Circulating water treatment closed-loop optimization method and system based on water quality prediction

Abstract

The invention relates to the technical field of circulating water treatment, and discloses a circulating water treatment closed-loop optimization method and system based on water quality prediction.The circulating water treatment closed-loop optimization method based on water quality prediction.The circulating water treatment closed-loop optimization method based on water quality predictioncomprises the steps that a time sequence knowledge graph supporting the time dimension is constructed and used for representing water quality parameters and time sequence evolution characteristics of the relation of the water quality parameters; constructing a multi-time scale prediction engine to predict a future water quality state; developing a predictive constraint generator, and converting future water quality prediction into current decision constraints; a space-time sensitive multi-objective optimization algorithm is introduced, and a medicament proportioning strategy is optimized; a multi-time-scale collaborative optimization system is realized, so that short-term response and long-term stability are further balanced; an environment adaptation active evolution system is constructed, and automatic adjustment of the system along with environment changes is achieved; according to the invention, the fundamental transformation of a circulating water treatment control mode is realized, and a brand new closed-loop optimization normal form is provided for a circulating water treatment system.

Description

Technical Field

[0001] The present invention relates to the technical field of circulating water treatment, and more particularly to a closed-loop optimization method and system for circulating water treatment based on water quality prediction. Background Art

[0002] Circulating water treatment systems are widely used in industrial cooling water, central air conditioning cooling towers, and chemical production processes. Their core goal is to ensure efficient and stable operation by monitoring and regulating water quality parameters. Existing circulating water treatment systems typically include infrastructure such as water quality monitoring equipment, chemical addition equipment, filtration equipment, and circulation pumps, combined with automated control methods to achieve water quality management.

[0003] However, existing technologies still have many deficiencies in the automation and intelligence of circulating water treatment, which are mainly reflected in the following aspects:

[0004] The current mainstream circulating water treatment control systems mostly adopt a feedback-based reactive control strategy, that is, adjustments are made only after it is detected that the water quality parameters exceed the set threshold. This control method has obvious lag, especially the effects of regulatory measures such as the addition of chemicals often take a certain amount of time to appear, resulting in the system being unable to respond to rapid changes in water quality in a timely manner, especially in the event of sudden water quality fluctuations, the system response is not timely enough, affecting the stability of water quality; in existing technologies, water quality prediction models and control systems are mostly independent of each other, and the prediction results are difficult to use directly for control decisions. Due to the lack of an efficient data interface and coordination mechanism between the prediction system and the control system, the prediction information cannot fully guide control operations such as the addition of chemicals, limiting the application of prediction-driven feedforward control strategies; existing circulating water treatment systems have limitations in terms of control requirements at multiple time scales, and some systems focus too much on short-term water quality optimization and ignore long-term operational stability. Qualitative; while other systems emphasize long-term stability at the expense of the ability to respond quickly to short-term water quality fluctuations. How to achieve a balance in water quality control effects at different time scales such as hourly, daily and even weekly levels is a problem that current technology urgently needs to solve; there are complex dynamic correlations between water quality parameters in the circulating water system (such as pH value, conductivity, turbidity, residual chlorine, etc.), and these correlations change with time and environmental conditions. Existing technologies mostly use static models or simple time series analysis methods, which are difficult to fully capture and model the temporal evolution laws between water quality parameters, resulting in limited understanding and prediction capabilities of the system's dynamic behavior; existing circulating water treatment systems generally lack the ability to adapt to external environmental changes (such as seasonal changes, raw water quality fluctuations, etc.), and system parameters usually need to rely on manual regular adjustment, making it difficult to achieve autonomous adaptation and continuous optimization to environmental disturbances, affecting the system's intelligence level and operational efficiency.

[0005] In summary, the existing circulating water treatment technology has obvious deficiencies in terms of response timeliness, prediction and control coordination, multi-time scale balance, complex time series relationship modeling and environmental adaptability. It is urgent to propose new technical solutions to improve the intelligence and optimization level of circulating water treatment systems. Summary of the Invention

[0006] The present invention provides a closed-loop optimization method and system for circulating water treatment based on water quality prediction, which solves the technical problems of response lag, separation of prediction and control, difficulty in multi-time scale balance and poor environmental adaptability in related technologies.

[0007] The present invention provides a closed-loop optimization method for circulating water treatment based on water quality prediction, comprising:

[0008] Construct a time-series knowledge graph that supports the time dimension to represent the temporal evolution characteristics of water quality parameters and their relationships;

[0009] Based on the time series knowledge graph, a multi-time scale prediction engine is constructed to predict future water quality status;

[0010] Develop a predictive constraint generator based on future water quality state prediction results to convert future water quality predictions into current decision constraints;

[0011] Introducing a time-space sensitive multi-objective optimization algorithm to optimize the drug ratio strategy based on the prediction results and current decision constraints;

[0012] Based on the optimized drug ratio strategy, a multi-time scale collaborative optimization system is realized to further balance short-term response and long-term stability;

[0013] Based on the optimization results of the multi-time-scale collaborative optimization system, an environmentally adaptive active evolutionary system is constructed to enable the system to automatically adjust as the environment changes.

[0014] Furthermore, the step of constructing a time series knowledge graph supporting the time dimension includes:

[0015] The water quality parameters are defined as entity nodes in the graph, each node contains parameter name, value, and timestamp attributes;

[0016] Apply the relationship discovery algorithm based on Granger causality test to identify the potential causal relationships between parameters and represent the potential causal relationships as directed edges in the graph;

[0017] A temporal pattern learning method based on graph neural network is used to extract the temporal evolution patterns of water quality parameters from historical data.

[0018] Furthermore, the multi-timescale prediction engine is built based on a hybrid architecture of graph convolutional networks and gated recurrent units, including:

[0019] Spatial feature extraction module, used to capture the spatial dependencies between nodes;

[0020] Time series feature extraction module, used to capture time series characteristics;

[0021] Attention fusion layer, used to fuse spatial features and temporal features;

[0022] The multi-timescale prediction head includes three parallel branches: short-term prediction, medium-term prediction, and long-term prediction.

[0023] Furthermore, the multi-time-scale prediction engine also introduces Monte Carlo dropout technology to provide a confidence interval for each prediction result. Through random dropout forward propagation, the prediction distribution is obtained, the mean is calculated as the final prediction value, and the variance is used as the uncertainty measure.

[0024] Furthermore, the predictive constraint generator includes:

[0025] Key threshold and target interval definition module, which defines the target range and warning threshold for each water quality parameter;

[0026] The prediction trajectory risk assessment module performs risk analysis on the predicted water quality parameter trajectory, taking into account parameter importance, deviation degree and probability of crossing the boundary;

[0027] The dynamic constraint generation module generates constraint conditions for control decisions based on risk assessment results.

[0028] Furthermore, the time-space sensitive multi-objective optimization algorithm includes:

[0029] Construct a reagent ratio model and a response model to describe the relationship between the reagent addition amount and water quality parameters;

[0030] Construct objective functions for the short-term, medium-term, and long-term time scales, focusing on response speed, stability, and cost-effectiveness respectively;

[0031] An improved particle swarm optimization algorithm is used to solve multi-objective optimization problems using continuous-discrete hybrid coding, adaptive penalty function and multi-swarm co-evolution strategy.

[0032] Furthermore, the multi-timescale collaborative optimization system includes:

[0033] Establish a three-tier control structure consisting of emergency response layer, daily control layer, and strategy optimization layer;

[0034] Build an inter-layer information transmission and decision coordination system to ensure that control decisions at different time scales are compatible and consistent;

[0035] According to the system status and prediction results, the activation priority of control layers at different time scales is dynamically adjusted.

[0036] Furthermore, the environmental adaptation active evolution system includes:

[0037] Model performance monitoring and evaluation module, which continuously monitors the performance of the prediction model and calculates the prediction error index by comparing the predicted value with the actual observed value;

[0038] The incremental learning and model update module uses incremental learning methods to update the time series knowledge graph and prediction model;

[0039] Seasonal pattern adaptation module, which identifies and learns seasonal variation patterns of water quality parameters;

[0040] The control effect feedback and optimization module records control operations and their effects, and uses reinforcement learning methods to optimize control strategies.

[0041] Furthermore, the incremental learning and model update module includes three key components: a selective memory module, an elastic weight adjustment algorithm, and a knowledge distillation framework, which are used to evaluate the value of new data, protect key knowledge, and balance new and old knowledge.

[0042] The present invention provides a closed-loop optimization system for circulating water treatment based on water quality prediction, which is used to execute the above-mentioned closed-loop optimization method for circulating water treatment based on water quality prediction, comprising:

[0043] The time series knowledge graph module is used to represent the time series evolution characteristics of water quality parameters and their relationships;

[0044] A multi-timescale prediction engine for predicting future water quality status based on a time-series knowledge graph;

[0045] A predictive constraint generator to convert future water quality predictions into current decision constraints;

[0046] Multi-objective optimization module, used to optimize the drug ratio strategy based on the prediction results and constraints;

[0047] A multi-timescale collaborative control module to balance short-term response and long-term stability;

[0048] The environment adaptation active evolution module is used to enable the system to automatically adjust to environmental changes.

[0049] The beneficial effects of the present invention are: capturing the dynamic correlation and temporal evolution of water quality parameters through a time-series knowledge graph, achieving high-precision prediction of water quality changes, and improving the prediction accuracy compared to traditional methods, especially in predicting water quality mutation events;

[0050] Early intervention based on multi-timescale prediction results transforms traditional passive response control into active preventive control, enabling the system to deal with potential water quality problems in advance and reduce the risk of water quality fluctuations;

[0051] Through multi-objective optimization and predictive constraint generation, the reagent ratio strategy is optimized while ensuring stable water quality, reducing the amount of reagents used and reducing energy consumption;

[0052] The environmental adaptation active evolution system enables the invention to autonomously adapt to external factors such as seasonal changes and fluctuations in raw water quality, and continuously maintain optimal performance without human intervention;

[0053] The multi-timescale collaborative optimization system enables the present invention to quickly respond to short-term disturbances while maintaining long-term stability, and the water quality stability is improved compared with traditional systems;

[0054] Through a unified timing-driven framework, the traditionally separate prediction system and control system are integrated into a single architecture, reducing the complexity of interface conversion, reducing the number of system parameters, and improving operational efficiency;

[0055] The present invention realizes a fundamental transformation of the circulating water treatment control mode, upgrading from the traditional "measurement, judgment, and reaction" mode to the "prediction, prevention, and optimization" mode, providing a new closed-loop optimization paradigm for the circulating water treatment system. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of a closed-loop optimization method for circulating water treatment based on water quality prediction in the present invention;

[0057] Figure 2 It is a line graph of the present invention's predicted trends for different water quality parameters within 24 hours;

[0058] Figure 3 is a scatter plot of the Pareto optimal solution set obtained by the multi-objective optimization algorithm of the present invention;

[0059] Figure 4 It is a radar chart showing the performance differences between the method of the present invention and the traditional control method in five dimensions;

[0060] Figure 5 It is an area chart showing the accuracy change trend of the incremental learning model and the traditional fixed model during long-term use;

[0061] Figure 6 It is a dual-axis graph showing the relationship between water temperature and microbial activity index in a circulating water system with seasonal changes;

[0062] Figure 7 It is a bar chart showing the performance differences between the traditional water quality control method and the method of the present invention in three key indicators. DETAILED DESCRIPTION

[0063] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0064] At least one embodiment of the present invention discloses a closed-loop optimization method for circulating water treatment based on water quality prediction, such as Figure 1 Shown, including:

[0065] Step 1: Construct a time-series knowledge graph that supports the time dimension to represent the time-series evolution characteristics of water quality parameters and their relationships;

[0066] This step first collects historical water quality parameter data for the circulating water system, including parameters such as pH, conductivity, turbidity, residual chlorine, hardness, and microbial counts, as well as environmental data such as temperature, humidity, and seasonal information. The collected data is preprocessed, including outlier detection, missing value filling, and data standardization. A time series knowledge graph is then constructed through the following sub-steps. The time series knowledge graph constructs key water quality parameters such as pH, conductivity, turbidity, residual chlorine, hardness, and microbial counts as graph nodes. The Granger causality test is used to discover dynamic associations between parameters (such as changes in microbial activity caused by changes in pH, the relationship between conductivity and scaling risk, etc.), and the evolution of these relationships over time is recorded:

[0067] Water quality parameters (such as pH, conductivity, turbidity, and residual chlorine) have different dimensions and numerical ranges and require normalization to map them to the same numerical range (e.g., 0-1 or -1 to 1) for fair comparison during subsequent map construction and model training. For parameters with a fixed range, such as pH, minimum-maximum normalization can be used. For parameters with theoretically unlimited upper limits, such as conductivity and turbidity, Z-score normalization or logarithmic transformation followed by normalization should be used. Categorical data, such as seasonal information, needs to be one-hot encoded and converted to numerical features.

[0068] Step 1.1, entity definition and attribute extraction;

[0069] Water quality parameters are defined as entity nodes in the graph, each of which contains attributes such as parameter name, value, and timestamp. Using a sliding time window method, statistical characteristics (such as mean, variance, and rate of change) are calculated for each parameter as the time series feature attributes of the node.

[0070] Step 1.2, relationship discovery and graph construction;

[0071] A Granger causality-based relationship discovery algorithm is applied to identify potential causal relationships between parameters and represent them as directed edges in the graph. The strength of the relationship is quantified using the time-lagged correlation coefficient, while the relationship type is determined by the direction of the correlation (positive or negative).

[0072] Before calculating time-lagged correlation coefficients between different water quality parameters, the parameters must be normalized to eliminate the effects of differences in dimension and numerical range. For example, pH values ​​typically range from 1 to 14, while conductivity can range from hundreds to thousands of microsiemens / cm. Directly calculating correlations can result in the larger parameter dominating the results. All parameter series should be Z-score normalized or min-max normalized to ensure fair correlation calculations.

[0073] The mathematical expression of the timing relationship is as follows:

[0074] ;

[0075] in, Representation parameters At the moment Parameters At the moment The strength of the influence relationship, is the time-lagged correlation function, and Represents parameters respectively and parameters time series data.

[0076] Time-lagged correlation function The specific implementation is to calculate the mutual correlation coefficient of two time series at a given time lag, and combine it with the p-value of the Granger causality test to form a comprehensive score. When the score exceeds the preset threshold, the existence of a causal relationship is confirmed and a directed edge connection is established.

[0077] Step 1.3, extraction of temporal evolution patterns;

[0078] A time series pattern learning method based on graph neural network is used to extract the evolution patterns of water quality parameters over time from historical data. A graph state transition matrix is ​​constructed to capture the evolution pattern of the system under different conditions. The dynamic evolution process of the time series graph can be expressed as: for any moment The system's dynamic characteristics are described by calculating the rate of change over time based on the state of the knowledge graph. This rate of change depends on three key factors: the current graph state (reflecting the current values ​​of water quality parameters and their relationship structure), the control vector (including the addition amounts of various reagents and adjustment operations), and the environmental factor vector (external conditions such as temperature and humidity). These three factors are comprehensively processed through a state transition function to obtain the rate of change of the graph state, that is, the immediate trend of change in the attributes and relationship strength of each node. The state transition function is implemented using a hybrid model that combines graph neural networks and differential equations: First, the graph neural network processes the current graph state and calculates the basic trend of change of each node under the current relationship structure; then, these trends are adjusted according to the control vector to reflect the impact of control operations on the system state; finally, the final state derivative value is generated, taking into account the modulation of the rate of change by environmental factors. Through this mechanism, the system can simulate and predict the continuous evolution of water quality parameters over time, providing a theoretical basis for prediction and control.

[0079] The temporal knowledge graph in this application has the following specific structure:

[0080] Multi-level node system: Graph nodes are divided into three categories: parameter nodes, aggregation nodes, and environmental nodes. Parameter nodes represent single water quality parameters; aggregation nodes represent parameter combinations, capturing the synergistic effects of multiple parameters; and environmental nodes represent external environmental factors.

[0081] Time-sensitive edges: Edges in the graph contain a time lag attribute, which represents the time it takes for an influence to propagate from a source node to a destination node. Furthermore, edge weights can change over time, reflecting the dynamic nature of the relationship strength.

[0082] Context subgraph: Build specialized context subgraphs for different operating conditions (such as high load, low temperature, etc.). Each subgraph captures the parameter relationship under specific conditions.

[0083] Memory structure: The graph consists of two parts: long-term steady-state relationships and short-term dynamic relationships. Long-term relationships represent stable causal relationships, while short-term relationships capture temporary strong correlations.

[0084] The temporal knowledge graph is kept up to date through a dynamic update module: when new observation data is received, the node attribute values ​​are first updated, then the edge relationship strength is calculated based on the new data, new edges are added or weakly related edges are removed when necessary, and finally the graph state transition matrix is ​​updated.

[0085] Step 2: Based on the time series knowledge graph, a multi-time scale prediction engine is built to predict future water quality status;

[0086] This step uses the time series knowledge graph constructed in step 1 to build a multi-time scale prediction engine that can provide short-term, medium-term and long-term predictions. The prediction engine provides differentiated predictions for different water quality parameter characteristics, such as hourly accurate predictions for fast-changing parameters such as pH, daily trend predictions for slowly changing parameters such as hardness, and weekly trend predictions for seasonal parameters such as microbial indicators. At the same time, it considers the interaction between parameters, such as the accelerated effect of rising temperature on the residual chlorine decay rate, to achieve accurate prediction of compound effects. Specifically, it includes:

[0087] Step 2.1, build a graph neural network prediction model;

[0088] The method proposed in this application builds a time series graph prediction model based on a hybrid architecture of a graph convolutional network (GCN) and a gated recurrent unit (GRU). This model takes graph node features and topology as input, extracts spatial dependencies through multi-layer graph convolution, and then captures time series characteristics through a recurrent network.

[0089] Specifically, the graph neural network prediction model of this application contains the following key components:

[0090] Spatial Feature Extraction Module: This module consists of a three-layer graph convolutional network, with each layer containing 32, 64, and 128 convolution kernels, respectively, to capture the spatial dependencies between nodes. The graph convolution operation is implemented as follows: first, the graph adjacency matrix is ​​obtained and self-loop connections are added to form an augmented adjacency matrix. The node degree matrix, representing the number of connections per node, is then calculated. The augmented adjacency matrix is ​​then normalized to a symmetric normalized form. The node feature matrix of the current layer is then multiplied by the normalized adjacency matrix, and then by a learnable weight matrix. Finally, the ReLU activation function is applied to the resulting feature representation to generate the node feature representation of the next layer. This process aggregates and transforms node features, enabling each node to incorporate information from its neighboring nodes.

[0091] Time series feature extraction module: This module consists of a bidirectional GRU with a hidden layer dimension of 256, which is used to capture time series characteristics. The historical feature sequence of each node is processed by the GRU to form a time series feature representation.

[0092] Attention fusion layer: A multi-head self-attention structure (8 attention heads) is used to fuse spatial features and temporal features and calculate the importance weights between different features.

[0093] Multi-timescale prediction head: This consists of three parallel branches, corresponding to short-term, medium-term, and long-term predictions. Each branch consists of two fully connected layers (with dimensions of 128 and 64, respectively) and an output layer with dimensions equal to the number of predicted parameters.

[0094] It should be noted that the model training adopts self-supervised learning, constructing training samples from historical data, using mean squared error as the loss function, and adding L2 regularization to prevent overfitting. The model is trained using the Adam optimizer with a learning rate of 0.001 and a batch size of 32.

[0095] Step 2.2, multi-time scale prediction framework construction;

[0096] A hierarchical forecasting strategy is employed, with specialized forecast heads constructed for different forecast time spans. Short-term forecasts (1-6 hours) prioritize detailed accuracy, medium-term forecasts (1-3 days) balance accuracy and trends, and long-term forecasts (1-2 weeks) focus on capturing macro trends. Each forecast head utilizes an attention structure to focus on different timescale features.

[0097] Step 2.3, quantification of prediction uncertainty;

[0098] Monte Carlo dropout technology is introduced to provide confidence intervals for each prediction result. Through multiple random dropout forward propagations, the prediction distribution is obtained, and the mean is calculated as the final prediction value, with the variance as a measure of uncertainty. The prediction confidence is used for risk assessment of subsequent control decisions.

[0099] Optionally, in some embodiments, an ensemble learning method may be used to train multiple models with different structures or parameters and comprehensively consider their prediction results to further improve prediction accuracy and robustness.

[0100] like Figure 2 The figure below shows the system's 24-hour prediction trends for different water quality parameters (pH, conductivity, and turbidity). The solid line represents the predicted value, and the dashed area represents the confidence interval for the prediction. This intuitively demonstrates the short-term prediction capabilities of the multi-timescale prediction engine. As can be seen from the figure, the system accurately captures the changing trends of water quality parameters and quantifies the uncertainty of the prediction using confidence intervals, providing a reliable basis for subsequent control decisions.

[0101] Step 3: Develop a predictive constraint generator based on the future water quality state prediction results to convert the future water quality prediction into current decision constraints;

[0102] This step generates constraints that guide current control decisions based on the prediction results of step 2. The constraint generator assigns priorities based on the importance of water quality parameters to achieve prediction-based feedforward control. The prediction-driven constraint generation mechanism transforms circulating water treatment from passive response to active prevention. The system no longer waits until water quality parameters actually exceed the standard before taking action, but instead makes precise interventions before problems occur. Specifically, it includes:

[0103] Step 3.1, definition of key thresholds and target intervals;

[0104] Define target ranges and warning thresholds for each water quality parameter, including ideal operating ranges, acceptable fluctuation ranges, and absolute limits. Assign weights based on the importance of different parameters and construct a parameter priority matrix.

[0105] Step 3.2, risk assessment of predicted trajectory;

[0106] Risk analysis is performed on the predicted water quality parameter trajectories to identify potential risk of crossing the boundary and unstable trends. The degree of risk is quantified by the deviation of the parameter prediction value from the target range and the probability of crossing the boundary. The risk assessment calculation method is as follows: First, the importance weight of each water quality parameter is determined. , reflecting the degree of impact of the parameter on system safety and performance. The normalized Euclidean distance between the predicted parameter value and the target value is then calculated, with a larger distance indicating a greater deviation from the target. The probability of the parameter value exceeding the allowable range is then calculated. This probability is based on the predicted uncertainty distribution and is obtained by calculating the probability density of the predicted distribution falling outside the allowable range. Finally, the parameter weight, distance metric, and out-of-bounds probability are multiplied together to obtain a comprehensive risk assessment value. This risk assessment comprehensively considers the parameter's importance, degree of deviation, and likelihood of out-of-bounds, providing a basis for subsequent constraint generation. For high-risk parameters, the system generates stricter control constraints; for low-risk parameters, it provides greater control flexibility.

[0107] The target values ​​and tolerances for different water quality parameters vary significantly. Directly calculating the Euclidean distance can lead to the parameter with the larger tolerance dominating the risk assessment. Instead, the deviation of each parameter should be divided by its tolerance to obtain the relative deviation, which is then used to calculate the normalized Euclidean distance. For the calculation of the probability of exceeding the tolerance, the uncertainty distribution needs to be normalized to a unified probability space to ensure that the risk of exceeding the tolerance for different parameters is comparable.

[0108] Step 3.3, dynamic constraint generation;

[0109] Based on the risk assessment results, constraints are generated for the current control decision. For high-risk predictions, mandatory constraints are generated; for medium-risk predictions, soft constraints are generated; and for low-risk predictions, optimization objectives are generated. Constraints are expressed as upper and lower limits or target intervals for control variables (such as the dosage of the drug).

[0110] It should be understood that in some embodiments, the constraint generator may also consider factors such as the operating cost of the system and the life of the equipment to generate multi-dimensional constraints to achieve more comprehensive optimization control.

[0111] Step 4: Introduce a time-space sensitive multi-objective optimization algorithm to optimize the drug ratio strategy based on the prediction results and current decision constraints;

[0112] This step uses a multi-objective optimization algorithm based on the constraints generated in step 3 to optimize the efficiency of chemical use and energy consumption while meeting the water quality parameter targets. Specifically, it includes:

[0113] Step 4.1, drug ratio model construction;

[0114] Construct a response model of the relationship between the amount of reagent added and water quality parameters, taking into account the interaction and hysteresis effects between reagents. The model is trained based on historical control data and can predict the impact of different reagent ratio schemes on water quality parameters. The optimal control strategy is determined by first defining the set of control variables , including various combinations of chemical addition amounts and timings. A cost function is then defined, comprehensively considering multiple target indicators under the current and predicted future state of the map. A cost value is then calculated for each possible control strategy, and the control strategy with the lowest cost is selected as the optimal solution. The response model for the relationship between chemical addition amount and water quality parameters utilizes a hybrid modeling approach, combining a multi-level recurrent neural network with physicochemical constraints. The model inputs include the current water quality state vector (pH, conductivity, turbidity, and other parameters), a 24-hour historical water quality trajectory, a chemical operation vector (chemical addition amounts and timings), an environmental condition vector (temperature, humidity, system load), and a system state identifier. The outputs are the evolution of water quality parameters over the next 1-24 hours, a chemical effect index, a parameter fluctuation risk assessment, and a prediction of chemical residual concentration. Chemical interactions are modeled using a dedicated interaction feature layer to capture synergistic or antagonistic relationships between chemicals, such as the neutralization of oxidizing biocides and reducing cleaning agents, or the impact of scale inhibitors on the effectiveness of pH regulators. Multi-order interaction features are automatically extracted using a nonlinear mapping function. Hysteresis effects are accounted for using a time-delayed embedding method to differentiate between fast-response relationships (minutes), medium-response relationships (hours), and slow-response relationships (days). The residual impact of historical operations is dynamically calculated using memory units. The online learning mechanism is based on incremental gradient descent and includes sliding window sample management, differentiated learning rate scheduling, abnormal sample processing, model structure adaptation, and parameter importance assessment.

[0115] The specific implementation of the cost function is as follows: first, each evaluation indicator (including chemical cost, energy consumption, water quality compliance and system stability) is normalized to make their dimensions consistent and the numerical range unified; then, a weight coefficient is assigned to each indicator, and these weight coefficients are dynamically adjusted through system status monitoring and historical data analysis; then, the weighted sum of each indicator is calculated, where the chemical cost part takes into account the unit price and usage of various types of chemicals, the energy consumption part integrates pump power, heating / cooling energy consumption, etc., the water quality compliance part evaluates the degree of consistency between each water quality parameter and the target value, and the system stability part calculates the fluctuation range of key parameters; finally, a smoothing function is applied to handle boundary conditions to ensure that there is still a reasonable optimization direction under extreme conditions.

[0116] The pharmaceutical proportioning model in this application is specifically implemented as a three-layer structure:

[0117] Agent response sub-models: A separate response model is built for each agent to describe its impact on various water quality parameters. These sub-models are based on a combination of physical and chemical principles and regression models trained on historical data, capable of handling nonlinear response relationships.

[0118] Drug interaction layer: Modeling the synergistic and antagonistic effects between different drugs, using second-order interaction terms to capture the mutual influence between drugs.

[0119] Time response layer: The exponential decay function is used to simulate the curve of the change of the agent effect over time, accurately reflecting the dynamic response process of water quality parameters after the addition of the agent.

[0120] The drug ratio model training adopts a two-stage approach: first, the basic response relationship is initialized using historical data, and then the model parameters are continuously optimized through online learning to adapt to changes in system characteristics.

[0121] Step 4.2, multi-time scale objective function construction;

[0122] Construct objective functions for three timescales: short-term (response speed), medium-term (stability), and long-term (cost-effectiveness), focusing on different optimization priorities. A weighted sum method is used to integrate these three objectives into a comprehensive evaluation function, with weight coefficients dynamically adjusted based on system status.

[0123] The short-term objective function focuses on quickly correcting deviations. Its calculation method is as follows: first calculate the absolute deviation between the predicted value and the target value of each water quality parameter at the short-term prediction time point; then multiply the deviation of each parameter by the corresponding parameter importance coefficient , representing the weights of the different parameters' impact on system performance. Finally, all weighted deviations are summed to obtain the overall short-term objective function value. A smaller value indicates higher short-term water quality control accuracy.

[0124] The medium-term objective function focuses on parameter stability. The calculation method is as follows: first, the variance of each water quality parameter in the medium-term prediction period is calculated. The variance value represents the degree of fluctuation of the parameter in the period; then the variance of each parameter is multiplied by the corresponding stability weight coefficient. , reflecting the differences in stability requirements for different parameters. Finally, all weighted variances are summed to obtain the overall medium-term objective function value. The smaller this value, the higher the stability of the medium-term water quality parameters.

[0125] The long-term objective function optimizes resource utilization efficiency. The calculation method is as follows: first calculate the cost of pharmaceuticals, energy consumption costs, and equipment maintenance costs respectively; then multiply these three types of costs by the corresponding cost weight coefficients. 、 and , reflecting the importance of different cost types. Finally, the weighted costs are summed to obtain the overall long-term objective function value. The smaller the value, the more efficient the resource utilization.

[0126] The short-term objective function focuses on parameter deviation, the medium-term objective function focuses on parameter variance, and the long-term objective function focuses on cost-effectiveness. These indicators have different dimensions and numerical ranges. Before performing weighted sum calculations, the objective function values ​​for the three time scales need to be normalized. This can be done by using proportional normalization relative to historical typical values ​​or Z-score standardization to ensure comparability of objectives across different time scales in the comprehensive evaluation.

[0127] Parameter importance coefficient The specific implementation involves analyzing data from historical water quality issues, calculating the correlation between each parameter and the overall system performance and safety, combining expert knowledge to set basic weights, and dynamically adjusting them based on the current system operating status. Key safety indicators (such as pH and residual chlorine) are given higher weights, while parameters with less impact are given lower weights.

[0128] Stability weight coefficient The specific implementation is based on the allowable fluctuation range of each parameter and its impact on system stability. Parameters that are sensitive to fluctuations (such as microbiological indicators) are given a higher weight, while parameters that allow a certain fluctuation range (such as conductivity) are given a lower weight. The weights are calculated by analyzing the differences between historical stable operating data and problem data, and fine-tuned based on operational experience.

[0129] Cost weight coefficient 、 and The specific implementation involves determining the relative importance of three cost types (chemical costs, energy consumption, and equipment maintenance costs) based on the primary system optimization goal and user needs. These weights are dynamically adjusted based on different system states and operational stages. For example, the weighting of energy costs is increased during high-energy consumption seasons, maintenance costs during equipment aging, and chemical costs during periods of rising chemical costs. Weight adjustments are based on system status monitoring, cost analysis, and projected economic benefit calculations.

[0130] The comprehensive evaluation function integrates the objective functions of the above three time scales through the weighted sum method. The calculation method is: multiply the short-term objective function value by the short-term weight coefficient , the mid-term objective function value multiplied by the mid-term weight coefficient , the long-term objective function value multiplied by the long-term weight coefficient , and then the three are added together to obtain a comprehensive evaluation value. The weight coefficient is dynamically adjusted according to the system status, increasing the long-term weight when the system is operating stably and increasing the short-term weight when facing potential risks, to balance the relationship between immediate response and long-term optimization.

[0131] Step 4.3, optimization algorithm execution;

[0132] The method provided in this application uses an improved particle swarm optimization algorithm to solve multi-objective optimization problems. The algorithm incorporates improvements specifically tailored to the characteristics of the drug ratio problem, including handling discrete variables, constraint handling, and enhanced local search strategies. The algorithm outputs an optimal drug ratio solution, including the dosage and timing of each drug.

[0133] The improved particle swarm optimization algorithm in this application has the following characteristics:

[0134] Mixed coding strategy: Use continuous-discrete mixed coding scheme, where continuous variables represent the amount of drug added and discrete variables represent the timing of addition.

[0135] Constraint processing module: Adopts adaptive penalty function method to process hard constraints and soft constraints to ensure the feasibility of the solution.

[0136] Multi-population co-evolution: maintain multiple sub-populations for parallel search, each sub-population focuses on a different optimization direction, and shares high-quality solutions through regular information exchange.

[0137] Local search enhancement: Apply local optimization techniques such as pattern search to promising solutions to accelerate the convergence process.

[0138] Adaptive parameter adjustment: Dynamically adjust algorithm parameters (inertia weight, learning factor, etc.) according to search progress to balance global exploration and local development.

[0139] The algorithm execution process is as follows:

[0140] Initialize multiple subpopulations, each subpopulation targeting different optimization objectives;

[0141] Evaluate the objective function value and constraint violation degree for each particle; update the individual optimal solution and the global optimal solution; adjust the particle position and velocity according to the update rule; apply local search to promising solutions; exchange information between sub-populations; dynamically adjust algorithm parameters; repeat until the termination condition is met; output the Pareto optimal solution set and select the final solution based on the decision maker's preferences.

[0142] like Figure 3 The figure shows the Pareto optimal solution set obtained by the multi-objective optimization algorithm of the present invention. Each point in the figure represents a feasible reagent ratio scheme. The horizontal axis represents system resource consumption (including reagent cost and energy consumption), and the vertical axis represents system stability (the inverse of the degree of fluctuation of water quality parameters). The curve in the figure is the Pareto front, which shows the trade-off between system resource consumption and system stability, verifying the effectiveness of the multi-objective optimization algorithm of the present invention. Decision makers can select an appropriate balance point on this frontier as the final control solution based on actual needs.

[0143] Step 5: Based on the optimized drug ratio strategy, a multi-time scale collaborative optimization system is implemented to further balance short-term response and long-term stability;

[0144] This step builds a multi-level control architecture to coordinate control strategies at different time scales, enabling the system to respond quickly to short-term disturbances while maintaining long-term stability. Specifically, it includes:

[0145] Step 5.1, constructing a hierarchical control framework;

[0146] Establish a three-tier control architecture consisting of an emergency response layer, a daily control layer, and a strategy optimization layer. The emergency response layer handles sudden anomalies with a response time of minutes; the daily control layer performs forecast-based feedforward control with a time scale of hours; and the strategy optimization layer is responsible for long-term parameter optimization with a time scale of days or weeks.

[0147] Step 5.2, inter-layer coordination system;

[0148] Build an inter-layer information transmission and decision coordination system to ensure that control decisions at different time scales are compatible and consistent. The upper layer provides constraints and guidance to the lower layer, while the lower layer provides execution feedback and status updates to the upper layer.

[0149] Step 5.3, dynamically adjust the module;

[0150] Based on system status and forecast results, the activation priorities of control layers at different time scales are dynamically adjusted. During normal operation, long-term optimization strategies take the lead; when facing potential risks, medium-term control strengthens intervention; and in the event of an emergency, the short-term response layer takes over the system.

[0151] It should be noted that, in some embodiments, the method provided in the present application may further provide a transition buffer module between each control layer to ensure a smooth transition of control strategy switching between layers and avoid control oscillation.

[0152] like Figure 4 As shown in the figure, the performance difference between the method of the present invention and the traditional control method is comprehensively evaluated from the five dimensions of prediction accuracy, response speed, resource efficiency, environmental adaptability and system stability. The blue polygons in the figure represent the performance of the multi-time scale collaborative optimization system of the present invention, and the red polygons represent the performance of the traditional control method. It can be clearly seen from the figure that the method of the present invention is superior to the traditional method in all five dimensions, especially in prediction accuracy, environmental adaptability and system stability, which fully demonstrates the comprehensive performance advantages of the method and the effect of multi-time scale collaborative optimization.

[0153] Step 6: Based on the optimization results of the multi-timescale collaborative optimization system, an environment-adaptive active evolution system is constructed to enable the system to automatically adjust to environmental changes.

[0154] This step enables the system to autonomously adapt to environmental changes, continuously optimize models and control strategies, and form a true closed-loop optimization system. Specifically, it includes:

[0155] Step 6.1, model performance monitoring and evaluation;

[0156] Continuously monitor the performance of the forecast model and calculate the forecast error metric by comparing the forecast value with the actual observation value. When the error exceeds the preset threshold or shows a continuous growth trend, the model update process is triggered.

[0157] In terms of specific implementation, this system adopts a multi-index comprehensive evaluation method, including:

[0158] Root mean square error: evaluates the average deviation between the predicted value and the actual value;

[0159] Mean absolute percentage error: evaluates relative prediction error;

[0160] R² coefficient of determination: assesses how well the model explains the variation in the data;

[0161] Cumulative sum of forecast deviations: Detects systematic deviations in forecasts;

[0162] Critical event prediction rate: evaluates the model's ability to predict important events (such as parameter mutations).

[0163] Metrics such as root mean square error, mean absolute percentage error, R² coefficient of determination, cumulative sum of prediction deviations, and critical event prediction rate have different numerical ranges and meanings. When setting trigger thresholds and conducting comprehensive evaluations, these metrics need to be standardized to a unified scoring range (e.g., 0-1). Percentile conversion based on historical performance distribution or fuzzy logic scoring methods can be used to enable comprehensive comparison of different performance metrics.

[0164] The specific implementation of the cumulative sum of forecast deviations is to calculate the signed cumulative value of all forecast deviations within a time window, rather than simply summing up the absolute errors. This method can detect the systematic deviation trend of the forecast model, such as continuous overestimation or underestimation of a parameter. When calculating, first obtain the continuous The difference between the predicted value and the actual value at each time point is calculated, retaining the positive and negative signs. The cumulative sum of these differences is then calculated. Finally, the cumulative sum is divided by the number of time points to obtain the standardized PBCS value. When the absolute value of the PBCS exceeds the preset threshold and shows a monotonically increasing trend, it indicates that the model has systematic deviations and the model adjustment mechanism needs to be triggered.

[0165] The system sets two levels of trigger thresholds: warning threshold and update threshold. When the performance index exceeds the warning threshold, the system increases the monitoring frequency; when it exceeds the update threshold or continuously When the value remains above the warning threshold for a certain period of time, the model update process is triggered.

[0166] Step 6.2, incremental learning and model updating;

[0167] An incremental learning approach is used to update the time series knowledge graph and prediction model without the need for full retraining. New data is first used to update the graph structure and then to fine-tune the prediction model parameters to maintain the model's sensitivity to the latest data.

[0168] The incremental learning system in this application consists of three key components:

[0169] Selective memory module: evaluates the information value of new data, screens high-value samples to add to the training set, and removes redundant or outdated samples to maintain the representativeness and timeliness of the training data.

[0170] Elastic Weight Adjustment Algorithm: Parameter importance is calculated based on the Fisher information matrix, imposing stronger regularization constraints on key parameters to protect learned key knowledge while allowing the model to adapt to new patterns. Water quality parameters are first divided into fast-changing parameter groups (pH, residual chlorine) and slow-changing parameter groups (conductivity, hardness). For each group, the second-order derivative matrix of the log-likelihood function with respect to the model parameters is calculated and weighted to form an overall Fisher matrix. The calculation specifically considers the sensitivity of parameters under different operating conditions (high temperature, low temperature, high load, etc.), and assigns higher weights to key points based on historical water quality fluctuations. When the system detects seasonal changes or sudden changes in raw water quality, the Fisher matrix guides the model to retain key knowledge (such as the long-term correlation between pH and scaling risk) while allowing for adaptive updates (such as the seasonal relationship between microbial activity and temperature). The knowledge distillation framework builds a circulating water treatment expert model library and uses empirical models (such as a model trained with 10 years of operating data from a chemical plant's cooling water system) as a teacher network to guide the new model's predictive behavior under different water quality conditions. This ensures that the model can quickly adapt to the latest water quality changes (such as the system response characteristics after adding corrosion inhibitors) while retaining the stable relationship pattern between the original water quality parameters (such as the mapping relationship between conductivity and corrosion rate), thereby ensuring the continued effectiveness of the circulating water system prediction model.

[0171] Knowledge distillation framework: Maintain a teacher model to preserve historical knowledge. While fitting new data, the new model minimizes the deviation from the teacher model's prediction to achieve a balance between new and old knowledge.

[0172] The update process adopts a phased strategy: first, the graph structure is updated, including adding / removing edges and adjusting edge weights; then the graph structure is fixed, and node representations and prediction model parameters are updated; finally, global fine-tuning is performed to integrate all components.

[0173] like Figure 5 The figure shows a comparison of the accuracy trends of the incremental learning model and the traditional fixed model over long-term use. The blue area in the figure represents the accuracy of the incremental learning model of the present invention, and the red area represents the accuracy of the traditional fixed model. The horizontal axis represents the system operation time (in months). As can be clearly seen from the figure, the accuracy of the traditional fixed model shows a clear downward trend over time, while the incremental learning model of the present invention is able to maintain a high level of accuracy and quickly restore performance through self-updates after changes in system characteristics (such as seasonal changes). This figure demonstrates the effectiveness of the incremental learning and model update mechanisms of the present invention, intuitively demonstrating the system's self-evolutionary capabilities and providing a guarantee for long-term stable operation.

[0174] Step 6.3, seasonal pattern adaptation;

[0175] Identify and learn seasonal variations in water quality parameters, building specialized knowledge subgraphs for each season. The system automatically switches and blends appropriate models and strategies based on seasonal and environmental conditions, achieving consistent performance year-round.

[0176] The seasonal pattern adaptation system includes the following core functions:

[0177] Seasonal pattern recognition: Detecting periodic patterns in time series using Fourier analysis and wavelet transforms, automatically identifying seasonal cycles and characteristics;

[0178] Seasonal Feature Extraction: A feature representation is constructed for each identified seasonal cycle, capturing the typical characteristics and relationship patterns of water quality parameters within that season. Seasonal feature extraction utilizes a multi-level feature analysis approach within the circulating water treatment system. First, wavelet transforms are used to decompose water quality time series data into different frequency components to identify seasonal cycles. Dimensionality reduction techniques are then applied to extract seasonal discriminant feature vectors from the high-dimensional water quality parameter space, including temperature-pH correlation features and conductivity-microbial activity joint features. A seasonal feature extraction template library is established based on the characteristics of the circulating water system: the summer template focuses on extracting feature combinations such as accelerated microbial growth, increased pH fluctuations, and increased residual chlorine consumption rates under conditions of elevated water temperature; the winter template focuses on features such as changes in scaling tendency and decreased microbial activity under low temperature conditions; and the spring-autumn transition season extracts turbidity fluctuations caused by rainfall changes. The system determines the current operating season mode by calculating the similarity between the current water quality parameter sequence and the templates for each season, and automatically calls the corresponding prediction model and control strategy. For example, in summer mode, the system proactively increases the frequency of biocide addition, adjusts the pH target range to 7.2-7.8, and adjusts the sewage discharge cycle based on the water temperature-microorganism relationship. In winter mode, the system optimizes the scale inhibitor formulation and adjusts the corrosion inhibitor to dispersant ratio to 2:1 to address the scaling risk changes under low temperature conditions, thus achieving intelligent seasonal regulation of the circulating water treatment system.

[0179] Seasonal sub-model library: Build specialized prediction models and control strategies for different seasonal conditions to form a model library;

[0180] Smooth transition system: Build a gradual switching algorithm during the seasonal transition period to achieve smooth model transition and avoid sudden changes in control strategies;

[0181] Seasonal omen detection: Based on environmental parameters and early indicators, seasonal changes are predicted in advance, and corresponding models are preset to achieve seamless adaptation.

[0182] In addition, in some embodiments, the method provided in this application can also be combined with historical annual data to construct a long-term memory library of seasonal patterns to further improve the accuracy of seasonal predictions.

[0183] like Figure 6As shown, the relationship between water temperature and microbial activity index in the circulating water system as they change with the seasons is shown. The bar graph in the figure represents the monthly average water temperature changes, and the line graph represents the corresponding microbial activity index changes. As can be seen from the figure, the microbial activity index and water temperature show an obvious seasonal correlation pattern, but the relationship is not a simple linear correspondence, reflecting the influence of complex seasonal factors in the system. This figure verifies the application scenario of the system's seasonal pattern adaptive function, intuitively demonstrates the necessity of the environmental adaptation-active evolution system, and proves that the present invention can effectively cope with changes in system characteristics brought about by seasonal changes.

[0184] Step 6.4, control effect feedback and optimization;

[0185] Record each control operation and its effects to build a control-response database. Regularly analyze control effects to evaluate the effectiveness of control strategies, and use reinforcement learning methods to optimize control strategies and improve overall system performance.

[0186] The control effect feedback and optimization module adopts a model-based reinforcement learning framework, which includes:

[0187] Control-response database: records each control operation, system status and response effect, and builds a structured database;

[0188] Effect evaluation engine: Comprehensively evaluates control effects through multi-dimensional indicators (water quality stability, resource efficiency, energy consumption, etc.);

[0189] Policy Optimizer: Based on a double-delayed deep deterministic policy gradient algorithm, it learns the optimal control policy from historical control data;

[0190] Simulation verification environment: Build a simulation environment based on the time-series knowledge graph to verify the effectiveness and security of new strategies;

[0191] Progressive deployment system: adopts a conservative strategy deployment plan, first tests new strategies in a limited scope, and then gradually expands the application scope after verification.

[0192] Through this closed-loop feedback and optimization system, the system can continuously learn from actual operating experience, continuously optimize control strategies, and form a truly environmentally adaptive and actively evolving intelligent control system.

[0193] like Figure 7The figure shows a direct comparison of the performance differences between traditional water quality control methods and the method of the present invention in terms of three key indicators: prediction accuracy, advance intervention time, and drug use efficiency. The blue columns in the figure represent traditional methods, and the green columns represent the method of the present invention. It can be clearly seen from the figure that the method of the present invention improves prediction accuracy by approximately 45%, increases advance intervention time from 0 hours (passive response) in traditional methods to an average of 12 hours (active prevention), and improves drug use efficiency by approximately 30%. These data intuitively demonstrate the technical advantages of the present invention, verify the actual performance of the above-mentioned technical effects, and fully demonstrate the innovative value and application potential of the method of the present invention in the field of circulating water treatment.

[0194] A circulating water treatment closed-loop optimization system based on water quality prediction, used to execute the above-mentioned circulating water treatment closed-loop optimization method based on water quality prediction, comprising:

[0195] The time series knowledge graph module is used to represent the time series evolution characteristics of water quality parameters and their relationships;

[0196] A multi-timescale prediction engine for predicting future water quality status based on a time-series knowledge graph;

[0197] A predictive constraint generator to convert future water quality predictions into current decision constraints;

[0198] Multi-objective optimization module, used to optimize the drug ratio strategy based on the prediction results and constraints;

[0199] A multi-timescale collaborative control module to balance short-term response and long-term stability;

[0200] The environment adaptation active evolution module is used to enable the system to automatically adjust to environmental changes.

[0201] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A closed-loop optimization method for circulating water treatment based on water quality prediction, characterized in that: include: Construct a time-series knowledge graph that supports the time dimension to represent the temporal evolution characteristics of water quality parameters and their relationships; Based on the time series knowledge graph, a multi-time scale prediction engine is built to predict future water quality status; Develop a predictive constraint generator based on future water quality state prediction results to convert future water quality predictions into current decision constraints; Introducing a time-space sensitive multi-objective optimization algorithm to optimize the drug ratio strategy based on the prediction results and current decision constraints; Based on the optimized drug ratio strategy, a multi-time scale collaborative optimization system is realized to further balance short-term response and long-term stability; Based on the optimization results of the multi-time-scale collaborative optimization system, an environmentally adaptive active evolutionary system is constructed to enable the system to automatically adjust as the environment changes.

2. A closed-loop optimization method for circulating water treatment based on water quality prediction according to claim 1, characterized in that: The steps of constructing a time series knowledge graph supporting the time dimension include: The water quality parameters are defined as entity nodes in the graph, each node contains parameter name, value, and timestamp attributes; Apply the relationship discovery algorithm based on Granger causality test to identify the potential causal relationships between parameters and represent the potential causal relationships as directed edges in the graph; A temporal pattern learning method based on graph neural network is used to extract the temporal evolution patterns of water quality parameters from historical data.

3. The closed-loop optimization method for circulating water treatment based on water quality prediction according to claim 1, characterized in that: The multi-timescale prediction engine is built based on a hybrid architecture of graph convolutional networks and gated recurrent units, including: Spatial feature extraction module, used to capture the spatial dependencies between nodes; Time series feature extraction module, used to capture time series characteristics; Attention fusion layer, used to fuse spatial features and temporal features; The multi-timescale prediction head includes three parallel branches: short-term prediction, medium-term prediction, and long-term prediction.

4. A closed-loop optimization method for circulating water treatment based on water quality prediction according to claim 3, characterized in that: The multi-time-scale prediction engine also introduces Monte Carlo dropout technology to provide a confidence interval for each prediction result. Through random dropout forward propagation, the prediction distribution is obtained, the mean is calculated as the final prediction value, and the variance is used as the uncertainty measure.

5. The closed-loop optimization method for circulating water treatment based on water quality prediction according to claim 1, characterized in that: The predictive constraint generator comprises: Key threshold and target interval definition module, which defines the target range and warning threshold for each water quality parameter; The prediction trajectory risk assessment module performs risk analysis on the predicted water quality parameter trajectory, taking into account parameter importance, deviation degree and probability of crossing the boundary; The dynamic constraint generation module generates constraint conditions for control decisions based on risk assessment results.

6. The closed-loop optimization method for circulating water treatment based on water quality prediction according to claim 1, characterized in that: The time-space sensitive multi-objective optimization algorithm includes: Construct a reagent ratio model and a response model to describe the relationship between the reagent addition amount and water quality parameters; Construct objective functions for the short-term, medium-term, and long-term time scales, focusing on response speed, stability, and cost-effectiveness respectively; An improved particle swarm optimization algorithm is used to solve multi-objective optimization problems using continuous-discrete hybrid coding, adaptive penalty function and multi-swarm co-evolution strategy.

7. The closed-loop optimization method for circulating water treatment based on water quality prediction according to claim 1, characterized in that: The multi-timescale collaborative optimization system includes: Establish a three-tier control structure consisting of emergency response layer, daily control layer, and strategy optimization layer; Build an inter-layer information transmission and decision coordination system to ensure that control decisions at different time scales are compatible and consistent; According to the system status and prediction results, the activation priority of control layers at different time scales is dynamically adjusted.

8. The closed-loop optimization method for circulating water treatment based on water quality prediction according to claim 1, characterized in that: The environmental adaptation active evolution system includes: Model performance monitoring and evaluation module, which continuously monitors the performance of the prediction model and calculates the prediction error index by comparing the predicted value with the actual observed value; The incremental learning and model update module uses incremental learning methods to update the time series knowledge graph and prediction model; Seasonal pattern adaptation module, which identifies and learns seasonal variation patterns of water quality parameters; The control effect feedback and optimization module records control operations and their effects, and uses reinforcement learning methods to optimize control strategies.

9. A closed-loop optimization method for circulating water treatment based on water quality prediction according to claim 8, characterized in that: The incremental learning and model update module includes three key components: a selective memory module, an elastic weight adjustment algorithm, and a knowledge distillation framework, which are used to evaluate the value of new data, protect key knowledge, and balance new and old knowledge.

10. A closed-loop optimization system for circulating water treatment based on water quality prediction, characterized in that: A closed-loop optimization method for circulating water treatment based on water quality prediction for executing any one of claims 1 to 9, comprising: The time series knowledge graph module is used to represent the time series evolution characteristics of water quality parameters and their relationships; A multi-timescale prediction engine for predicting future water quality status based on a time-series knowledge graph; A predictive constraint generator to convert future water quality predictions into current decision constraints; Multi-objective optimization module, used to optimize the drug ratio strategy based on the prediction results and constraints; A multi-timescale collaborative control module to balance short-term response and long-term stability; The environment adaptation active evolution module is used to enable the system to automatically adjust to environmental changes.

Images