Method for regulating water quality in land-based recirculating aquaculture

Abstract

This invention discloses a method for regulating water quality in land-based recirculating aquaculture systems, relating to the field of intelligent aquaculture management. The invention constructs a spatiotemporally synchronized multidimensional dynamic data stream by deploying a multi-source sensor network, enabling comprehensive real-time sensing of water quality parameters. It combines rigid threshold early warning with a support vector machine model to identify hidden abnormal fluctuation risks in advance. The invention utilizes a K-means clustering algorithm to trace and prioritize related abnormal root causes, generating automated control commands based on a pre-set structured strategy library. Through iterative closed-loop optimization and evaluation, it ensures stable water quality compliance. Simultaneously, it accumulates historical cases to drive the co-evolution of the model and strategy library. This method solves the problems of incomplete sensing, delayed early warning, difficulty in tracing root causes, inefficient control, and insufficient adaptability inherent in traditional methods, achieving precise, efficient, and sustainable water quality regulation.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for aquaculture, specifically a method for regulating water quality in land-based recirculating aquaculture. Background Technology

[0002] Water quality management in aquaculture is a crucial foundation for ensuring aquatic ecological balance and aquaculture efficiency, directly impacting food safety and sustainable industry development. With the continuous expansion of aquaculture scale, water quality control has become a key aspect of industry development, its importance self-evident. A good water quality environment not only improves the growth rate and health of farmed organisms but also effectively reduces disease risks, playing a decisive role in the stable operation of the entire industry chain. However, current water quality management methods have significant shortcomings in dynamic adaptability and comprehensive control capabilities. Many traditional methods often struggle to cope with complex changes in the aquatic environment, especially in the face of sudden pollution or rapid fluctuations in water quality parameters. They lack timely detection and precise intervention capabilities, meaning water quality problems are often only discovered after they have worsened, missing the optimal control opportunity and leading to increased aquaculture losses.

[0003] Against this backdrop, the core technical challenges of water quality management are becoming increasingly apparent, particularly in the synergy of real-time multi-parameter sensing and control methods. A primary problem is the difficulty in comprehensively capturing the dynamic changes of multiple key water indicators, such as dissolved oxygen, pH, and the concentration of harmful substances. The interactions between different parameters increase the complexity of monitoring. This complexity further evolves into another challenge: how to quickly match appropriate control strategies based on the fluctuation characteristics of different parameters, building upon the sensing capabilities. For example, in actual aquaculture, when dissolved oxygen suddenly drops while the concentration of harmful substances rises, if the correlation between the two cannot be determined in time and targeted measures cannot be taken, it may lead to an overall imbalance in water quality, or even cause large-scale fish mortality. Summary of the Invention

[0004] The purpose of this invention is to provide a method for regulating water quality in land-based recirculating aquaculture, which realizes fully automated management of the entire process from water quality problem discovery, root cause location, strategy matching, command execution to problem eradication, thereby improving the accuracy, timeliness and sustainability of water quality regulation.

[0005] The objective of this invention can be achieved through the following technical solutions: This application provides a method for regulating water quality in land-based recirculating aquaculture systems, comprising the following steps: S1. Deploy a sensor network in the aquaculture system to collect water quality monitoring data, equipment operating status data, and operation record data, and perform time alignment and fusion to output a multi-dimensional dynamic data stream; S2. Based on rigid water quality thresholds, perform real-time comparison and preliminary early warning of the multidimensional dynamic data stream, and use a support vector machine model to identify abnormal fluctuations that do not exceed the thresholds, and output the abnormal fluctuation feature judgment results. S3. When it is found that the decrease in dissolved oxygen is associated with the increase in the concentration of harmful substances, the K-means clustering algorithm is used to perform group analysis by combining the operating status and operation record data, and output the priority control groups that represent different root causes. S4. Based on the priority control group, map strategies from the preset structured control strategy library to generate a targeted control instruction sequence; S5. Issue and execute the targeted control command sequence, collect water quality parameters again and compare them with the preset standard range to generate preliminary water quality adjustment results; S6. If the preliminary water quality adjustment results do not meet the stability conditions, the anomaly identification and root cause analysis steps are repeated based on the adjusted data stream, and supplementary strategies are matched for iterative regulation until the water quality is stable. S7. Continuously record abnormal event data to form a historical case library, and optimize the support vector machine model, clustering rules and policy matching logic based on the library to achieve the co-evolution of the diagnostic model and the policy library.

[0006] The beneficial effects of this invention are as follows: This invention addresses the shortcomings of traditional methods in capturing and sensing the dynamic changes of multiple parameters in water bodies in a comprehensive and timely manner by deploying a multi-source sensor network and constructing a spatiotemporally synchronized multi-dimensional dynamic data stream. Through the coordinated monitoring and fusion analysis of key indicators such as dissolved oxygen and ammonia nitrogen, it achieves comprehensive and real-time perception of water quality status, laying a reliable data foundation for subsequent precise regulation. By combining rigid threshold early warning with support vector machine model recognition, it effectively solves the problem of delayed detection of hidden water quality problems that have not exceeded the threshold but have already shown abnormal fluctuation characteristics. It can identify potential risks in advance, transforming passive response into proactive early warning, significantly extending the regulation response window, and avoiding the accumulation and deterioration of water quality problems. By using the K-means clustering algorithm to group and prioritize correlated anomalies, the problem of difficulty in tracing the root causes and the lack of targeted control measures caused by the mutual influence of multiple parameters is solved. By decomposing complex correlation phenomena such as the decrease in dissolved oxygen and the increase in the concentration of harmful substances into clear root cause patterns and assigning control priorities, precise decision-making from symptomatic treatment to fundamental treatment is achieved. Through the use of a pre-built structured strategy library for targeted strategy mapping and intelligent instruction generation, the problems of traditional reliance on human experience, slow matching of control strategies and easy errors are solved. It can automatically match, combine and optimize equipment control and operation adjustment instructions according to different root causes, and realize the rapid, accurate and automated execution of control strategies. By implementing a closed-loop optimization mechanism of evaluation iteration, the problem of potentially incomplete single-stage regulation and difficulty in adapting to complex dynamic changes is solved. Based on feedback from the initial regulation effect, the remaining imbalance factors can be automatically identified and supplementary strategies can be matched for iterative regulation until the water quality stabilizes and meets the standards, significantly improving the success rate and final effect of regulation in complex scenarios. By continuously accumulating historical cases and driving the co-evolution of the model and strategy library, the problem of traditional static systems being unable to adapt to long-term changes in the aquaculture environment and new abnormal patterns is solved. By continuously optimizing the diagnostic model and regulation logic using historical data, the perception accuracy, diagnostic accuracy, and regulation effectiveness of the entire system can autonomously improve over time, possessing sustainable optimization capabilities. Attached Figure Description

[0007] To better understand and implement this application, the technical solution is described in detail below with reference to the accompanying drawings.

[0008] Figure 1 A schematic flowchart illustrating a method for regulating water quality in land-based recirculating aquaculture systems provided in Embodiment 1 of this application; Figure 2 A flowchart illustrating step S3 in a method for regulating water quality in land-based recirculating aquaculture provided in Embodiment 1 of this application; Figure 3 This is a flowchart illustrating step S4 of a land-based recirculating aquaculture water quality control method provided in Embodiment 1 of this application. Detailed Implementation

[0009] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, exemplary embodiments will be described in detail below, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0010] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0011] The following detailed description of the specific implementation methods, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided in detail.

[0012] Example 1 Please see Figures 1-3This embodiment provides a method for regulating water quality in land-based recirculating aquaculture systems, comprising the following steps: S1. Deploy a water quality sensor network in the breeding ponds, water storage tanks and tailwater treatment units of the aquaculture system to collect real-time monitoring data of dissolved oxygen, pH, ammonia nitrogen and nitrite, and simultaneously collect the operating status data of related equipment and the operation record data of feeding and water changing. Time-align and fuse the monitoring data, operating status data and operation record data to output a multi-dimensional dynamic data stream.

[0013] Furthermore, according to step S1, the specific steps include: Sensor nodes are deployed in aquaculture ponds, water storage ponds and effluent treatment units to collect real-time monitoring data on dissolved oxygen, pH, ammonia nitrogen and nitrite. Simultaneously, the operating status data of related equipment and operation records of feeding and water changing are collected. Based on the physical location of the data collection points, spatial partition identifiers are added to all incoming raw data to generate an initial partitioned data stream with clear spatial attributes. All monitoring data, equipment status data, and operation record data are stamped with a unified high-precision timestamp. Through the timestamp-driven event alignment engine, multi-source heterogeneous data are matched and calibrated in real time to eliminate timing deviations caused by sampling frequency or transmission delay, and to fuse and generate a standardized multi-dimensional data stream that is strictly synchronized in time and space. Based on the standardized multidimensional data stream, a sliding time window is used for continuous rolling calculation. Within each window, the key water quality parameter sequence is statistically aggregated in real time, and short-term statistical features including mean, standard deviation and trend are extracted. The original data stream is transformed into a feature vector sequence that reflects the dynamic state of the system in real time, and a multidimensional dynamic data stream is output.

[0014] Furthermore, a timestamp-driven event alignment engine performs real-time matching and calibration of multi-source heterogeneous data. Specifically, this includes: assigning millisecond-level timestamps based on the same reference clock to data streams from different sensor nodes, device controllers, and operating terminals; using the water quality monitoring data stream as the main timeline, and accurately associating device operating status signals with operational events such as feeding and water changing to the corresponding time points according to the set time window thresholds; dynamically completing data points missing due to different acquisition frequencies using linear interpolation algorithms to ensure the continuity of each parameter in the time dimension; and finally, resolving potential numerical inconsistencies in multi-source data through conflict detection and priority arbitration mechanisms, outputting a standardized multi-dimensional data stream that is strictly synchronized in time and space.

[0015] Furthermore, a sliding time window is used for continuous rolling calculations. Specifically, this includes: setting analysis windows with fixed durations and step sizes, and extracting time slice data from the standardized data stream; calculating the statistical mean of key water quality parameters such as dissolved oxygen and ammonia nitrogen within each window to reflect concentration levels, calculating the standard deviation to assess the degree of fluctuation, and obtaining the slope of the trend through linear regression analysis; simultaneously calculating the extreme value distribution and coefficient of variation of each parameter to construct a feature set containing multi-dimensional statistical features; and finally organizing the feature set of each time window into a continuous feature vector sequence in chronological order to form a multi-dimensional dynamic data stream that can dynamically characterize the evolution of water quality status.

[0016] Specifically, by deploying a multi-source sensor network in the aquaculture system and performing high-precision time alignment and spatial labeling on the collected water quality, equipment, and operational data, the core problems of fragmented monitoring information caused by dispersed multi-parameter sensing, heterogeneous data sources, and lack of a unified spatiotemporal benchmark in traditional water quality management are solved. By establishing a standardized data flow and sliding window feature extraction mechanism, a multi-dimensional data foundation capable of comprehensively, continuously, and structurally representing the dynamic state of the system is constructed, enabling holistic real-time perception of changes in multiple elements of the aquaculture environment and providing high-quality data support for subsequent intelligent diagnosis and precise control.

[0017] S2. Based on the rigid water quality threshold of the aquaculture regulations, the multidimensional dynamic data stream is compared in real time and a preliminary warning is given. A trained support vector machine model is used to identify water quality time series data that does not exceed the threshold but has preset abnormal fluctuation characteristics, and the abnormal fluctuation characteristic judgment result is output.

[0018] Furthermore, according to step S2, the specific steps include: Rigid water quality thresholds are set according to aquaculture regulations and standards. Each water quality parameter in the multidimensional dynamic data stream is compared in real time to determine whether it exceeds the threshold. If it exceeds the threshold, a primary warning signal is generated immediately. At the same time, the water quality time series data that did not trigger the primary warning are separated and their complete time series is retained for subsequent analysis. For the retained water quality time series data that did not exceed the threshold, a trained support vector machine model was used to identify abnormal fluctuation patterns, output binary classification judgment results, and then the water quality time series data segments that were judged to be abnormal fluctuations were selected, and the start and end timestamps of each abnormal segment were marked. By scanning the marked abnormal segments through a sliding window, the magnitude and rate of change of water quality parameters within each segment are calculated to determine the intensity level of abnormal fluctuations. If the intensity level of fluctuations reaches the preset conditions, a secondary warning signal is generated and associated with the time series data of the corresponding abnormal segment. The continuity of the secondary warning signal is verified by rule matching to determine whether there is overlap between abnormal segments in adjacent time windows. If there is overlap, they are merged into a continuous abnormal event, and the structured abnormal fluctuation feature judgment result is integrated and output.

[0019] Furthermore, a pre-trained support vector machine (SVM) model is used to identify abnormal fluctuation patterns, outputting binary classification results. Then, water quality time-series data segments identified as abnormal fluctuations are selected, and the start and end timestamps of each abnormal segment are marked. Specifically, this involves: constructing feature vectors from the water quality time-series data that did not trigger a primary warning in the multidimensional dynamic data stream using short-term statistical features such as mean, standard deviation, and trend slope extracted by a sliding window; and inputting these vectors into a pre-trained SVM classification model for inference in real time. This model, trained based on historical normal and abnormal water quality fluctuation patterns, can identify composite anomalies exceeding rigid thresholds and outputs a binary label of "normal" or "abnormal" for each input window. Based on the continuous label sequence output by the model, all continuous time windows identified as "abnormal" are automatically located. By scanning their time indices, the start and end timestamps of each abnormal fluctuation event are precisely marked, forming independent abnormal segments with clear time boundaries. Subsequently, each abnormal segment is associated with the complete original water quality curve, equipment status change logs, and operation records within its occurrence time period, forming an enhanced abnormal event data object, providing a complete context for subsequent analysis.

[0020] Furthermore, by scanning the marked abnormal segments using a sliding window, the magnitude and rate of change of water quality parameters within each segment are calculated to determine the intensity level of the abnormal fluctuations. Specifically, this includes: for each marked abnormal segment, a second sliding window scan is performed within its time boundary at a higher temporal resolution to calculate the magnitude and rate of change of key water quality parameters within each high-resolution window in real time; based on this, the calculation results of all windows within the abnormal segment are aggregated and analyzed to extract multi-dimensional intensity features such as the maximum and average values ​​of the magnitude of change, the peak value of the rate of change, and the proportion exceeding the baseline threshold; according to the preset intensity grading rules, the above aggregated features are mapped to discrete intensity levels, such as "low", "medium", and "high", and a "intensity-parameter-context" correlation map is constructed simultaneously to clearly identify the main water quality parameter types that cause the anomaly and the associated operational or equipment factors, providing a quantitative basis for root cause localization and the generation of control strategies.

[0021] Furthermore, a rule-based matching method is used to verify the continuity of the secondary warning signals, determining whether there is overlap in abnormal segments within adjacent time windows. Specifically, this includes maintaining a real-time updated list of secondary warning events; when a new warning event is added, its time interval is automatically compared with the time intervals of warning events in the same monitoring unit that are not closed in the list, and the overlap duration is calculated. If the overlap exceeds a preset threshold, these events are determined to have temporal continuity and belong to different stages of the same persistent anomaly. Subsequently, the system integrates the continuous events according to the merging rules to generate a unified persistent anomaly event record. This record covers the total time span after merging, the evolution sequence of intensity levels at each stage, and all associated contextual data. At the same time, the original independent events are updated to sub-events of the persistent event, realizing the structured representation and tracking of long-cycle, recurring anomalies.

[0022] The trained support vector machine model includes: a model trained on multi-dimensional water quality time-series data collected during historical aquaculture cycles, with training data covering normal water quality fluctuation patterns and various preset abnormal patterns, including but not limited to stepwise decreases in dissolved oxygen, covert cumulative increases in ammonia nitrogen, periodic abnormal oscillations in pH, and atypical collaborative change patterns among multiple parameters; the model uses radial basis functions as kernel functions, and optimizes the optimal penalty coefficient and kernel parameters through grid search and cross-validation, enabling it to identify complex nonlinear abnormal boundaries in high-dimensional feature space; the model takes the short-term statistical feature vectors (including mean, standard deviation, trend slope, range, and coefficient of variation) extracted in the aforementioned steps as input, outputs the probability value of each analysis window belonging to "abnormal fluctuation", and converts it into a binary classification result of "normal" and "abnormal" by setting a classification threshold.

[0023] The intensity level of abnormal fluctuations is determined based on a comprehensive assessment of the quantitatively calculated magnitude and rate of change, and is divided into the following three levels: Low-intensity fluctuations: The maximum change in key parameters (such as dissolved oxygen) within the abnormal segment is less than 1.5 times its daily fluctuation range, and the average rate of change is slow. Such fluctuations are usually caused by routine operational disturbances or slight environmental disturbances. The system only records them but does not immediately trigger high-intensity regulation. Medium-intensity fluctuations: The change in key parameters is between 1.5 and 3 times their daily fluctuation range, or the average rate of change shows a clear trend (such as a continuous decline); this level of fluctuation indicates potential risks or a decline in equipment efficiency, and the system will generate an early warning and suggest preparatory checks or adjustments; High-intensity fluctuations: The change in key parameters exceeds three times the daily fluctuation range, or the rate of change is rapid (such as a sudden change in parameter value within a short period of time). This level of fluctuation is usually related to emergency situations such as equipment failure, serious operational errors, or leakage of toxic substances. The system will immediately trigger the highest level of warning and initiate an emergency control and intervention sequence.

[0024] The abnormal fluctuation characteristic judgment result is a structured data object that encapsulates a complete feature profile of each water quality anomaly event identified through a multi-level analysis process. Specifically, this result object includes: a unique identifier for the anomaly event, its geographical location, and precise start and end timestamps; key water quality parameters that triggered the anomaly (such as dissolved oxygen) and their quantitative fluctuation characteristics, including the magnitude of change, the rate of change, and the determined intensity level (low, medium, high); warning signal information associated with the event; and a status identifier representing the continuity of the event. If the event is part of a longer, persistent anomaly, the identifier and sequence position of its associated persistent event are also recorded. This judgment result, as a standardized output, provides a complete input basis for subsequent root cause clustering analysis and the formulation of precise control strategies.

[0025] Specifically, by combining rigid water quality threshold early warning with machine learning identification based on support vector machine models, the system effectively solves the problems of difficulty in detecting and delayed early warning of early and complex water quality issues that have not exceeded the threshold but have already shown abnormal fluctuation characteristics in traditional water quality management. It can not only immediately alarm when parameters exceed the standard, but also identify complex abnormal patterns caused by the coupling of multiple factors that exceed simple threshold rules through intelligent models. By classifying the intensity of the identified anomalies and continuously verifying them, the system has realized the transformation from simple alarms to intelligent diagnosis. It can accurately quantify the severity and development trend of the problem, thereby providing a precise and structured decision-making basis for subsequent root cause tracing and graded intervention, significantly improving the early identification capability and the effectiveness of early warning of water quality risks.

[0026] S3. Based on the abnormal fluctuation characteristics, if it is found that the decrease in dissolved oxygen is associated with the increase in the concentration of harmful substances, then the K-means clustering algorithm is used to group the associated abnormal parameters and context data in combination with the operating status data and operation record data, and output the priority control groups that represent different root causes.

[0027] Furthermore, according to step S3, the specific steps include: S31. When an abnormal correlation between a decrease in dissolved oxygen and an increase in ammonia nitrogen concentration is detected in a specific aquaculture pond, the system automatically extracts equipment operation status data, refined operation record data, and relevant historical context data during the abnormal period to construct a feature data set with spatiotemporal markers containing multidimensional feature parameters. S32. The feature data set is analyzed and processed using the K-means clustering algorithm. Based on data similarity, different abnormal cases are divided into several cluster groups that represent potential root cause patterns. By analyzing the correlation strength between each cluster group and water quality anomalies, and combining historical control records, a quantitative priority control level is assigned to each root cause group. S33. Based on the cluster analysis results and priority level assessment, generate a structured set of priority control groups containing group identifiers, root cause descriptions, priority levels, and key feature fields, providing a hierarchical and classified decision-making basis for the accurate matching of subsequent control strategies.

[0028] Furthermore, the K-means clustering algorithm is used to analyze and process the feature data set. Specifically, in a land-based recirculating aquaculture system scenario, when an anomaly of "dissolved oxygen decrease - ammonia nitrogen increase" is identified in a specific aquaculture pond, all relevant standardized equipment operating parameters (such as microporous aeration pipe ventilation pressure, Roots blower frequency, and circulating water pump flow rate) and refined operation records (such as automatic feeder feeding time and weight, and water exchange valve opening duration) within that period are extracted and integrated with historical contextual data (such as recent cumulative feeding amount and biological filter backwash data). The data sets are recorded to form a multi-dimensional feature set. First, the feature set is standardized and dimensionality reduced to eliminate differences in dimensions and extract key features. In the offline stage, based on the historical abnormal case library, the optimal number of clusters K is determined by the silhouette coefficient method, and a K-means clustering model is trained. When applied online, the feature data set of the current associated anomalies is input into the trained K-means clustering model and divided into several cluster groups. Each group represents a potential root cause pattern, such as "overfeeding leads to overload of the biological filter" or "continuous decline in the efficiency of the aeration system".

[0029] By analyzing the correlation strength between each cluster group and water quality anomalies, and combining this with historical control records, a quantitative priority control level is assigned to each root cause group. Specifically, this includes: calculating the similarity between the feature vector of the current anomaly and the central vector of each cluster group, as a correlation strength score; and simultaneously, reviewing the historical case database to statistically analyze the average control response time, water quality recovery time, and control cost of historical events corresponding to each root cause group. Combining the current correlation strength with historical control effectiveness data, a quantitative priority index is calculated for each root cause group using a pre-defined priority evaluation model, and mapped to levels such as "P0 - Emergency Response," "P1 - High Priority," and "P2 - Medium Priority." For example, the "aeration equipment failure" root cause group, which is highly similar to the current anomaly and whose historical records indicate that delayed handling could lead to severe losses, will be assigned the highest priority level, thereby ensuring that control resources are accurately and efficiently directed to the most pressing root causes of the anomaly.

[0030] Specifically, by using the K-means clustering algorithm to intelligently trace the root causes and prioritize correlated water quality anomalies, this effectively solves the problem in traditional water quality management where it is difficult to trace the core causes and distinguish the urgency of problems when facing complex correlated issues (such as simultaneous anomalies in dissolved oxygen and harmful substance concentrations). This leads to weak targeted control measures and inefficient resource allocation. The algorithm can automatically identify different potential root cause patterns from multi-dimensional operational data and dynamically allocate intervention priorities based on historical control costs and current correlation strength. This transforms vague water quality anomalies into a clear list of root cause control measures with varying degrees of urgency, achieving a leap from "superficial treatment" to "root cause treatment." This provides a scientific basis for subsequent precise and efficient differentiated control.

[0031] S4. Based on the priority control groups, perform targeted strategy mapping from a pre-set structured control strategy library. This strategy library is established according to the technical requirements of the aquaculture regulations and contains combinations of control strategies logically associated with different root cause groups. Based on the mapping results, a targeted control instruction sequence containing equipment control instructions and operation adjustment instructions is generated; Furthermore, according to step S4, the specific steps include: S41. Analyze the priority control group and extract each root cause element; use the root cause element as an index to search and match in the preset structured control strategy library; if the match is missing, fill in the default security strategy. S42. Based on the strategy combination obtained from the mapping, generate corresponding device control instructions and operation adjustment instructions; sort and initially assemble the instructions according to the correspondence between root cause elements and strategies. S43. Perform conflict detection on the initially assembled instruction sequence; if a conflict exists, arbitrate and resolve it according to the priority level of the corresponding root cause element; finally output the optimized targeted control instruction sequence; the constituent elements of the instruction sequence correspond one-to-one with the root cause elements in the priority control group.

[0032] Furthermore, the system performs retrieval and matching within a pre-built structured control strategy library. Specifically, in land-based recirculating aquaculture scenarios, after extracting specific root causes such as insufficient aeration efficiency or overloaded biofilters, a matching engine combining rules and semantics is used to search the strategy library constructed based on relevant aquaculture procedures. For example, for the factor of insufficient aeration efficiency, the library has pre-set associated strategy groups, including strategy options such as increasing the operating frequency of the Roots blower, checking and cleaning the microporous aeration pipes, or starting backup pure oxygen aeration. The system calculates the matching degree between the root cause factor characteristics and strategy tags, selects one or more most suitable strategies for each factor, and forms a preliminary control plan; if no results are found, a default safety process, such as executing a system warning and prompting manual verification, is automatically triggered to ensure the integrity of the response chain.

[0033] Furthermore, based on the strategy combinations obtained from the mapping, corresponding equipment control instructions and operation adjustment instructions are generated. Specifically, this includes converting the matched abstract strategies into executable instructions that can directly drive physical equipment or adjust management parameters. For example, for the strategy of increasing the operating frequency of the Roots blower, a plaintext instruction is generated: Increase the operating frequency of the Roots blower (equipment ID: BF-01) associated with pond 1 from 35Hz to 45Hz for 30 minutes; for the strategy of reducing the feeding load, an operation instruction is generated: Reduce the daily feeding rate of the automatic feeder in pond 3 from 3% to 1.5% within the next 12 hours. Each instruction clearly points to a specific device or operation and includes precise control parameters, execution time, or conditions.

[0034] Furthermore, conflict detection is performed on the initially assembled instruction sequence, specifically including: After completing the initial sorting and assembly of instructions, the system activates the conflict detection module. This module, based on a resource occupancy model and a logical rule base, performs pairwise comparisons and analyses of all instructions in the sequence. In aquaculture scenarios, typical conflicts include equipment resource conflicts (e.g., two instructions attempting to exclusively use the same water pump during the same time period), logical conflicts (e.g., one instruction requests aeration to increase dissolved oxygen, while another instruction plans a large-scale water change that may lead to dissolved oxygen loss), and parameter target conflicts (e.g., setting conflicting expected values ​​for the same water quality parameter). Through simulation and rule verification, all potential conflict points are identified.

[0035] Furthermore, arbitration and resolution are conducted based on the priority level of the corresponding root cause factors. The final output is an optimized sequence of targeted control instructions, specifically including: when a conflict is detected, an arbitration mechanism is immediately initiated. The core principle is to trace the original root cause factors corresponding to the conflicting instructions and strictly adjudicate based on the priority levels of these factors as assessed in step S3. Instructions corresponding to higher-priority root cause factors are retained, while lower-priority instructions are modified, postponed, or eliminated. For example, if an aeration instruction corresponding to an aeration equipment malfunction (P0 level) conflicts with a regular water exchange instruction (P2 level), the system prioritizes retaining the aeration instruction and may adjust the water exchange instruction to postpone execution or reduce the water exchange ratio. After arbitration, the system performs final logical reorganization and time-based optimization of the instruction sequence, outputting a conflict-free, executable instruction sequence that maximizes the synergistic achievement of multi-objective control.

[0036] The pre-built structured control strategy library is constructed by systematically integrating aquaculture regulations, technical requirements, expert domain knowledge, and historical control case data. Specifically, it includes: structurally coding recommended treatment measures for different water quality anomalies (such as insufficient dissolved oxygen and excessive ammonia nitrogen) in aquaculture management regulations, based on relevant national and local aquaculture standards and specifications; simultaneously, collecting and analyzing historical successful control cases, extracting effective equipment operation combinations and management adjustments, and verifying and optimizing the logical correlation, applicable conditions, and execution priorities of the strategies through expert experience, ultimately forming a standardized knowledge base indexed by root cause type and containing multi-dimensional control strategy combinations.

[0037] Specifically, based on root cause analysis results, a pre-built structured strategy library is used for precise strategy mapping and intelligent instruction generation. This effectively solves the problems of traditional water quality management, such as reliance on manual experience, low matching efficiency, and susceptibility to logical conflicts in control strategies. According to different priority root cause groups, the system automatically matches and combines the optimal control strategy, transforming it into directly executable equipment control and management operation instructions. Through built-in conflict detection and priority arbitration mechanisms, the generated instruction sequences are ensured to be conflict-free at the logical and resource levels and can be executed collaboratively. This achieves a seamless connection from "root cause diagnosis" to "precise execution," automating and standardizing complex control decisions. It significantly improves the targeting, timeliness, and reliability of control measures, providing core decision-making and execution capabilities for intelligent water quality control.

[0038] S5. The targeted control command sequence is sent to the IoT actuator group to drive the corresponding oxygenation device, dosing device, or sewage discharge device to perform the operation. After execution, key water quality parameters are collected again and compared with preset standard ranges to generate preliminary water quality adjustment results.

[0039] Furthermore, according to step S5, the specific steps include: The IoT group receives the targeted control command sequence, parses and generates an executable operation command set for specific oxygenation devices, dosing devices or sewage discharge devices, and clarifies the priority and execution order of the commands; drives the corresponding actuator group to perform operations in sequence, and records the operation status, start and stop time and key operating parameters of each device in real time, forming a complete operation execution feedback data stream; After the preset equipment operation execution time window ends, the water quality parameters of the aquaculture pond and key nodes are collected again through the sensor network, covering key indicators such as dissolved oxygen, pH, ammonia nitrogen and nitrite. The collected raw data is checked for integrity and screened for outliers to obtain an accurate and reliable latest water quality parameter dataset. The water quality parameter dataset obtained from the retest is compared and analyzed item by item with the preset standard range to identify and record parameters that exceed the allowable range and their degree of deviation. Using a pre-trained support vector machine model, the water quality status after regulation is classified and predicted, and classification labels representing the adjustment effect are output (such as "significantly improved", "partially improved", "not as expected"). Combining the deviation records and classification labels, a preliminary water quality adjustment result report is generated, which also serves as the basis for evaluating the effectiveness of this regulation and triggering subsequent optimized regulation.

[0040] Furthermore, a pre-trained support vector machine (SVM) model is used to classify and predict the post-regulation water quality status, outputting classification labels representing the adjustment effect. Specifically, this involves fusing the latest water quality parameter dataset obtained from retesting (including instantaneous values ​​of dissolved oxygen, pH, ammonia nitrogen, nitrite, etc., relative changes compared to before regulation, and short-term trends) with real-time collected equipment operating status data to form a multi-dimensional "post-regulation state feature vector." This feature vector is then input into a pre-trained SVM classification model. This model is trained based on a large number of historical regulation cases and their final effects (success, partial success, failure) labels, and can learn the nonlinear mapping relationship between complex combinations of water quality parameters and regulation effects. After analyzing this feature vector, the model outputs a classification label representing the immediate effect of this regulation, such as: "Class A - Significant Improvement" (all key parameters have entered and stabilized within the target range), "Class B - Partial Improvement" (some parameters meet the target, but some indicators still deviate slightly), or "Class C - Not Meeting Expectations" (key parameters have not shown significant improvement or continue to deteriorate). This classification result provides a more intelligent and comprehensive assessment for subsequent decision-making than simple threshold comparison.

[0041] The preset standard ranges are set based on the farmed species, growth stage, and farming procedures. Specifically, they cover the target ranges for the following key water quality parameters: dissolved oxygen concentration is usually maintained between 6 mg / L and 10 mg / L; pH value is maintained within the range of 7.2 to 7.8; for adult fish farming ponds, ammonia nitrogen concentration must be below 0.2 mg / L, and nitrite concentration must be below 0.2 mg / L; for juvenile fish farming ponds, ammonia nitrogen concentration is allowed to be slightly higher but must be below 1.0 mg / L, and nitrite concentration must be below 0.1 mg / L. This standard range serves as the benchmark for water quality comparison and assessment of whether the control measures meet the standards.

[0042] Specifically, by precisely sending intelligently generated control commands to physical execution devices and conducting real-time retesting and intelligent evaluation of the control effects based on preset standards, the problem of "disconnect between execution and effect" and lack of closed-loop verification in traditional water quality control is effectively solved. It not only realizes the automated connection from decision-making to physical execution, but also uses machine learning models to conduct multi-dimensional intelligent evaluation of complex water quality improvement effects, thereby providing quantitative effect feedback for each control action and establishing a complete closed loop of "decision-execution-evaluation". This significantly improves the traceability, execution reliability and effect evaluability of control measures.

[0043] S6. If the preliminary water quality adjustment results do not meet the preset stability conditions, then based on the adjusted data stream, the support vector machine model analysis as described in step S2 and the root cause localization step as described in step S3 are re-executed to identify the remaining imbalance factors. Furthermore, according to step S6, the specific steps include: S61. When the classification labels and deviation records in the preliminary water quality adjustment results indicate that the water quality has not reached the preset stable conditions, based on the new data stream generated after the control is executed, the support vector machine model analysis and root cause localization steps are re-executed to identify the remaining imbalance factors that cause the water quality parameters to still not meet the standards. S62. For the identified remaining imbalance factors, match the corresponding supplementary intervention strategies from the structured control strategy library, generate an optimized control instruction sequence, and issue it for execution; S63. After executing the optimized control command sequence, re-collect water quality parameters for evaluation; if the key parameters are still not stable within the target range, repeat steps S61 and S62 until the preset stability conditions are met.

[0044] Among them, residual imbalance factors refer to the root causes of persistent water quality abnormalities that, after initial water quality control, key water parameters are still not stable within the target range as identified through effect evaluation based on new data streams generated after control. These causes are identified through support vector machine model analysis and root cause localization steps and are either not effectively addressed by the initial control strategy or have newly emerged. These factors are usually the result of multiple intertwined and mutually influential factors. For example, while initial aeration may alleviate insufficient dissolved oxygen, it may fail to simultaneously resolve the persistent accumulation of ammonia nitrogen caused by blockage of the biological filter, or the efficiency of aeration equipment may be adjusted but still fail to reach the expected level, among other deep-seated and secondary problems.

[0045] Supplementary intervention strategies refer to a set of additional control measures and operational instructions matched, selected, and combined from a pre-set structured control strategy library to specifically address the remaining imbalance factors identified above. These strategies are targeted supplements and optimizations to the initial control plan. For example, in addition to the already implemented aeration, specific operations such as "enhancing backwashing of biological filters," "supplementing compound microbial agents to improve nitrification efficiency," or "inspecting and calibrating aeration equipment" are matched and executed to collaboratively address the remaining and more complex imbalances and promote water quality convergence towards a stable target.

[0046] Specifically, by establishing an iterative closed-loop mechanism of "assessment-identification-re-regulation", the problem of incomplete effects and difficulty in achieving stable water quality standards due to the complexity of the problem or the limitations of the strategy in a single water quality regulation is effectively solved. It realizes adaptive and gradual correction of complex and dynamic water quality imbalance problems, and significantly improves the system's success rate and stability of the final effect in dealing with complex scenarios such as multiple factors and long-term anomalies.

[0047] S7. Continuously record the diagnostic data, executed control command sequence, and final control effect of each abnormal event to form a historical case library. Based on the historical case library, the parameters, clustering rules, and policy matching logic of the support vector machine model are optimized through machine learning methods, thereby achieving the co-evolution of the structured regulation strategy library and the diagnostic model.

[0048] Furthermore, according to step S7, the specific steps include: After each water quality anomaly event is handled, the complete data package of the anomaly event is automatically archived in a structured manner. The data package includes at least: the multi-dimensional dynamic data stream when the anomaly was triggered, the diagnostic analysis process data, the sequence of control instructions issued in various versions, the equipment execution feedback, and the final water quality stabilization result; all data packages are stored in the historical case library in chronological order to form a traceable and analyzable data set. Based on the accumulated historical case library, the correlation patterns between the control command sequence and the final control effect are analyzed periodically to evaluate the effectiveness of the existing strategy matching logic and to revise inefficient or contradictory strategy matching rules. At the same time, cases of misjudgment or fuzzy grouping generated in root cause clustering analysis are reviewed to optimize the feature weights and grouping rules of the clustering algorithm. By using the optimized strategy matching logic and grouping rules, as well as the newly added historical case data, the support vector machine classification model is incrementally trained or retrained to update its model parameters, thereby improving its recognition accuracy for complex abnormal fluctuation patterns. Then, the updated diagnostic model and the optimized structured regulation strategy library are synchronously loaded into the online system to achieve the co-evolution of diagnostic capabilities and regulation knowledge.

[0049] Furthermore, the effectiveness of existing strategy matching logic is evaluated, including establishing a multi-dimensional efficiency quantitative system based on aquaculture scenarios; and defining differentiated evaluation index systems for different aquaculture stages and seasonal characteristics. For example, in high-density aquaculture scenarios in summer, the system combines water quality recovery timeliness indicators with equipment energy consumption indicators for comprehensive evaluation; during peak feeding seasons, the focus is on examining correlation indicators such as ammonia nitrogen control efficiency and biological filter operating load. Through analysis of historical cases, it was found that for specific root causes, certain strategy combinations outperform the original strategies in terms of overall efficiency. Based on this, the system automatically optimizes the strategy matching logic, increasing the matching priority of efficient strategies.

[0050] Furthermore, the clustering and grouping rules are optimized, including feature backtracking analysis of misclassified cases in aquaculture scenarios; by analyzing the feature distribution of seasonal misclassified cases, key feature dimensions leading to classification bias are identified. For example, for common water quality anomaly patterns in spring, the system identifies under-considered periodic fluctuation features through feature analysis, dynamically adjusts the feature weighting system accordingly, and can adaptively add new feature dimensions to improve classification accuracy; at the same time, for new fluctuation patterns that appear in specific production stages such as feed conversion, corresponding clustering dimensions can be created in the feature space to achieve rapid adaptation to emerging anomaly patterns.

[0051] Furthermore, the support vector machine model update mechanism includes incremental learning and periodic reconstruction to adapt to changes in the aquaculture scenario; it establishes a model-specific optimization mechanism based on seasonal changes and production cycles, and conducts targeted reinforcement learning for anomaly types that occur frequently in different seasons. When new fluctuation patterns are detected in succession, the system can automatically trigger an emergency incremental training process to quickly update model parameters while ensuring service continuity, thus ensuring that diagnostic capabilities keep pace with the times.

[0052] Furthermore, the co-evolutionary mechanism includes phased verification and deployment in conjunction with the aquaculture operation cycle. The deployment of the new version model and strategy library follows the principle of phased verification: comprehensive testing is conducted during the aquaculture intermission period, parallel verification is performed in shadow mode during the stable production period, and version stability is maintained during critical operation periods. The system establishes a complete performance comparison and verification process, and will only be fully rolled out if the new version performs stably and is superior to the old version. At the same time, the system has an emergency rollback mechanism. When the new version encounters compatibility issues under specific aquaculture conditions, it can automatically switch to a verified stable version to ensure the system reliability of the evolutionary process.

[0053] Specifically, by constructing a closed loop of continuous learning and evolution driven by historical cases, the fundamental problems of traditional static control systems being unable to adapt to long-term environmental changes, new abnormal patterns, and the limitations of their own strategies are effectively solved. This enables the water quality control system to learn autonomously and continuously improve from historical experience, achieving a leap from the automated execution of fixed rules to an intelligent system with adaptive and self-optimizing capabilities. As a result, it maintains and continuously improves its control efficiency and reliability in actual complex aquaculture environments.

[0054] Example 2 This embodiment provides another method for regulating water quality in land-based recirculating aquaculture systems. By constructing and maintaining a stable floc community composed of bacteria, protozoa, etc., it makes the floc community the core functional unit for degrading pollutants such as ammonia nitrogen and nitrite. This solves the fundamental problems of traditional land-based recirculating aquaculture systems, which rely excessively on expensive and complex external physical and biological filtration equipment, resulting in high system construction and maintenance costs, long start-up cycles, and insufficient buffering capacity when dealing with high pollution loads. This method achieves dynamic and refined management of water purification capacity and long-term stable operation of the system through real-time monitoring and active intervention (such as adjusting the carbon-nitrogen ratio and aeration mode) of state parameters such as total floc volume and activity.

[0055] The specific steps are as follows: Construction and maintenance of biofloc system: In the recirculating water treatment unit of the aquaculture system, by adding carbon source and microbial agent, and by adjusting the carbon-nitrogen ratio and aeration intensity in the water, a stable biofloc community composed of bacteria, protozoa, microalgae and other organisms is actively cultivated and maintained, making it the core functional unit for degrading aquaculture metabolites. Monitoring and evaluating the status of bioflocs: Real-time monitoring and collection of key parameters directly related to the functional status of the biofloc system, including but not limited to the total amount of bioflocs in the water, floc particle size distribution, sedimentation performance, and activity indicators of key enzymes for nitrification and denitrification; Active regulation based on state parameters: The state parameters of the bioflocs monitored in step (2) are compared with the preset optimization threshold range. If the parameters deviate from the optimization range, the corresponding active regulation operation is triggered and executed. The active regulation operation includes: controlling the carbon-nitrogen ratio by adjusting the carbon source addition rate and type, adjusting the aeration intensity and mode to optimize the floc structure and dissolved oxygen distribution, or periodically discharging excess or aged flocs through a hydraulic separation device to maintain system activity. Evaluation of control effects and system fine-tuning: After implementing active control operations, key water quality parameters and biofloc state parameters are monitored again to assess the overall stability of the system. If the expected stable state is not achieved, the active control steps based on the state parameters are repeated to fine-tune the parameters until the water quality and biofloc ecosystem of the aquaculture system are restored and maintained in a dynamic equilibrium state.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for regulating water quality in land-based recirculating aquaculture systems, characterized in that: This application provides a method for regulating water quality in land-based recirculating aquaculture systems, comprising the following steps: S1. Deploy a sensor network in the aquaculture system to collect water quality monitoring data, equipment operating status data, and operation record data, and perform time alignment and fusion to output a multi-dimensional dynamic data stream; S2. Based on rigid water quality thresholds, perform real-time comparison and preliminary early warning of the multidimensional dynamic data stream, and use a support vector machine model to identify abnormal fluctuations that do not exceed the thresholds, and output the abnormal fluctuation feature judgment results. S3. When it is found that the decrease in dissolved oxygen is associated with the increase in the concentration of harmful substances, the K-means clustering algorithm is used to perform group analysis by combining the operating status and operation record data, and output the priority control groups that represent different root causes. S4. Based on the priority control group, map strategies from the preset structured control strategy library to generate a targeted control instruction sequence; S5. Issue and execute the targeted control command sequence, collect water quality parameters again and compare them with the preset standard range to generate preliminary water quality adjustment results; S6. If the preliminary water quality adjustment results do not meet the stability conditions, the anomaly identification and root cause analysis steps are repeated based on the adjusted data stream, and supplementary strategies are matched for iterative regulation until the water quality is stable. S7. Continuously record abnormal event data to form a historical case library, and optimize the support vector machine model, clustering rules and policy matching logic based on the library to achieve the co-evolution of the diagnostic model and the policy library.

2. The method for regulating water quality in land-based recirculating aquaculture systems according to claim 1, characterized in that: According to step S1, the specific steps include: Sensor nodes are deployed in aquaculture ponds, water storage ponds and effluent treatment units to collect real-time monitoring data on dissolved oxygen, pH, ammonia nitrogen and nitrite. Simultaneously, the operating status data of related equipment and operation records of feeding and water changing are collected. Based on the physical location of the data collection points, spatial partition identifiers are added to all incoming raw data to generate an initial partitioned data stream with clear spatial attributes. All monitoring data, equipment status data, and operation record data are stamped with a unified high-precision timestamp. Through a timestamp-driven event alignment engine, multi-source heterogeneous data are matched and calibrated in real time to eliminate timing deviations and fuse and generate a standardized multi-dimensional data stream that is strictly synchronized in time and space. Based on the standardized multidimensional data stream, a sliding time window is used for continuous rolling calculation. Within each window, the key water quality parameter sequence is statistically aggregated in real time, and short-term statistical features including mean, standard deviation and trend are extracted. The original data stream is transformed into a feature vector sequence that reflects the dynamic state of the system in real time, and a multidimensional dynamic data stream is output.

3. The method for regulating water quality in land-based recirculating aquaculture systems according to claim 1, characterized in that: According to step S2, the specific steps include: Rigid water quality thresholds are set according to aquaculture regulations and standards. Each water quality parameter in the multidimensional dynamic data stream is compared in real time to determine whether it exceeds the threshold. If it exceeds the threshold, a primary warning signal is generated immediately, and the water quality time series data that has not triggered the primary warning are separated. For the retained water quality time series data that did not exceed the threshold, a trained support vector machine model was used to identify abnormal fluctuation patterns, output binary classification judgment results, and then the water quality time series data segments that were judged to be abnormal fluctuations were selected, and the start and end timestamps of each abnormal segment were marked. By scanning the marked abnormal segments through a sliding window, the magnitude and rate of change of water quality parameters in each segment are calculated to determine the intensity level of abnormal fluctuations. If the intensity level of fluctuations reaches the preset conditions, a level-two warning signal is generated and associated with the time series data of the corresponding abnormal segment. The continuity of the secondary warning signal is verified by rule matching. It is determined whether there is overlap between abnormal segments in adjacent time windows. If there is overlap, they are merged into a continuous abnormal event, and the structured abnormal fluctuation feature judgment result is integrated and output.

4. The method for regulating water quality in land-based recirculating aquaculture systems according to claim 3, characterized in that: By scanning marked anomalous segments using a sliding window, the magnitude and rate of change of water quality parameters within each segment are calculated to determine the intensity level of anomalous fluctuations. Specifically, this involves: performing a secondary sliding window scan at a higher temporal resolution for each marked anomalous segment, calculating the magnitude and rate of change of key water quality parameters within each high-resolution window in real time; then performing aggregate analysis on the calculation results of all windows within the anomalous segment to extract the maximum and average values ​​of the magnitude of change, the peak value of the rate of change, and the proportion of the rate of change exceeding the baseline threshold—multidimensional intensity features; and mapping the aggregated features to discrete intensity levels according to preset intensity grading rules, constructing an intensity-parameter-context association map to identify the main types of water quality parameters causing the anomalies and the associated operational or equipment factors.

5. The method for regulating water quality in land-based recirculating aquaculture systems according to claim 1, characterized in that: According to step S3, the specific steps include: S31. When an abnormal correlation between a decrease in dissolved oxygen and an increase in ammonia nitrogen concentration is detected in a specific aquaculture pond, the system automatically extracts equipment operation status data, operation record data, and relevant historical context data during the abnormal period to construct a feature data set with spatiotemporal markers containing multidimensional feature parameters. S32. The feature data set is analyzed and processed using the K-means clustering algorithm. Based on data similarity, different abnormal cases are divided into several cluster groups that represent potential root cause patterns. By analyzing the correlation strength between each cluster group and water quality anomalies, and combining historical control records, a quantitative priority control level is assigned to each root cause group. S33. Based on the cluster analysis results and priority level assessment, generate a structured set of priority control groups that includes group identifiers, root cause descriptions, priority levels, and key feature fields.

6. The method for regulating water quality in land-based recirculating aquaculture systems according to claim 5, characterized in that: By analyzing the correlation strength between each cluster group and water quality anomalies, and combining historical control records, a quantitative priority control level is assigned to each root cause group. Specifically, this includes: calculating the similarity between the feature vector of the current abnormal event and the central vector of each cluster group as a correlation strength score; simultaneously, reviewing the historical case database to statistically analyze the average control response time, water quality recovery time, and control cost of historical events corresponding to each root cause group; and combining the current correlation strength with historical control effectiveness data, a quantitative priority index is calculated for each root cause group using a pre-set priority evaluation model, and the priority index is mapped to a level.

7. The method for regulating water quality in land-based recirculating aquaculture systems according to claim 1, characterized in that: According to step S4, the specific steps include: S41. Analyze the priority control group and extract each root cause element; use the root cause element as an index to search and match in the preset structured control strategy library; if the match is missing, fill in the default security strategy. S42. Based on the strategy combination obtained from the mapping, generate corresponding device control instructions and operation adjustment instructions; sort and initially assemble the instructions according to the correspondence between root cause elements and strategies. S43. Perform conflict detection on the initially assembled instruction sequence; if a conflict exists, arbitrate and resolve it according to the priority level of the corresponding root cause element; finally output the optimized targeted control instruction sequence; the constituent elements of the instruction sequence correspond one-to-one with the root cause elements in the priority control group.

8. The method for regulating water quality in land-based recirculating aquaculture systems according to claim 1, characterized in that: According to step S5, the specific steps include: The IoT group receives the targeted control command sequence, parses and generates an executable operation command set for specific oxygenation devices, dosing devices or sewage discharge devices, and clarifies the priority and execution order of the commands; drives the corresponding actuator group to perform operations in sequence, and records the operation status, start and stop time and key operating parameters of each device in real time, forming a complete operation execution feedback data stream; After the preset equipment operation execution time window ends, the water quality parameters of the aquaculture pond and key nodes are collected again through the sensor network. The integrity of the collected raw data is verified and outlier screening is performed to obtain an accurate and reliable latest water quality parameter dataset. The water quality parameter dataset obtained from the retest is compared and analyzed item by item with the preset standard range to identify and record parameters that exceed the allowable range and their degree of deviation. Using a pre-trained support vector machine model, the water quality status after regulation is classified and predicted, and classification labels representing the adjustment effect are output. Combining the deviation records and classification labels, a preliminary water quality adjustment result report is generated.

9. The method for regulating water quality in land-based recirculating aquaculture systems according to claim 1, characterized in that: According to step S6, the specific steps include: S61. When the classification labels and deviation records in the preliminary water quality adjustment results indicate that the water quality has not reached the preset stable conditions, based on the new data stream generated after the control is executed, the support vector machine model analysis and root cause localization steps are re-executed to identify the remaining imbalance factors that cause the water quality parameters to still not meet the standards. S62. For the identified remaining imbalance factors, match the corresponding supplementary intervention strategies from the structured control strategy library, generate an optimized control instruction sequence, and issue it for execution; S63. After executing the optimized control command sequence, re-collect water quality parameters for evaluation; if the key parameters are still not stable within the target range, repeat steps S61 and S62 until the preset stability conditions are met.

10. The method for regulating water quality in land-based recirculating aquaculture systems according to claim 1, characterized in that: According to step S7, the specific steps include: After each water quality anomaly event is handled, the complete data package of the anomaly event is automatically archived in a structured manner. The data package includes at least: the multi-dimensional dynamic data stream when the anomaly was triggered, the diagnostic analysis process data, the sequence of control instructions issued in various versions, the equipment execution feedback, and the final water quality stabilization result; all data packages are stored in the historical case library in chronological order to form a traceable and analyzable data set. Based on the accumulated historical case library, the correlation patterns between the control command sequence and the final control effect are analyzed periodically to evaluate the effectiveness of the existing strategy matching logic and to revise inefficient or contradictory strategy matching rules; at the same time, misjudgment or fuzzy grouping cases generated in root cause clustering analysis are reviewed to optimize the feature weights and grouping rules of the clustering algorithm. Using the optimized strategy matching logic and grouping rules, as well as the newly added historical case data, the support vector machine classification model is incrementally trained or retrained to update its model parameters. Then, the updated diagnostic model and the optimized structured regulation strategy library are synchronously loaded into the online system.

Images