A method and device for automatic grading and quality-separating treatment of aquaculture wastewater
By combining real-time data acquisition and an ecological causal weight matrix with an XGBoost model for automatic graded and quality-based treatment, the problem of existing devices failing to differentiate treatment based on pollution levels has been solved. This enables precise monitoring and efficient equipment management, reducing water treatment costs and energy consumption.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HAINAN UNIVERSITY SANYA NANFAN RESEARCH INSTITUTE
- Filing Date
- 2026-04-20
- Publication Date
- 2026-06-30
AI Technical Summary
Existing aquaculture wastewater treatment devices lack the ability to treat wastewater in stages and according to its quality, failing to differentiate treatment based on the degree of pollution. This results in over-treatment of low-pollution wastewater, increasing costs, and incomplete treatment of heavily polluted wastewater, failing to achieve efficient resource utilization and precise wastewater treatment.
By collecting data in real time using multi-parameter water quality sensors, an ecological causal weight matrix is constructed to complete the data and replace outliers. Combined with the XGBoost gradient boosting tree prediction model, the sensor cleaning parameters are dynamically adjusted to achieve automatic graded and quality-differentiated treatment.
It achieves accurate monitoring and efficient equipment management, avoids over- or substandard treatment caused by data distortion, reduces water treatment costs and energy consumption, and improves treatment efficiency.
Smart Images
Figure CN122064975B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aquaculture wastewater treatment technology, and in particular to an automatic grading and quality-separating treatment method and apparatus for aquaculture wastewater. Background Technology
[0002] With the large-scale and intensive development of aquaculture, the amount of wastewater discharged during the aquaculture process has increased significantly. Aquaculture wastewater contains large amounts of pollutants such as uneaten feed, feces, ammonia nitrogen, nitrite, and reactive phosphate. If discharged directly without effective treatment, it will lead to eutrophication of receiving water bodies, damage the aquatic ecosystem, and hinder the sustainable development of aquaculture itself. Currently, domestic and international technologies for treating aquaculture wastewater mainly fall into four categories: physical treatment, chemical treatment, biological treatment, and combined treatment technologies. However, in practical applications, existing devices have the drawback of not classifying and treating wastewater according to its pollution level. This results in over-treatment of low-pollution wastewater, increasing costs, while heavily polluted wastewater is not thoroughly treated, leading to substandard discharge and hindering efficient resource utilization and precise wastewater treatment.
[0003] Existing methods for treating aquaculture wastewater objectively have the following problems: Existing aquaculture wastewater treatment devices adopt a single treatment mode, lack the ability to treat wastewater in stages and according to different quality levels, and do not design differentiated treatment based on the degree of wastewater pollution; existing devices do not form an automated linkage mechanism, and cannot automatically adjust the opening and closing of the sewage valve according to the real-time water quality of the wastewater, and adopt a uniform treatment path, resulting in low treatment efficiency; existing devices lack the ability to predict treatment load, and cannot adjust the operating status of treatment units in advance, which easily leads to waste of treatment resources or insufficient treatment.
[0004] Therefore, this invention proposes an automatic grading and quality-separating treatment method and device for aquaculture wastewater to solve the above problems. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention develops an automatic grading and quality-differentiating treatment method and device for aquaculture wastewater. By grading the water quality data of aquaculture wastewater and maintaining the sensors, this invention forms a closed-loop management system, achieving precise monitoring and efficient equipment management.
[0006] On the one hand, the technical solution of this invention to solve the technical problem is an automatic grading and quality-separating treatment method for aquaculture wastewater, comprising the following steps:
[0007] S1. Real-time water quality data of aquaculture wastewater is collected through multi-parameter water quality sensors. The collection frequency is set, and the actual water quality classification result is labeled for each set of collected water quality data. All collected data and corresponding labels constitute a dataset.
[0008] S2. Based on the aquaculture ecological coupling relationship between the collected water quality data, construct an ecological causal weight matrix. Use the ecological causal weight matrix to fill in missing values and replace outliers to obtain the preprocessed dataset.
[0009] S3. Perform dual-dimensional dynamic threshold calibration. Based on the feature library of aquaculture species and growth stages, determine the benchmark threshold and stage correction coefficient of the static dimension and the real-time water quality fluctuation coefficient of the dynamic dimension. Then, calculate the dynamic grading threshold through the dual-dimensional results.
[0010] S4. Based on the analytic hierarchy process and the integration of ecological causal weights, a comprehensive weight is generated. The water quality data and dynamic classification thresholds are simultaneously processed for differential standardization. The comprehensive pollution index is obtained by weighted calculation of pollution contribution. The comprehensive classification threshold is obtained by weighted calculation of dynamic classification threshold. The comprehensive pollution index is then used to classify water quality based on the comprehensive classification threshold.
[0011] S5. Construct an XGBoost gradient boosting tree prediction model, calculate the load level based on the comprehensive pollution index, dissolved oxygen concentration, breeding density and comprehensive grading threshold, calculate the sensor drift coefficient based on sensor detection value and standard detection value, and then execute the load-cleaning linkage adaptive algorithm based on the load level and sensor drift coefficient to dynamically adjust the sensor cleaning parameters.
[0012] S1 is as follows:
[0013] Water quality sensors are submerged and deployed at the drainage outlets and sedimentation tanks of aquaculture ponds.
[0014] The water quality data collected by the multi-parameter water quality sensor includes dissolved oxygen, ammonia nitrogen, nitrite, pH value, turbidity, chemical oxygen demand, total phosphorus, and salinity.
[0015] Adjust the data collection frequency according to the breeding situation and environmental changes;
[0016] Water quality is classified into three levels: slightly safe, moderately warning, and severely dangerous.
[0017] A dynamic labeling mechanism is adopted, in which professionals in the field classify the water quality of each set of collected water quality data. The labeling is based on seasonality, compound pollution, and sensor calibration results.
[0018] S2 is as follows:
[0019] Based on the theory of ecological aquaculture, the causal relationship between various water quality data is determined. The Pearson correlation coefficient between various water quality data is calculated through historical data to obtain the correlation strength between various water quality data. After normalizing the causal relationship and the Pearson correlation coefficient, the data are then fused using the weighted average method to obtain the ecological causal weight factor between any two water quality data, and thus obtain the ecological causal weight matrix.
[0020] Extract all ecological causal weight factors related to missing or outlier values from the ecological causal weight matrix, multiply them by the standardized water quality data of the same sample (both non-missing and normal), sum the multiplication results and divide by the sum of all ecological causal weight factors related to missing or outlier values to obtain the imputation or replacement values, thus obtaining the preprocessed dataset.
[0021] The details of S3 are as follows:
[0022] (1) Static grading:
[0023] 1) Baseline threshold:
[0024] Based on the feature library of aquaculture species and growth stages, historical water quality data under the same conditions are determined, and the data sample size is determined. The standard deviation and mean of historical water quality data are calculated, and the benchmark threshold is calculated by "mean + k standard deviation". k represents the settable confidence coefficient, and different k correspond to different water quality grades.
[0025] For two-way water quality data, the maximum and minimum values and standard deviations are determined based on historical water quality data, and the baseline threshold is calculated by "maximum and minimum values ± k standard deviations".
[0026] 2) Stage correction factor:
[0027] The current stage duration is obtained directly. The total duration of the current stage, the influencing factor of the current stage, and the correction coefficient range of the current stage are obtained from the feature library of breeding species and growth stages. The stage progress factor is calculated by using the current stage duration and the total duration of the current stage. The stage correction coefficient is calculated by using the stage progress factor, the correction coefficient range of the current stage, and the influencing factor of the current stage.
[0028] (2) Dynamic grading:
[0029] Calculate the real-time water quality fluctuation coefficient.
[0030] The time window length is determined based on the data acquisition frequency of the sensor, and the real-time water quality fluctuation coefficient is calculated based on the time window length. The attributes of the water quality data are unidirectional deterioration, unidirectional optimization, and bidirectional indicators. The higher the unidirectional deterioration value, the worse the water quality; the higher the unidirectional optimization value, the better the water quality; and the more the bidirectional indicator value deviates from the range, the worse the water quality.
[0031] 1) Unidirectional degradation;
[0032] The mean and standard deviation of water quality data are calculated based on the length of the time window. The ratio of the standard deviation to the mean is the coefficient of variation of water quality data. The upward trend of water quality data in time period T is fitted by linear regression. The unidirectional deterioration fluctuation coefficient is calculated by combining the coefficient of variation and the upward trend.
[0033] 2) Unidirectional optimization:
[0034] Similar to unidirectional deterioration, the coefficient of variation is calculated. The downward trend of water quality data in time period T is fitted by linear regression. The unidirectional optimization fluctuation coefficient is calculated by combining the coefficient of variation and the downward trend.
[0035] 3) Two-way indicators:
[0036] The maximum and minimum values are determined based on historical water quality data corresponding to the feature database of aquaculture species and growth stages. The deviation index is calculated based on the maximum and minimum values and the length of the time window. The trend deviation factor of the water quality data in time period T is fitted by linear regression. The two-way index fluctuation coefficient is calculated by combining the deviation index and the trend deviation factor.
[0037] (3) Dynamic grading threshold:
[0038] The benchmark threshold and stage correction coefficient of static grading are combined with the three types of volatility coefficients in dynamic grading to obtain the dynamic grading threshold.
[0039] The specific operation of S4 is as follows:
[0040] A priority system for the toxic effects of water quality indicators was constructed. The priority weights of each water quality data were calculated by combining the analytic hierarchy process with the priority system. The priority weights were then integrated with the ecological causal weight factors in the ecological causal weight matrix to generate a comprehensive weight.
[0041] Differential standardization was performed on the water quality data of each attribute after preprocessing, and differential standardization was also performed on the dynamic grading thresholds of each water quality data.
[0042] The pollution contribution of each water quality data point is calculated based on the standardized water quality data and the corresponding comprehensive weight. The pollution contribution of each water quality data point is added together to obtain the threshold-coupled comprehensive pollution index.
[0043] The comprehensive grading threshold is obtained by weighted summation based on the dynamic grading thresholds and corresponding comprehensive weights after standardization of various water quality data.
[0044] Water quality is classified based on a threshold-coupled comprehensive pollution index using a comprehensive classification threshold.
[0045] S5 is detailed below:
[0046] The calculation process for the load level is as follows:
[0047] The comprehensive pollution index, dissolved oxygen concentration, aquaculture density, and comprehensive grading threshold were normalized to eliminate the influence of dimensions. Ratio features were constructed, with the ratio of dissolved oxygen concentration to comprehensive pollution index as the self-purification capacity pressure ratio and the ratio of aquaculture density to comprehensive grading threshold as the ecological carrying capacity pressure ratio. In addition, the water quality data corresponding to the two largest ecological causal weight factors in the ecological causal weight matrix, excluding dissolved oxygen, were selected as additional input features.
[0048] The XGBoost gradient boosting tree prediction model is trained based on the comprehensive pollution index, dissolved oxygen concentration, aquaculture density, comprehensive grading threshold, historical water quality data, and corresponding load levels. The residual gradient is iteratively calculated, the decision tree nodes are split to maximize the gain, and the prediction results are updated to obtain the trained XGBoost gradient boosting tree prediction model. The comprehensive pollution index, dissolved oxygen concentration, aquaculture density, comprehensive grading threshold, ratio features, and additional input features are then input into the trained XGBoost gradient boosting tree prediction model to obtain the load index.
[0049] The calculation process for the sensor drift coefficient is as follows:
[0050] While the sensor collects water quality data, the aquaculture wastewater is tested using laboratory standard methods to obtain standard values. The ratio of the difference between the sensor detection value and the standard value to the standard value is calculated to obtain the instantaneous drift coefficient. Then, a time decay factor and a load correlation factor are introduced to dynamically correct the instantaneous drift coefficient. The time decay factor is obtained by fitting the sensor's natural decay characteristics over time with historical data, and the load correlation factor is obtained by mapping the load index through the Sigmoid activation function. Finally, the dynamic sensor drift coefficient is obtained.
[0051] The load-cleaning linkage adaptive cleaning strategy is as follows:
[0052] The sensor cleaning trigger value is obtained by weighting the load index and the dynamic sensor drift coefficient. A preset cleaning degree strategy is then implemented, with different cleaning degrees for different sensor cleaning trigger values. The cleaning degree strategy includes cleaning frequency, cleaning duration, and cleaning intensity.
[0053] On the other hand, the present invention also provides an automatic grading and quality-separating treatment device for aquaculture wastewater, which performs an automatic grading and quality-separating treatment method for aquaculture wastewater, including a main unit, a multi-parameter water quality sensor and an ultrasonic cleaning ring, specifically including the following units:
[0054] Real-time water quality monitoring unit: includes a multi-parameter water quality sensor. The sensor body is a cylindrical structure and is installed at the drainage outlet of the aquaculture pond and the sedimentation tank via a 304 stainless steel mounting bracket. The depth of the sensor probe can be changed by adjusting the adjustable bolts on the stainless steel mounting bracket.
[0055] The sensor data cable is inserted into the sensor interface on the back of the chassis through a waterproof connector to establish a connection with the data acquisition and transmission module inside the host. Data transmission is carried out via RS485 wired connection.
[0056] An ultrasonic cleaning ring is set around the sensor probe, and the ultrasonic cleaning ring is connected to the central controller.
[0057] Central control and data processing unit: The main unit integrates a PLC controller, data acquisition and transmission module, edge computing module and power supply module. A rectangular operation window with an operation display screen is opened on the front of the chassis.
[0058] Graded and differentiated sewage discharge valve assembly: includes 3 electric sewage discharge valves, each valve is equipped with an independent electric actuator, and the inlet end of each of the 3 sewage discharge valves is connected to the main outlet pipe of the aquaculture pond through the sewage discharge valve pipe;
[0059] Power supply and data export components: One end of the power cord is inserted into the main unit, and the other end is connected to the power socket in the aquaculture farm; historical water quality data is exported from the data acquisition and transmission module of the main unit.
[0060] The specific operating procedures for the device are as follows:
[0061] After the power is turned on, the host is started and initialized. The PLC controller receives multi-parameter sensor data stored in the data acquisition and transmission module, summarizes and packages the data and sends it to the edge computing module.
[0062] On the host operation display screen, multiple parameter weighting coefficients, breeding species, and initial release date are set by operation. The host system automatically loads the corresponding species' feature library and dynamic grading threshold parameters.
[0063] After preprocessing the received data, the edge computing module calculates the comprehensive pollution index, compares it with the dynamic grading threshold parameters, and determines the pollution level of the effluent in real time. Based on the pollution level, it activates the corresponding electric drain valve. It further calculates the load level and sensor drift coefficient to control the operation of the ultrasonic cleaning ring.
[0064] The effects described in the invention are merely those of the embodiments, and not all the effects of the invention. The above technical solutions have the following advantages or beneficial effects:
[0065] This invention does not employ traditional methods such as mean-filling and nearest-neighbor filling for data supplementation and correction during data preprocessing. Instead, it is based on the ecological logic of ecological aquaculture theory, considering the inherent ecological correlations between water quality data for supplementation and correction. This avoids erroneous associations caused by statistical coincidences. Furthermore, by normalizing and integrating causal relationships and correlation strengths, it ensures that weight allocation does not deviate from the ecological coupling relationship of the aquaculture species, thus obtaining accurate and reliable data for subsequent calculations. This invention balances practical adaptability to aquaculture with real-time dynamic accuracy through a combination of static and dynamic approaches, taking into account the impact of differences in aquaculture species and growth stages on water quality, as well as the differences in water quality data attributes. This invention avoids grading results from deviating from reality and prevents short-term fluctuations from affecting the judgment of water quality status, thereby accurately capturing water quality change trends and achieving precise water quality grading. The invention calculates comprehensive weights for various water quality data by integrating an ecological causal weight matrix and a toxicity impact priority weight. This considers both the impact of ecological laws and the hazard priority of different pollutants, avoiding the limitations of single weights. In the comprehensive pollution index calculation process, it focuses on the true pollution sources and weakens secondary pollution sources. A comprehensive grading threshold composed of dynamic grading thresholds is used to grade the comprehensive pollution index for water quality, achieving a closed-loop logic of multi-dimensional fusion, precise quantification, and adaptive grading. Furthermore, the invention uses the XGBoost gradient boosting tree prediction model to capture the nonlinear relationship between water quality indicators and load levels. Compared to traditional linear models, the load index prediction results of the XGBoost gradient boosting tree prediction model are more consistent with the dynamic changes of complex aquaculture ecosystems. At the same time, the model's input takes into account four dimensions: pollution level, water self-purification, ecological carrying capacity, and indicator correlation, avoiding prediction bias caused by single features.
[0066] This invention ultimately forms a complete data generation-processing-evaluation-maintenance-regeneration chain. After calculating the water quality classification results, the sensor cleaning trigger value is further calculated. While classifying the water quality, the impact of aquaculture wastewater on the accuracy of sensor data acquisition is also taken into account, thus forming a management closed loop and realizing accurate monitoring and efficient equipment management. Attached Figure Description
[0067] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.
[0068] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation
[0069] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific implementation methods and in conjunction with the accompanying drawings.
[0070] Example 1
[0071] like Figure 1 As shown, an automatic grading and quality-separating treatment method for aquaculture wastewater includes the following steps:
[0072] S1. Real-time water quality data of aquaculture wastewater is collected through multi-parameter water quality sensors. The collection frequency is set, and the actual water quality classification result is labeled for each set of collected water quality data. All collected data and corresponding labels constitute a dataset.
[0073] S2. Based on the aquaculture ecological coupling relationship between the collected water quality data, construct an ecological causal weight matrix. Use the ecological causal weight matrix to fill in missing values and replace outliers to obtain the preprocessed dataset.
[0074] S3. Perform dual-dimensional dynamic threshold calibration. Based on the feature library of aquaculture species and growth stages, determine the benchmark threshold and stage correction coefficient of the static dimension and the real-time water quality fluctuation coefficient of the dynamic dimension. Then, calculate the dynamic grading threshold through the dual-dimensional results.
[0075] S4. Based on the analytic hierarchy process and the integration of ecological causal weights, a comprehensive weight is generated. The water quality data and dynamic classification thresholds are simultaneously processed for differential standardization. The comprehensive pollution index is obtained by weighted calculation of pollution contribution. The comprehensive classification threshold is obtained by weighted calculation of dynamic classification threshold. The comprehensive pollution index is then used to classify water quality based on the comprehensive classification threshold.
[0076] S5. Construct an XGBoost gradient boosting tree prediction model, calculate the load level based on the comprehensive pollution index, dissolved oxygen concentration, breeding density and comprehensive grading threshold, calculate the sensor drift coefficient based on sensor detection value and standard detection value, and then execute the load-cleaning linkage adaptive algorithm based on the load level and sensor drift coefficient to dynamically adjust the sensor cleaning parameters.
[0077] In a specific implementation, S1 is as follows:
[0078] Water quality sensors are submerged and deployed at the drainage outlets and sedimentation tanks of aquaculture ponds.
[0079] The water quality data collected by the multi-parameter water quality sensor includes dissolved oxygen, ammonia nitrogen, nitrite, pH value, turbidity, chemical oxygen demand, total phosphorus, and salinity.
[0080] Adjust the data collection frequency according to the breeding situation and environmental changes;
[0081] Water quality is classified into three levels: slightly safe, moderately warning, and severely dangerous.
[0082] Mild safety corresponds to Level 1, where aquaculture wastewater can be directly recycled for aquaculture or safely discharged into natural water bodies;
[0083] A moderate warning corresponds to level two, requiring the drain valve to be activated and the situation addressed as soon as possible.
[0084] The severe danger corresponds to level three, requiring the immediate activation of the drain valve; the aquaculture wastewater must not be discharged at this time.
[0085] A dynamic labeling mechanism is adopted, in which professionals in the field classify the water quality of each set of collected water quality data. The labeling is based on seasonality, compound pollution, and sensor calibration results.
[0086] S2 is as follows:
[0087] Based on ecological aquaculture theory, the causal relationships between various water quality data points are determined. Pearson correlation coefficients are calculated using historical data to determine the correlation strength between these data points. After normalizing the causal relationships and Pearson correlation coefficients, a weighted average method is used to fuse the data, yielding the ecological causal weight factor between any two water quality data points. This leads to the ecological causal weight matrix. , The representation is as follows:
[0088] ,
[0089] Indicates the first Water quality data and the first Ecological causal weighting factors among water quality data items hour, ;
[0090] From the ecological causal weight matrix Extract all ecological causal weight factors related to missing or outlier values, multiply them by the standardized water quality data of the same sample (both non-missing and normal), sum the multiplication results and divide by the sum of all ecological causal weight factors related to missing or outlier values to obtain the imputation or replacement values, thus obtaining the preprocessed dataset.
[0091] This invention constructs a weight matrix based on ecological coupling relationships, fills in missing values and replaces outliers, thereby correcting data deviations. This ensures that water quality data truly reflects the pollution status and avoids over-treatment of mild effluent or severe non-compliance due to data distortion.
[0092] The details of S3 are as follows:
[0093] (1) Static grading:
[0094] 1) Baseline threshold:
[0095] Based on the feature library of aquaculture species and growth stages, historical water quality data under the same conditions are determined, and the data sample size is determined. The standard deviation and mean of historical water quality data are calculated, and the benchmark threshold is calculated by "mean + k standard deviation". k represents the settable confidence coefficient, and different k correspond to different water quality grades.
[0096] For two-way water quality data, the maximum and minimum values and standard deviations are determined based on historical water quality data, and the baseline threshold is calculated by "maximum and minimum values ± k standard deviations".
[0097] 2) Stage correction factor:
[0098] The current stage duration is obtained directly. The total duration of the current stage, the influencing factor of the current stage, and the correction coefficient range of the current stage are obtained from the feature library of breeding species and growth stages. The stage progress factor is calculated by using the current stage duration and the total duration of the current stage. The stage correction coefficient is calculated by using the stage progress factor, the correction coefficient range of the current stage, and the influencing factor of the current stage.
[0099] (2) Dynamic grading:
[0100] Calculate the real-time water quality fluctuation coefficient.
[0101] The time window length is determined based on the data acquisition frequency of the sensor, and the real-time water quality fluctuation coefficient is calculated based on the time window length. The attributes of the water quality data are unidirectional deterioration, unidirectional optimization, and bidirectional indicators. The higher the unidirectional deterioration value, the worse the water quality; the higher the unidirectional optimization value, the better the water quality; and the more the bidirectional indicator value deviates from the range, the worse the water quality.
[0102] 1) Unidirectional degradation;
[0103] The mean and standard deviation of water quality data are calculated based on the length of the time window. The ratio of the standard deviation to the mean is the coefficient of variation of water quality data. The upward trend of water quality data in time period T is fitted by linear regression. The unidirectional deterioration fluctuation coefficient is calculated by combining the coefficient of variation and the upward trend.
[0104] 2) Unidirectional optimization:
[0105] Similar to unidirectional deterioration, the coefficient of variation is calculated. The downward trend of water quality data in time period T is fitted by linear regression. The unidirectional optimization fluctuation coefficient is calculated by combining the coefficient of variation and the downward trend.
[0106] 3) Two-way indicators:
[0107] The maximum and minimum values are determined based on historical water quality data corresponding to the feature database of aquaculture species and growth stages. The deviation index is calculated based on the maximum and minimum values and the length of the time window. The trend deviation factor of the water quality data in time period T is fitted by linear regression. The two-way index fluctuation coefficient is calculated by combining the deviation index and the trend deviation factor.
[0108] (3) Dynamic grading threshold:
[0109] The benchmark threshold and stage correction coefficient of static grading are combined with the three types of volatility coefficients in dynamic grading to obtain the dynamic grading threshold.
[0110] The specific operation of S4 is as follows:
[0111] A priority system for the toxic effects of water quality indicators was constructed. The priority weights of each water quality data were calculated by combining the analytic hierarchy process with the priority system. The priority weights were then integrated with the ecological causal weight factors in the ecological causal weight matrix to generate a comprehensive weight.
[0112] Differential standardization was performed on the water quality data of each attribute after preprocessing, and differential standardization was also performed on the dynamic grading thresholds of each water quality data.
[0113] The pollution contribution of each water quality data point is calculated based on the standardized water quality data and the corresponding comprehensive weight. The pollution contribution of each water quality data point is added together to obtain the threshold-coupled comprehensive pollution index.
[0114] The comprehensive grading threshold is obtained by weighted summation based on the dynamic grading thresholds and corresponding comprehensive weights after standardization of various water quality data.
[0115] Water quality is classified based on a threshold-coupled comprehensive pollution index using a comprehensive classification threshold.
[0116] S5 is detailed below:
[0117] In a specific implementation, the calculation process for the load level is as follows:
[0118] The comprehensive pollution index, dissolved oxygen concentration, aquaculture density, and comprehensive grading threshold were normalized to eliminate the influence of dimensions. Ratio features were constructed, with the ratio of dissolved oxygen concentration to comprehensive pollution index as the self-purification capacity pressure ratio and the ratio of aquaculture density to comprehensive grading threshold as the ecological carrying capacity pressure ratio. In addition, the water quality data corresponding to the two largest ecological causal weight factors in the ecological causal weight matrix, excluding dissolved oxygen, were selected as additional input features.
[0119] The XGBoost gradient boosting tree prediction model is trained based on the comprehensive pollution index, dissolved oxygen concentration, aquaculture density, comprehensive grading threshold, historical water quality data, and corresponding load levels. The residual gradient is iteratively calculated, the decision tree nodes are split to maximize the gain, and the prediction results are updated to obtain the trained XGBoost gradient boosting tree prediction model. The comprehensive pollution index, dissolved oxygen concentration, aquaculture density, comprehensive grading threshold, ratio features, and additional input features are then input into the trained XGBoost gradient boosting tree prediction model to obtain the load index.
[0120] In a specific implementation, the calculation process for the sensor drift coefficient is as follows:
[0121] While the sensor collects water quality data, the aquaculture wastewater is tested using laboratory standard methods to obtain standard values. The ratio of the difference between the sensor detection value and the standard value to the standard value is calculated to obtain the instantaneous drift coefficient. Then, a time decay factor and a load correlation factor are introduced to dynamically correct the instantaneous drift coefficient. The time decay factor is obtained by fitting the sensor's natural decay characteristics over time with historical data, and the load correlation factor is obtained by mapping the load index through the Sigmoid activation function. Finally, the dynamic sensor drift coefficient is obtained.
[0122] In a specific implementation, the load-cleaning linkage adaptive cleaning strategy is as follows:
[0123] The sensor cleaning trigger value is obtained by weighting the load index and the dynamic sensor drift coefficient. A preset cleaning degree strategy is then implemented, with different cleaning degrees for different sensor cleaning trigger values. The cleaning degree strategy includes cleaning frequency, cleaning duration, and cleaning intensity.
[0124] Example 2
[0125] An automatic grading and quality-separating treatment device for aquaculture wastewater, implementing an automatic grading and quality-separating treatment method for aquaculture wastewater, includes a main unit, a multi-parameter water quality sensor, and an ultrasonic cleaning ring, specifically comprising the following units:
[0126] Real-time water quality monitoring unit: includes a multi-parameter water quality sensor. The sensor body is a cylindrical structure and is installed at the drainage outlet of the aquaculture pond and the sedimentation tank via a 304 stainless steel mounting bracket. The depth of the sensor probe can be changed by adjusting the adjustable bolts on the stainless steel mounting bracket to adapt to different water levels.
[0127] The sensor data cable is inserted into the sensor interface on the back of the main unit through a waterproof connector to establish a connection with the data acquisition and transmission module inside the main unit. Data transmission is carried out via RS485 wired method, and the data acquisition frequency is preset to once per minute.
[0128] An ultrasonic cleaning ring is set around the sensor probe and connected to the central controller. The ultrasonic cleaning ring is used to remove the biofilm on the surface of the sensor probe.
[0129] Central Control and Data Processing Unit: The main unit is a vertical, rectangular chassis with four rubber feet at the bottom for stability and levelness. Inside, it integrates a PLC controller, data acquisition and transmission module, edge computing module, and power supply module. A rectangular operation window with a display screen is located on the front of the chassis. Above the window are a power button and a USB port for data export. The back of the display screen connects to the RS485 interface of the PLC controller inside the main unit via an RS485 data cable for bidirectional data transmission. The PLC controller and data acquisition and transmission module are spaced 50mm apart to avoid electromagnetic interference. A power socket, multi-parameter sensor interface, and three drain valve interfaces are located on the lower rear side of the chassis. The edge computing module connects to the PLC via an internal RS485 bus for data interaction. An SD card can be inserted into the main unit to store recent historical data and model parameters, eliminating the need for large-capacity storage.
[0130] Graded and differentiated sewage discharge valve assembly: includes 3 electric sewage discharge valves, each valve is equipped with an independent electric actuator, and the inlet end of each of the 3 sewage discharge valves is connected to the main outlet pipe of the aquaculture pond through the sewage discharge valve pipe;
[0131] The specific connection method is as follows: a four-way connecting pipe is installed on the main outlet pipe. The three branch holes of the four-way pipe are bonded to one end of the drain valve connecting pipe with UPVC glue. The other end of the connecting pipe is connected to the drain valve inlet. The outlet ends of the three drain valves flow to the corresponding treatment process units through branch pipes. The three drain valves correspond to light, medium and heavy water quality respectively. Each electric actuator is connected to the corresponding drain valve interface on the back of the chassis through a connecting wire, thereby forming a circuit path with the output terminal of the PLC controller in the main unit.
[0132] Power supply and data export components: One end of the power cord is inserted into the main unit, and the other end is connected to the power socket of the aquaculture farm. When the power button is pressed, the power module inside the main unit starts, providing power to the PLC controller, display screen, sensors, actuators and other components. The button indicator light (green) lights up, indicating that the device is in working condition. The data export USB port supports inserting a USB flash drive, which can export historical water quality data from the data acquisition and transmission module of the main unit.
[0133] The specific operating procedures for the device are as follows:
[0134] After the power is turned on, the host is started and initialized. The PLC controller receives multi-parameter sensor data stored in the data acquisition and transmission module, summarizes and packages the data and sends it to the edge computing module.
[0135] On the host operation display screen, multiple parameter weighting coefficients, breeding species, and initial release date are set by operation. The host system automatically loads the corresponding species' feature library and dynamic grading threshold parameters.
[0136] After preprocessing the received data, the edge computing module calculates the comprehensive pollution index, compares it with the dynamic grading threshold parameters, and determines the pollution level of the effluent in real time. Based on the pollution level, it activates the corresponding electric drain valve. It further calculates the load level and sensor drift coefficient to control the operation of the ultrasonic cleaning ring.
[0137] Example 3
[0138] To verify the stability and control precision of the automatic grading and quality-separating treatment method and device for aquaculture wastewater described in this invention during actual operation, the following experiments were conducted in conjunction with an aquaculture company:
[0139] Automatic grading and quality-differentiating treatment devices for aquaculture wastewater were deployed in adult fish rearing ponds and frying ponds, respectively. Sensors in the adult fish rearing ponds were deployed in the lower middle section, while sensors in the frying ponds were deployed in the upper middle section. Total dissolved solids (TTS) were set for the three drain valves: TTS less than 30,000 mg / L for light drain valves, TTS between 30,000 and 50,000 mg / L for moderate drain valves, and TTS greater than 50,000 mg / L for heavy drain valves. Portable water quality parameter instruments were used to measure the TTS of the water at the drain outlet of the drain valves. After the sensors of both devices started collecting data, aquaculture water samples were collected manually for water quality grading and labeling. Each sample was 500 mL in size and transported to the laboratory in a refrigerated insulated box (4℃) to avoid changes in sample composition.
[0140] Valve opening and closing accuracy is a core indicator for measuring the precision of device control. The accuracy of the sewage valve's opening and closing was determined by measuring the TSS (Total Sediment Saturation) using a portable water quality parameter meter. Higher accuracy indicates more accurate predictions from the automatic grading and quality classification method for aquaculture wastewater. The entire experiment involved continuous dynamic monitoring for 72 hours, recording 144 sets of valid data. The final results showed that the electric sewage valve's opening and closing accuracy reached 98.6%, specifically demonstrated in the following three aspects:
[0141] (1) Threshold matching accuracy: When the measured TSS of the portable instrument was <30000 mg / L, the mild drain valve was triggered to open 32 times, while the moderate and severe drain valves did not malfunction. When the measured TSS was in the range of 30000-50000 mg / L, the moderate drain valve opened accurately 45 times, with only one instance of a mild valve malfunction caused by a momentary disturbance in the water body (which was automatically corrected within 5 minutes). When the measured TSS was >50000 mg / L, the severe drain valve opened promptly 58 times with no response delay. The error rate was less than 3%, ensuring the accuracy of the threshold judgment.
[0142] (2) Action response speed: After the TSS value reaches the corresponding sewage discharge threshold, the average response time of the device valve from receiving the signal to fully opening is 3.2 seconds. The response speed of the device in the adult fish breeding pond is slightly faster (average 2.8 seconds), while the response time of the device in the seedling pond is slightly longer due to the lower water flow (average 3.6 seconds). Both meet the needs of rapid discharge of pollutants during the breeding process and avoid the accumulation of pollutants.
[0143] (3) Continuous operation stability: During the 72-hour continuous operation, neither of the two sets of devices experienced valve jamming or accidental closure. In the complex environment of the gentian grouper aquaculture pond with a pH value of 7.8-8.5 and a water temperature of 26-28℃, the valve sealing performance was good and there was no leakage. The mechanical structure and electronic control system were well compatible.
[0144] In addition, regarding water treatment costs, the adoption of an automatic grading and quality-separating treatment method and device for aquaculture wastewater significantly reduced water treatment costs. Specifically, based on a single pond (20m³) operating for an average of 16 hours per day, the comparison of water treatment costs before and after the device's implementation shows that the automatic grading and quality-separating treatment device for aquaculture wastewater can reduce the average daily water treatment cost per pond by 21.3%, demonstrating a significant cost optimization effect. The specific calculations are as follows:
[0145] (1) Chemical cost: Before the device was used, in order to control the concentration of pollutants in the water, about 0.8 kg of flocculants, disinfectants and other chemicals were added to a single pool per day, and the chemical purchase cost was RMB 12 per day; after the device was used, due to the reduction of the total amount of pollutants by the graded and precise sewage discharge, the amount of chemicals added was reduced to 0.5 kg per day, and the daily chemical cost was reduced to RMB 7.5, a reduction of 37.5%.
[0146] (2) Energy consumption cost: The operating power of the device is 0.3kW, and the daily power consumption of a single pool is 4.8kWh for 16 hours of operation. Based on the local industrial electricity price of 0.8 yuan / kWh, the daily energy consumption cost is 3.84 yuan. Before the device was used, the traditional sewage discharge method required the use of a high-power water pump for circulation, with a daily power consumption of 8kWh per pool and an energy cost of 6.4 yuan. After the device was used, the energy consumption cost was reduced by 40%.
[0147] (3) Labor cost: Before the device was used, a dedicated person was required to inspect the sewage discharge every 2 hours for each pool, with an average daily labor input of 1.5 hours. Based on the average daily wage of 200 yuan for aquaculture workers (8-hour work system), the average daily labor cost was 37.5 yuan. After the device achieved fully automatic and precise sewage discharge, it only needs to be inspected once in the morning and once in the evening. The average daily labor input was reduced to 0.5 hours, and the labor cost was reduced to 12.5 yuan, a reduction of 66.7%.
[0148] (4) Wastewater treatment cost: Traditional wastewater treatment methods involve mixed discharge of pollutants, with an average daily discharge of 8m³ per pool and a wastewater treatment plant cost of 5 yuan / m³, resulting in an average daily wastewater treatment cost of 40 yuan. After the device performs graded wastewater discharge, the volume of high-concentration pollutant wastewater is reduced to 3m³, and low-concentration wastewater can be partially recycled. The actual discharge volume is 4m³, and the average daily wastewater treatment cost is reduced to 20 yuan, a reduction of 50%.
[0149] Based on comprehensive calculations, the average daily water treatment cost per pond after using the automatic grading and quality-separating treatment device for aquaculture wastewater decreased from 90.4 yuan to 71.34 yuan. Calculated based on 100 PP ponds for the enterprise, the average daily total cost can be reduced by 1,906 yuan, resulting in annual cost savings of approximately 695,000 yuan.
[0150] Although the specific embodiments of the invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the invention. Based on the technical solutions of the invention, various modifications or variations that can be made by those skilled in the art without creative effort are still within the scope of protection of the invention.
Claims
1. An automatic grading and quality-separating treatment method for aquaculture wastewater, characterized in that, Includes the following steps: S1. Real-time water quality data of aquaculture wastewater is collected through multi-parameter water quality sensors. The water quality data includes dissolved oxygen, ammonia nitrogen, nitrite, pH value, turbidity, chemical oxygen demand, total phosphorus and salinity. The collection frequency is set, and the actual water quality classification result is labeled for each set of collected water quality data. All collected data and corresponding labels constitute a dataset. S2. Based on the aquaculture ecological coupling relationship between the collected water quality data, construct an ecological causal weight matrix. Use the ecological causal weight matrix to fill in missing values and replace outliers to obtain the preprocessed dataset. S3. Perform dual-dimensional dynamic threshold calibration. Based on the feature library of aquaculture species and growth stages, determine the benchmark threshold and stage correction coefficient of the static dimension and the real-time water quality fluctuation coefficient of the dynamic dimension. Then, calculate the dynamic grading threshold through the dual-dimensional results. The details of S3 are as follows: (1) Static grading: 1) Baseline threshold: Based on the feature library of aquaculture species and growth stages, historical water quality data under the same conditions are determined, and the data sample size is determined. The standard deviation and mean of historical water quality data are calculated, and the benchmark threshold is calculated by "mean + k standard deviation". k represents the settable confidence coefficient, and different k correspond to different water quality grades. For two-way water quality data, the maximum and minimum values and standard deviations are determined based on historical water quality data, and the baseline threshold is calculated by "maximum and minimum values ± k standard deviations". 2) Stage correction factor: The current stage duration is obtained directly. The total duration of the current stage, the influencing factor of the current stage, and the correction coefficient range of the current stage are obtained from the feature library of breeding species and growth stages. The stage progress factor is calculated by using the current stage duration and the total duration of the current stage. The stage correction coefficient is calculated by using the stage progress factor, the correction coefficient range of the current stage, and the influencing factor of the current stage. (2) Dynamic grading: Calculate the real-time water quality fluctuation coefficient. The time window length is determined based on the data acquisition frequency of the sensor, and the real-time water quality fluctuation coefficient is calculated based on the time window length. The attributes of the water quality data are unidirectional deterioration, unidirectional optimization, and bidirectional indicators. The higher the unidirectional deterioration value, the worse the water quality; the higher the unidirectional optimization value, the better the water quality; and the more the bidirectional indicator value deviates from the range, the worse the water quality. 1) Unidirectional degradation; The mean and standard deviation of water quality data are calculated based on the length of the time window. The ratio of the standard deviation to the mean is the coefficient of variation of water quality data. The upward trend of water quality data in time period T is fitted by linear regression. The unidirectional deterioration fluctuation coefficient is calculated by combining the coefficient of variation and the upward trend. 2) Unidirectional optimization: Similar to unidirectional deterioration, the coefficient of variation is calculated. The downward trend of water quality data in time period T is fitted by linear regression. The unidirectional optimization fluctuation coefficient is calculated by combining the coefficient of variation and the downward trend. 3) Two-way indicators: The maximum and minimum values are determined based on historical water quality data corresponding to the feature database of aquaculture species and growth stages. The deviation index is calculated based on the maximum and minimum values and the length of the time window. The trend deviation factor of the water quality data in time period T is fitted by linear regression. The two-way index fluctuation coefficient is calculated by combining the deviation index and the trend deviation factor. (3) Dynamic grading threshold: The benchmark threshold and stage correction coefficient of static grading are fused with the three types of volatility coefficients in dynamic grading to obtain the dynamic grading threshold. S4. Based on the analytic hierarchy process and the integration of ecological causal weights, a comprehensive weight is generated. The water quality data and dynamic classification thresholds are simultaneously processed for differential standardization. The comprehensive pollution index is obtained by weighted calculation of pollution contribution. The comprehensive classification threshold is obtained by weighted calculation of dynamic classification threshold. The comprehensive pollution index is then used to classify water quality based on the comprehensive classification threshold. S5. Construct an XGBoost gradient boosting tree prediction model, calculate the load level based on the comprehensive pollution index, dissolved oxygen concentration, breeding density and comprehensive grading threshold, calculate the sensor drift coefficient based on sensor detection value and standard detection value, and then execute the load-cleaning linkage adaptive algorithm based on the load level and sensor drift coefficient to dynamically adjust the sensor cleaning parameters.
2. The automatic grading and quality-separating treatment method for aquaculture wastewater according to claim 1, characterized in that, S1 is as follows: Water quality sensors are submerged and deployed at the drainage outlets and sedimentation tanks of aquaculture ponds. Adjust the data collection frequency according to the breeding situation and environmental changes; Water quality is classified into three levels: slightly safe, moderately warning, and severely dangerous. A dynamic labeling mechanism is adopted, in which professionals in the field classify the water quality of each set of collected water quality data. The labeling is based on seasonality, compound pollution, and sensor calibration results.
3. The automatic grading and quality-separating treatment method for aquaculture wastewater according to claim 1, characterized in that, S2 is as follows: Based on the theory of ecological aquaculture, the causal relationship between various water quality data is determined. The Pearson correlation coefficient between various water quality data is calculated through historical data to obtain the correlation strength between various water quality data. After normalizing the causal relationship and the Pearson correlation coefficient, the data are then fused using the weighted average method to obtain the ecological causal weight factor between any two water quality data, and thus obtain the ecological causal weight matrix. Extract all ecological causal weight factors related to missing or outlier values from the ecological causal weight matrix, multiply them by the standardized water quality data of the same sample (both non-missing and normal), sum the multiplication results and divide by the sum of all ecological causal weight factors related to missing or outlier values to obtain the imputation or replacement values, thus obtaining the preprocessed dataset.
4. The automatic grading and quality-separating treatment method for aquaculture wastewater according to claim 1, characterized in that, The specific operation of S4 is as follows: A priority system for the toxic effects of water quality indicators was constructed. The priority weights of each water quality data were calculated by combining the analytic hierarchy process with the priority system. The priority weights were then integrated with the ecological causal weight factors in the ecological causal weight matrix to generate a comprehensive weight. Differential standardization was performed on the water quality data of each attribute after preprocessing, and differential standardization was also performed on the dynamic grading thresholds of each water quality data. The pollution contribution of each water quality data point is calculated based on the standardized water quality data and the corresponding comprehensive weight. The pollution contribution of each water quality data point is added together to obtain the threshold-coupled comprehensive pollution index. The comprehensive grading threshold is obtained by weighted summation based on the dynamic grading thresholds and corresponding comprehensive weights after standardization of various water quality data. Water quality is classified based on a threshold-coupled comprehensive pollution index using a comprehensive classification threshold.
5. The automatic grading and quality-separating treatment method for aquaculture wastewater according to claim 1, characterized in that the load... The calculation process for the degree is as follows: The comprehensive pollution index, dissolved oxygen concentration, aquaculture density, and comprehensive grading threshold were normalized to eliminate the influence of dimensions. Ratio features were constructed, with the ratio of dissolved oxygen concentration to comprehensive pollution index as the self-purification capacity pressure ratio and the ratio of aquaculture density to comprehensive grading threshold as the ecological carrying capacity pressure ratio. In addition, the water quality data corresponding to the two largest ecological causal weight factors in the ecological causal weight matrix, excluding dissolved oxygen, were selected as additional input features. The XGBoost gradient boosting tree prediction model is trained based on the comprehensive pollution index, dissolved oxygen concentration, aquaculture density, comprehensive grading threshold, historical water quality data, and corresponding load levels. The residual gradient is iteratively calculated, the decision tree nodes are split to maximize the gain, and the prediction results are updated to obtain the trained XGBoost gradient boosting tree prediction model. The comprehensive pollution index, dissolved oxygen concentration, aquaculture density, comprehensive grading threshold, ratio features, and additional input features are then input into the trained XGBoost gradient boosting tree prediction model to obtain the load index.
6. The automatic grading and quality-separating treatment method for aquaculture wastewater according to claim 5, characterized in that, The calculation process for the sensor drift coefficient is as follows: While the sensor collects water quality data, the aquaculture wastewater is tested using laboratory standard methods to obtain standard values. The ratio of the difference between the sensor detection value and the standard value to the standard value is calculated to obtain the instantaneous drift coefficient. Then, a time decay factor and a load correlation factor are introduced to dynamically correct the instantaneous drift coefficient. The time decay factor is obtained by fitting the sensor's natural decay characteristics over time with historical data, and the load correlation factor is obtained by mapping the load index through the Sigmoid activation function. Finally, the dynamic sensor drift coefficient is obtained.
7. The automatic grading and quality-separating treatment method for aquaculture wastewater according to claim 6, characterized in that, The load-cleaning linkage adaptive cleaning strategy is as follows: The sensor cleaning trigger value is obtained by weighting the load index and the dynamic sensor drift coefficient. A preset cleaning degree strategy is then implemented, with different cleaning degrees for different sensor cleaning trigger values. The cleaning degree strategy includes cleaning frequency, cleaning duration, and cleaning intensity.
8. An automatic grading and quality-separating treatment device for aquaculture wastewater, executing an automatic grading and quality-separating treatment method for aquaculture wastewater as described in any one of claims 1 to 7, comprising a main unit, a multi-parameter water quality sensor, and an ultrasonic cleaning ring, characterized in that, Includes the following units: Real-time water quality monitoring unit: includes a multi-parameter water quality sensor. The sensor body is a cylindrical structure and is installed at the drainage outlet of the aquaculture pond and the sedimentation tank via a 304 stainless steel mounting bracket. The depth of the sensor probe can be changed by adjusting the adjustable bolts on the stainless steel mounting bracket. The sensor data cable is inserted into the sensor interface on the back of the chassis through a waterproof connector to establish a connection with the data acquisition and transmission module inside the host. Data transmission is carried out via RS485 wired connection. An ultrasonic cleaning ring is set around the sensor probe, and the ultrasonic cleaning ring is connected to the central controller. Central control and data processing unit: The main unit integrates a PLC controller, data acquisition and transmission module, edge computing module and power supply module. A rectangular operation window with an operation display screen is opened on the front of the chassis. Graded and differentiated sewage discharge valve assembly: includes 3 electric sewage discharge valves, each valve is equipped with an independent electric actuator, and the inlet end of each of the 3 sewage discharge valves is connected to the main outlet pipe of the aquaculture pond through the sewage discharge valve pipe; Power supply and data export components: One end of the power cord is inserted into the main unit, and the other end is connected to the power socket in the aquaculture farm; historical water quality data is exported from the data acquisition and transmission module of the main unit.
9. The automatic grading and quality-separating treatment device for aquaculture wastewater according to claim 8, characterized in that, The specific operating procedures for the device are as follows: After the power is turned on, the host is started and initialized. The PLC controller receives multi-parameter sensor data stored in the data acquisition and transmission module, summarizes and packages the data and sends it to the edge computing module. On the host operation display screen, multiple parameter weighting coefficients, breeding species, and initial release date are set by operation. The host system automatically loads the corresponding species' feature library and dynamic grading threshold parameters. After preprocessing the received data, the edge computing module calculates the comprehensive pollution index, compares it with the dynamic grading threshold parameters, and determines the pollution level of the effluent in real time. Based on the pollution level, it activates the corresponding electric drain valve. It further calculates the load level and sensor drift coefficient to control the operation of the ultrasonic cleaning ring.