into intuitive decision-making basis, enabling a global perception of the governance status and timely response to abnormal indicators.

[0198] like Figure 7 As shown, core indicators of the four systems are collected in real time through a three-level data access system. The collected data is standardized using Z-Score or Min-Max, mapping indicators of different dimensions to the [0,1] interval, thus obtaining the standardized score norm(S) for each system. i This lays a comparable foundation for subsequent index calculations. Based on standardized scores, the system quantifies the synergistic effect of the four systems using a synergistic index formula.

[0199] When SCI ≥ 0.6, the system is considered to have good synergy. In this case, the visualization module uses 3D rendering technology to generate an efficiency heatmap of the wastewater treatment unit, visually displaying the operating status of each area with varying color intensity and plotting the dynamic changes of key indicators. It also supports interactive queries; clicking on any indicator will display the calculation results output by the analysis and decision-making module. Otherwise, a red alert is automatically triggered, and related indicators are highlighted to help operators quickly identify weaknesses.

[0200] 4. The application adaptation layer possesses powerful scenario adaptation capabilities, supporting both watershed-level (cross-plant collaboration) and plant-level (single-plant optimization) scenarios. For the watershed level, it adapts to the differences in water quality standards across different watersheds, generating region-specific models through federated learning to promote cross-regional collaborative governance. For the plant level, it focuses on issues such as resource waste and efficiency shortcomings in single plants, outputting customized optimization solutions for each plant to accurately meet different governance needs. This four-layer architecture progresses layer by layer, jointly supporting the efficient operation of the wastewater treatment plant AI collaborative governance system.

[0201] In cross-plant collaborative governance, cross-regional data privacy issues are addressed through federated learning and cross-regional optimization. The specific process is as follows: Figure 8 As shown, it includes:

[0202] a. Multi-region initialization: Connect to clients of different watersheds or plant types, mark regional characteristics (such as total phosphorus limits), and lay the foundation for model adaptation.

[0203] b. Local model training: Each factory independently trains SBM-DEA and LSTM models based on local data, outputting only the local parameters θ. i The original data is stored locally to protect privacy.

[0204] c. Encrypted Parameter Aggregation: Parameters are transmitted using homomorphic encryption and aggregated using the FedAvg algorithm with weighted aggregation. Where, N i Let N be the sample size of the i-th sample, and N be the total sample size, integrating regional commonalities and individual characteristics.

[0205] d. Regional adaptability assessment: Verify the adaptability of the global model to different watersheds (e.g., the total phosphorus prediction error in the Yangtze River Basin is ≤10%). If the standard is not met, adjust the weights and re-aggregate.

[0206] e. Global model deployment: Generate watershed-specific models, securely transmit them to clients, and support cross-regional strategy collaborative optimization.

[0207] This process ensures data privacy while enabling regional model collaboration, thereby improving the accuracy and efficiency of cross-basin governance.

[0208] This invention deeply integrates the 3E1S theory with AI technology, and through four-system coupled modeling, full-element index design, adaptive SBM-DEA efficiency assessment, TreeSHAP index analysis, scenario-based strategy deduction, and federated learning cross-regional optimization, it provides a quantifiable and implementable full-chain solution for the low-carbon, refined, and intelligent governance of wastewater treatment plants, promoting the upgrading of the wastewater treatment industry towards a high-quality development model of multi-system synergistic efficiency.

[0209] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0210] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A wastewater treatment plant treatment efficiency evaluation and dynamic simulation system based on the 3E1S theory, characterized in that, include: The system comprises a core theoretical layer, a core technology layer, and a functional module layer; wherein the functional module layer includes a data layer, a computing layer, and a decision-making layer. The core theoretical layer, based on the 3E1S theoretical framework, defines the interaction rules between the environmental system, energy system, economic system, and social system. It uses the environmental system as the foundation, the energy system as the driver, the economic system as the constraint, and the social system as the feedback. A directed weighted graph is used to represent the interaction relationships between the four systems, with the node set represented as... ; in, For environmental systems, used to characterize water treatment effectiveness, the core indicators are COD removal rate and sludge moisture content; For energy systems, it is used to characterize energy efficiency, with the core indicators being electricity consumption per unit of water treatment and the proportion of photovoltaic power supply; For the economic system, it is used to characterize the treatment cost, with the core indicators being the cost per ton of water treated and carbon trading revenue; For social systems, it is used to characterize social impact, with core indicators being the complaint rate from surrounding residents and the degree of implementation of environmental protection policies; Integrating expert knowledge and machine learning to calculate edge weights : in, For experts to review the system arrive Impact score; This represents the training results of the Random Forest algorithm on historical data. The weighted fusion coefficient; The core theoretical layer also constructs a three-level tree-like architecture of a full-element indicator system, consisting of a target layer, a criterion layer, and an indicator layer. The target layer is for the 3E1S collaborative governance efficiency assessment of wastewater treatment plants. The criterion layer corresponds to the environmental system, energy system, economic system, and social system, respectively. The indicator layer is divided into positive indicators, negative indicators, and constraint indicators for each system. The core technology layer is used to load various algorithm models and runtime environments; The data layer is used to collect multi-source heterogeneous data in four dimensions: environment, energy, economy, and society. The computing layer is used to retrieve relevant algorithm models based on the collected multi-source heterogeneous data to quantify resource waste and efficiency values, and to dynamically simulate governance paths by combining preset strategy libraries and custom strategies to perform iterative optimization and obtain the optimal strategy package. The decision-making layer is used to present the coordinated state among the four systems—environmental, energy, economic, and social—through a visual early warning system. The algorithm models carried by the core technology layer include SBM-DEA model, LSTM model, Monte Carlo algorithm and TreeSHAP algorithm; The computation layer includes a computation module and a scenario simulation module. The computation module is used to retrieve the SBM-DEA model to quantify resource waste and efficiency. The scenario simulation module is used to generate a wastewater treatment plant treatment strategy, retrieve the LSTM model and Monte Carlo model to simulate and extrapolate the current strategy, and iteratively optimize the parameter values ​​of relevant indicators under the current strategy until the optimal strategy package is output. The decision-making layer includes an analysis and decision-making module and a visualization and display module. The analysis and decision-making module is used to retrieve the TreeSHAP algorithm to calculate the correlation strength between different indicators and the four systems. The visualization and display module is used to visualize the calculation results of the analysis and decision-making module and provide a user interface.

2. The wastewater treatment plant treatment efficiency evaluation and dynamic simulation system based on the 3E1S theory according to claim 1, characterized in that, The process by which the computing module quantifies resource waste and efficiency loss using the SBM-DEA model includes: Preprocess the collected indicators; For any given indicator, the information entropy of that indicator is calculated using the entropy weight method. and through Obtain objective weights Subjective assignment method is introduced to obtain the subjective weights of the indicators. ; through the fusion formula The dynamic weight of this indicator is obtained. ; Construct a relaxation matrix to quantify resource usage from three dimensions: input, expected output, and unexpected output; The efficiency value is calculated using the non-radial SBM-DEA model, and the formula is as follows: in, Indicates the efficiency value; Indicates the number of input indicators: This represents the dynamic weight of the i-th input indicator. This represents the dynamic weight of the r-th expected output indicator. Let be the dynamic weight of the t-th undesirable output indicator; Indicates the first The first sewage treatment plant The actual value of each input indicator; Indicates the expected output: Indicates the first The first sewage treatment plant The actual value of the expected output indicator; Indicates the number of unexpected outputs: Indicates the first The first sewage treatment plant The actual value of an undesirable output indicator; Indicates the input slack variable; This represents the expected output slack variable; This represents the penalty coefficient for undesirable output; This represents the maximum value of the undesired output; The computational task of the SBM-DEA model is decomposed into multiple parallel tasks using the Spark distributed computing framework. The slack variables and efficiency values ​​are solved iteratively. When the difference in the efficiency value between two iterations is less than 0.01, the model is considered to have converged, and the final resource waste quantification result and efficiency value are output.

3. The wastewater treatment plant treatment efficiency evaluation and dynamic simulation system based on the 3E1S theory according to claim 1, characterized in that, The steps of the scenario simulation module in deriving the treatment strategies for wastewater treatment plants include: You can call a wastewater treatment plant treatment strategy from the preset strategy library or input a custom wastewater treatment plant treatment strategy. Each wastewater treatment plant treatment strategy contains parameter values ​​for multiple indicators. The LSTM model is used to capture the long-term and short-term dependencies of seasonal and lagging indicators in wastewater treatment plant governance strategies, and to obtain the future time series data of the corresponding indicators. The Monte Carlo algorithm is used to randomly sample the parameter values ​​of indicators that are susceptible to uncertainty in the current wastewater treatment plant treatment strategy. The output is the probability distribution of the predicted effect of the current strategy, and the potential benefits and risk boundaries of the strategy are presented. The four systems are set to have a synergy index (SCI) ≥ 0.6, and the simulation error is > 0.1, as a dual-condition iteration mechanism. If the simulation result does not meet this dual-condition iteration mechanism, the relevant indicator parameters in the current strategy are adjusted, and the simulation is repeated until the dual-condition iteration mechanism is met, outputting the optimal strategy package. The optimal strategy package includes: environmental benefits, economic benefits, and implementation path. Among them, environmental benefits include the increase in COD removal rate and carbon emission reduction; economic benefits include the percentage decrease in cost per ton of water and investment payback period; and the implementation path includes a phased transformation plan.

4. The wastewater treatment plant treatment efficiency evaluation and dynamic simulation system based on the 3E1S theory according to claim 3, characterized in that, The synergy index (SCI) of the four systems is calculated as follows: in, Standardized scores are given for environmental, energy, economic, and social systems, ranging from 0 to 1. The weights of the four systems, environmental system Energy system economic system Social system .

5. The wastewater treatment plant treatment efficiency evaluation and dynamic simulation system based on the 3E1S theory according to claim 3, characterized in that, The specific process by which the analysis and decision-making module uses the TreeSHAP algorithm to calculate the correlation strength between different indicators and the four systems includes: Multiple regression trees are generated iteratively through a pre-built gradient boosting tree model to capture the nonlinear relationship between various indicators and the four system objectives; The TreeSHAP algorithm is used to calculate the contribution of each indicator to a given prediction result, i.e., the SHAP value. The calculation formula is as follows: in, Let be the SHAP value of the i-th indicator. The larger the absolute value of the SHAP value, the more significant the impact of the indicator on the four system objectives. This represents a sample drawn from data distribution D. This represents the sample after removing the i-th indicator; This represents a sample drawn from the data distribution D. Seeking expectations, This represents the specific value of the i-th indicator; Indicates the target values ​​of the four systems; The visualization module outputs three types of core icons based on the SHAP value of each indicator: an indicator weight ranking chart, an indicator interaction relationship chart, and a feature dependency chart. The indicator weight ranking chart sorts the key indicators from largest to smallest according to their absolute SHAP values, displaying the top n key indicators and their SHAP values. The indicator interaction relationship chart displays the synergistic or restrictive relationships between indicators in the form of a heatmap, with color depth representing the intensity of the interaction. The feature dependency chart plots a nonlinear relationship curve between a single indicator value and the four system objectives.

6. The wastewater treatment plant treatment efficiency evaluation and dynamic simulation system based on the 3E1S theory according to claim 4, characterized in that, The visualization module is also used to automatically trigger a red alert and highlight related indicators when the four-system synergy index (SCI) is less than 0.6; when the four-system synergy index (SCI) is greater than or equal to 0.6, it displays a normal state, generates an efficiency heatmap of the wastewater treatment unit using 3D rendering technology, intuitively displays the operating status of each area with varying color shades, and plots the dynamic changes of key indicators; it also supports interactive queries, allowing users to click on any indicator to display the calculation results output by the analysis and decision-making module.

7. The wastewater treatment plant treatment efficiency evaluation and dynamic simulation system based on the 3E1S theory according to claim 1, characterized in that, The system also includes an application adaptation layer that supports two types of scenarios: watershed level and plant level. For watershed level, the built-in models of terminals in different regions are trained collaboratively through a federated learning architecture. For plant level, the system outputs the resource waste and efficiency values ​​of a single plant.

Images