Intelligent control method and system for tail water treatment of saline-alkali land water-based recirculating aquaculture system based on multi-stage purification linkage

Abstract

The invention belongs to the technical field of water treatment, and particularly relates to an intelligent control method and system for tail water treatment of a saline-alkali land water-based circulating aquaculture system based on multi-stage purification linkage. Comprising the following steps: collecting original water quality data through intelligent water quality monitoring stations arranged at a water inlet and a water outlet of each purification unit, and preprocessing the original water quality data to obtain preprocessed water quality data; a principal component analysis method is adopted for extracting principal components with the accumulative contribution rate reaching a preset threshold value or above from the preprocessed water quality data, feature dimension reduction is conducted, and a fused water quality feature vector is obtained; according to the method, a data fusion method combining Kalman filtering and principal component analysis is adopted, the interference of sensor noise and data missing is eliminated, the reliability of water quality monitoring data is improved, meanwhile, credible data support is provided for intelligent decision making, and the purification efficiency is evaluated through a dynamic weight calculation method based on the deviation degree.

Description

Technical Field

[0001] This invention belongs to the field of water treatment technology, specifically a smart control method and system for wastewater treatment of saline-alkali land-based recirculating aquaculture systems based on multi-stage purification linkage. Background Technology

[0002] Land-based recirculating aquaculture systems are widely used due to their small footprint, high stocking density, and strong environmental controllability. In saline-alkali areas, utilizing saline-alkali water for aquaculture can effectively develop and utilize saline-alkali water resources and achieve comprehensive land use. However, the wastewater produced by saline-alkali aquaculture contains pollutants such as ammonia nitrogen, nitrite, suspended organic matter, and uneaten feed and feces. Direct discharge without treatment will cause eutrophication pollution of the water body. Currently, aquaculture wastewater treatment technologies mainly include physical, chemical, and biological methods. Physical methods remove suspended particulate matter through sedimentation and filtration; chemical methods use flocculants and disinfectants for water purification; and biological methods utilize microorganisms, aquatic plants, and filter-feeding organisms to absorb and transform nutrients. In practical applications, multi-stage purification processes are often used.

[0003] Existing aquaculture-farm integrated industrialized aquaculture wastewater treatment systems purify wastewater through fish, shrimp, and crab farming, shellfish filter feeding, and salt-tolerant plant absorption. However, these systems lack effective coordination and control mechanisms among the various treatment units, relying primarily on natural ecological processes. Hydraulic retention time and treatment load are difficult to dynamically adjust based on actual water quality changes. Pond-based ecological recirculation industrialized aquaculture systems offer advantages in solid waste resource utilization, but lack systematic technical means for monitoring purification efficiency and controlling flow rates, making it difficult to achieve synergistic optimization between purification units. Methods utilizing filter-feeding organisms such as sea cucumbers and abalone to treat industrialized aquaculture wastewater are unique in their biological filter feeding capabilities, but control methods are relatively passive, lacking real-time water quality monitoring and intelligent regulation. Existing intelligent water quality control systems based on multimodal sensing mainly target water quality control within the aquaculture area, failing to address the coordinated control between multi-level purification units and neglecting the special adaptability requirements of purification systems in saline-alkali water environments.

[0004] In summary, existing recirculating aquaculture wastewater treatment technologies suffer from several problems. Each purification unit operates independently with fixed operating parameters, lacks a linkage control mechanism based on real-time water quality feedback, lacks the ability to predict the future state of the multi-stage purification system, and cannot take preventive control measures in advance. As a result, the system cannot adjust the hydraulic distribution and treatment load in a timely manner according to fluctuations in influent water quality or changes in the efficiency of purification units. Summary of the Invention

[0005] (1) Technical problems to be solved The purpose of this invention is to provide an intelligent control method and system for tailwater treatment of a land-based recirculating aquaculture system based on multi-stage purification linkage, so as to solve the problem that it is impossible to adjust the hydraulic distribution and treatment load in a timely manner according to the fluctuation of influent water quality or the change of purification unit efficiency in a multi-stage purification system.

[0006] (2) Technical solution To achieve the above objectives, on the one hand, this invention provides an intelligent control method for wastewater treatment in a land-based recirculating aquaculture system in saline-alkali water based on multi-stage purification linkage, comprising: Step S1: Collect raw water quality data through the intelligent water quality monitoring stations set at the inlet and outlet of each purification unit. Preprocess the raw water quality data to obtain preprocessed water quality data. Use principal component analysis to extract principal components with a cumulative contribution rate of more than a preset threshold from the preprocessed water quality data and perform feature dimensionality reduction to obtain the fused water quality feature vector.

[0007] Step S2: Calculate the removal efficiency of each purification unit for each pollutant based on the concentration of pollutants in the influent and effluent of each purification unit. Calculate the dynamic weight of each pollutant index based on the current water quality status using an adaptive adjustment method based on deviation. Calculate the weighted sum of the removal efficiency and the dynamic weight to obtain the comprehensive purification efficiency of each purification unit. Calculate the purification unit health index using the health assessment formula based on the comprehensive purification efficiency, the rate of change of purification efficiency, and the continuous operating time.

[0008] Step S3: The fused water quality feature vector sequence of the past n time steps is used as input. Temporal feature learning is performed through a long short-term memory network model to output the predicted value sequence of key water quality parameters for the next m time steps. When the prediction results show that the key water quality parameters will exceed the warning threshold, the preventive control mode is triggered. The key water quality parameters include ammonia nitrogen concentration and chemical oxygen demand.

[0009] Step S4: Obtain the water salinity based on the current conductivity, obtain the current pH value based on the pH sensor, calculate the bioactivity coefficient through the salinity-pH coupling effect bioactivity inhibition model, determine the inhibition interval based on the bioactivity coefficient and the preset inhibition interval threshold, and determine the corresponding graded adaptive adjustment strategy based on the determination result.

[0010] Step S5: The health index of the purification unit, the predicted value of water quality parameters, and the level of the regulation strategy are used as the initial conditions for the model predictive control optimization problem. An optimization objective function is established with the goal of maximizing purification efficiency and minimizing energy consumption. A set of constraints is set, including hydraulic balance constraints, upper and lower limits of flow, water quality compliance constraints, and control variable change rate constraints. The optimal control variable vector is obtained by solving the problem using a sequential quadratic programming algorithm.

[0011] Step S6: Convert the optimal control variable vector into the opening setpoint of each electric valve and the frequency setpoint of each water pump. Output control signals to the actuator through the PLC controller. The control signals adopt a ramp-type gradual output mode to avoid hydraulic impact.

[0012] Furthermore, the step of extracting principal components with a cumulative contribution rate exceeding a preset threshold from the pretreated water quality data using principal component analysis to obtain a fused water quality feature vector includes: The intelligent water quality monitoring station uses a multi-parameter water quality sensor array to collect raw data vectors according to a preset collection cycle. The raw data vector includes temperature, pH, dissolved oxygen, conductivity, turbidity, ammonia nitrogen, nitrite, nitrate, total nitrogen, total phosphorus, and chemical oxygen demand.

[0013] Each parameter value in the original data vector is compared with the mean and standard deviation of that parameter within a sliding window. When the deviation of the parameter value from the mean exceeds 3 times the standard deviation, the data point is marked as an outlier.

[0014] For data points marked as outliers or missing values, a Kalman filter is used for state estimation and data imputation. The state equation of the Kalman filter is: ; The observation equation is: ; in, Here is the state transition matrix. To control the input matrix, For control vectors, For process noise, For the observation matrix, To observe the noise, the optimal state estimate is obtained through iterative calculation of the prediction and update steps of the Kalman filter and used as the fill value.

[0015] A covariance matrix is ​​constructed from the preprocessed multi-source water quality data and eigenvalue decomposition is performed. The eigenvalues ​​are sorted from largest to smallest, and the top principal components with a cumulative contribution rate of over 95% are selected. The original data vector is projected onto the direction of the selected principal components to obtain the fused water quality feature vector.

[0016] Furthermore, step S2 includes: Based on the pollutant concentration data collected by the monitoring stations at the inlet and outlet of each purification unit, the formula is used: ; Calculation yields the first The purification unit for the first Removal efficiency of various pollutants ,in, The influent concentration is... This refers to the concentration of the effluent.

[0017] Obtain the basic weights and emission standard limits for each pollutant. Based on the deviation between the measured concentration of each pollutant and the emission standard limit at the current moment, use the formula: ; Calculate dynamic weights ,in, For the first The basic weights of pollutants, For adjustment coefficients, For the current moment The measured concentrations of the pollutants, These are emission standard limits.

[0018] The removal efficiency is weighted and summed with dynamic weights, using the formula: ; Calculation yields the first Overall purification efficiency of each purification unit .

[0019] Obtain the design purification efficiency reference value, design maintenance cycle, and current continuous operating time of the purification unit. Calculate the difference between the current overall purification efficiency and the overall purification efficiency of adjacent data collection cycles to obtain the purification efficiency change rate, using the formula: ; The health index of the purification unit was calculated. ,in To design a reference value for purification efficiency, The rate of change in purification efficiency. For continuous running time, To design maintenance cycles, , , The weighting coefficients are satisfied. .

[0020] Furthermore, step S3 includes: Construct time-series input data, and arrange the fused water quality feature vectors from the past n time steps in chronological order to form a feature vector sequence. As input to the Long Short-Term Memory (LSTM) network model, The value ranges from 12 to 24, corresponding to 1 to 2 hours of historical data.

[0021] For each time step in the input sequence, the current input... Compared to the previous hidden state After concatenation, input the data into the forget gate, and then use the formula: Calculate the output of the forget gate ,in It is the Sigmoid activation function. Here is the forget gate weight matrix. This is the bias vector.

[0022] The concatenated vector is input into the input gate, using the formula: The input gate output is calculated. At the same time, through the formula: Calculate the candidate memory state .

[0023] Based on the output of the forget gate, selective forgetting of the previous memory state is performed; based on the output of the input gate, selective memorization of candidate memory states is performed, using the formula: Update to get the current memory state .

[0024] The concatenated vector is input to the output gate, using the formula: The output of the output gate is calculated. The output of the output gate is compared with the current memory state. Transformation multiplication, using the formula: Calculate the current hidden state ; The hidden state output at the last time step is linearly transformed through a fully connected output layer to map the predicted sequence of key water quality parameters for the next m time steps. ,in The value ranges from 6 to 12, corresponding to a prediction duration of 30 minutes to 1 hour.

[0025] Furthermore, step S4 includes: The current conductivity is collected using a conductivity sensor. The water salinity is obtained based on the conversion relationship between conductivity and salinity; the current pH value is collected using a pH sensor. .

[0026] Obtain the preset optimal salinity for microorganisms Optimal pH value Salinity inhibition coefficient and pH inhibition coefficient The bioactivity coefficient was calculated using a bioactivity inhibition model based on the salinity-pH coupling effect. The calculation formula for the bioactivity inhibition model is: ; in, This represents the microbial activity coefficient under optimal conditions.

[0027] The calculated bioactivity coefficient The system compares the current condition with a preset inhibition threshold to determine the current inhibition interval and the appropriate adjustment strategy level. When it is determined to be a mild inhibition region, when When it is determined to be a moderate inhibition region, when When it is determined to be a severely inhibited region, when It is determined to be in extreme condition protection mode.

[0028] Furthermore, the specific adjustment measures of the hierarchical adaptive adjustment strategy are as follows: If the inhibition range is mild, maintain normal operating parameters and increase the aeration rate of the biological filter to 110% to 120% of the normal value to compensate for the decrease in microbial activity, without adjusting the hydraulic distribution.

[0029] If the water level is in the moderate inhibition range, reduce the influent flow rate of the biological purification unit to 70% to 85% of the normal flow rate to prolong the hydraulic retention time, activate the bypass diversion to directly introduce 15% to 30% of the water flow into the tertiary ecological stabilization unit, and increase the aeration rate to 125% to 135% of the normal value.

[0030] If the water flow is in a severely suppressed zone, more than 50% of the water flow will be diverted to the third-level ecological stabilization unit through a bypass. The return system will be activated to return some of the third-level effluent to the first-level sedimentation tank for dilution to reduce the salinity and alkalinity entering the biological purification unit. The influent flow rate of the biological purification unit will be reduced to less than 50% of the normal value.

[0031] In extreme condition protection mode, water intake to the biological purification unit is suspended, the inlet valve of the biological purification unit is closed, and all water flows through the bypass to the third-level ecological stabilization unit for emergency purification. At the same time, an alarm signal is sent to the monitoring system to prompt maintenance personnel to intervene.

[0032] Furthermore, the process of obtaining the optimal control variable vector through the sequential quadratic programming algorithm includes: Define control variable vector ,in These refer to the flow rates flowing into the primary physical purification unit, the secondary biological purification unit, and the tertiary ecological stabilization unit, respectively. To divert traffic via bypass, This is the return flow.

[0033] Establish the optimization objective function: ; in, For the first The system effluent pollutant concentration vector at each predicted time point For the target effluent concentration vector, To control the amount of change in variables, For system energy consumption, , , These are the weighting coefficients.

[0034] Set a set of constraints, including hydraulic balance constraints. Flow upper and lower limits constraints Water quality compliance constraints and control variable rate of change constraints .

[0035] The constraints are dynamically adjusted according to the level of the regulation strategy, and in the moderate suppression range, the constraints are adjusted accordingly. Reduced to 70% to 85% of normal values, in the severe inhibition range Reduced to below 50% of normal value, in extreme condition protection mode Set to zero.

[0036] The objective function is solved iteratively under constraints using a sequential quadratic programming algorithm. Each iteration involves constructing a quadratic programming subproblem for local optimization until the convergence condition is met and the optimal control variable vector is obtained.

[0037] Furthermore, it also includes steps for effect verification and model update: In the next collection cycle, actual water quality data is acquired, and the actual water quality change value is compared with the predicted value to calculate the prediction deviation. The prediction deviation is compared with a preset deviation threshold. When the prediction deviation exceeds the deviation threshold, the weight parameters of the long short-term memory network model are fine-tuned and updated using an online learning algorithm. At a preset time each day, the daily running data is uploaded to the cloud server. The cloud server periodically retrains the long short-term memory network model offline based on the accumulated running data. After training is completed, the updated model parameters are sent to the edge computing device to replace the original model parameters.

[0038] Based on the same inventive concept, this invention also provides an intelligent control system for wastewater treatment of a land-based recirculating aquaculture system in saline-alkali water based on multi-level purification linkage, including a data acquisition module, a purification analysis module, a regulation and early warning module, a purification strategy module, a treatment decision module, and a system control module.

[0039] The data acquisition module is used to collect raw water quality data through the intelligent water quality monitoring stations set at the inlet and outlet of each purification unit. The raw water quality data is preprocessed to obtain preprocessed water quality data. The principal component analysis method is used to extract the principal components whose cumulative contribution rate reaches a preset threshold from the preprocessed water quality data, and feature dimensionality reduction is performed to obtain the fused water quality feature vector.

[0040] The purification analysis module is used to calculate the removal efficiency of each purification unit for each pollutant based on the concentration of pollutants in the influent and effluent of each purification unit. Based on the current water quality status, it calculates the dynamic weight of each pollutant index using an adaptive adjustment method based on deviation. The removal efficiency and dynamic weight are weighted and summed to obtain the comprehensive purification efficiency of each purification unit. Based on the comprehensive purification efficiency, the rate of change of purification efficiency, and the continuous operating time, the health index of the purification unit is calculated using the health assessment formula.

[0041] The regulation and early warning module takes the fused water quality feature vector sequence of the past n time steps as input, performs temporal feature learning through a long short-term memory network model, and outputs the predicted value sequence of key water quality parameters for the next m time steps. When the prediction results show that the key water quality parameters will exceed the early warning threshold, the preventive regulation mode is triggered.

[0042] The purification strategy module is used to obtain the salinity of the water body based on the current conductivity, obtain the current pH value based on the pH sensor, calculate the bioactivity coefficient through the bioactivity inhibition model of salinity-pH coupling effect, determine the inhibition interval based on the comparison between the bioactivity coefficient and the preset inhibition interval threshold, and determine the corresponding graded adaptive adjustment strategy based on the judgment result.

[0043] The processing decision module uses the purification unit health index, predicted water quality parameters, and regulation strategy level as initial conditions for the model predictive control optimization problem. It establishes an optimization objective function with the goal of maximizing purification efficiency and minimizing energy consumption, and sets a set of constraints including hydraulic balance constraints, flow upper and lower limit constraints, water quality compliance constraints, and control variable change rate constraints. The optimal control variable vector is obtained by solving the problem using a sequential quadratic programming algorithm.

[0044] The system control module is used to convert the optimal control variable vector into the opening setpoint of each electric valve and the frequency setpoint of each water pump. The control signal is output to the actuator through the PLC controller. The control signal adopts a ramp-type gradual output mode to avoid hydraulic impact.

[0045] (3) Beneficial effects Compared with existing technologies, this invention employs a data fusion method combining Kalman filtering and principal component analysis to eliminate interference from sensor noise and missing data, improving the reliability of water quality monitoring data and providing credible data support for intelligent decision-making. Furthermore, it evaluates purification efficiency through a dynamic weight calculation method based on deviation, automatically adjusting the weights of each pollutant indicator according to the degree of deviation between the measured concentration and the emission standard, prioritizing indicators with a higher risk of exceeding standards, and introducing a health index to evaluate the operating status of the purification unit, making it more adaptable than the fixed-weight method. A water quality prediction model is constructed using an LSTM network to predict water quality change trends in advance, realizing a shift from passive response to proactive predictive control, improving the system's resistance to shock loads. A biological activity inhibition model based on salinity-pH coupling effect is established, and a graded adaptive adjustment strategy is formulated accordingly to specifically address the inhibitory effect of saline-alkali water environment on the biological purification process. This solves the technical challenges of intelligent control in saline-alkali water effluent treatment in existing technologies. A model predictive control strategy is used to collaboratively optimize multi-stage purification units, achieving energy consumption optimization while ensuring effluent meets standards, further reducing system energy consumption. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating the intelligent control method for wastewater treatment of a land-based recirculating aquaculture system in saline-alkali water based on multi-level purification linkage, as described in an embodiment of the present invention. Figure 2 This is a block diagram of the intelligent control system module for tailwater treatment of a land-based recirculating aquaculture system based on multi-level purification linkage, as described in an embodiment of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Throughout this specification, references to "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "an embodiment," "an example," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0050] like Figure 1 As shown, this embodiment uses a saline-alkali land-based recirculating aquaculture system as an example to describe in detail the intelligent control method for wastewater treatment of the saline-alkali land-based recirculating aquaculture system based on multi-stage purification linkage. The farm cultivates Litopenaeus vannamei, the salinity of the culture water is 8‰ to 15‰, the pH value is 7.8 to 8.6, and the daily wastewater discharge is approximately 500 cubic meters.

[0051] The tailwater treatment system is configured as follows: The primary physical purification unit uses an inclined tube sedimentation tank and a rotary drum microfilter connected in series. The effective volume of the inclined tube sedimentation tank is 120 cubic meters, the microfilter has a filtration accuracy of 60 microns, and the designed hydraulic retention time is 1.5 to 3 hours.

[0052] The secondary biological purification unit adopts an aerated biological filter with an effective volume of 180 cubic meters. The packing material is a mixture of volcanic rock and ceramsite. The hydraulic retention time is designed to be 3 to 6 hours. It is equipped with two Roots blowers, and the aeration volume can be adjusted by frequency conversion.

[0053] The third-level ecological stability unit adopts a parallel configuration of subsurface flow artificial wetland and surface flow ecological pond. The wetland area is 800 square meters, planted with Suaeda salsa, reeds and calamus; the ecological pond area is 500 square meters, stocked with filter-feeding shellfish such as oysters and clams, and designed hydraulic retention time of 6 to 12 hours.

[0054] The monitoring and control equipment is configured as follows: Water quality monitoring station configuration: Four water quality monitoring stations are set up at the primary inlet, primary outlet, secondary outlet, and tertiary outlet. Each monitoring station is equipped with one multi-parameter water quality instrument, including sensors for temperature, pH, dissolved oxygen, conductivity, and turbidity; one online ammonia nitrogen analyzer; one online COD analyzer; and one online total nitrogen analyzer. The data acquisition cycle is set to 5 minutes.

[0055] Actuator configuration: Each pipeline node is equipped with an electric butterfly valve with an adjustable valve opening range of 0% to 100%; each pipeline is equipped with an electromagnetic flow meter; each water pump is equipped with a frequency converter.

[0056] Control system configuration: The edge layer uses one industrial-grade edge computing gateway, equipped with an Intel Core i7 processor, 16GB of memory, and a 512GB solid-state drive, running a Linux operating system, and deploying a data preprocessing module, an LSTM prediction model, and an MPC optimized control algorithm; it is also equipped with one PLC controller, which communicates with the edge gateway via the Modbus protocol and outputs 4 to 20mA analog control signals.

[0057] Step S1: Multi-source heterogeneous water quality data fusion and acquisition Each monitoring station collects water quality data every 5 minutes. After receiving the raw data, the edge gateway first performs outlier detection. For the first... Raw data vectors collected at different times Set the sliding window length to 12 acquisition cycles and calculate the mean value of each parameter within the window. and standard deviation When the deviation of a parameter value from the mean exceeds... If the data point is an outlier, mark it as such.

[0058] For data points marked as outliers or missing values, a Kalman filter is used for state estimation and data imputation. The state equation of the Kalman filter is: The observation equation is In this embodiment, Here is the state transition matrix. To control the input matrix, For control vectors, For process noise, For the observation matrix, For observation noise, where the state transition matrix is... Take the identity matrix and the observation matrix. Take the identity matrix and the process noise covariance matrix. and observation noise covariance matrix The optimal state estimate is determined based on historical data statistics. Iterative calculations using the prediction and update steps of a Kalman filter are employed to obtain the estimate, which is then used as an imputation value for outliers or missing values.

[0059] A covariance matrix was constructed from the preprocessed 11-dimensional water quality data, and eigenvalue decomposition was performed, with the eigenvalues ​​sorted from largest to smallest. In this embodiment, the cumulative contribution rate of the first 5 principal components reached 96.3%, so the first 5 principal components were selected, and the original 11-dimensional data vector was projected onto the 5-dimensional principal component space to obtain the fused water quality feature vector.

[0060] Step S2: Multi-target purification efficiency assessment Based on the pollutant concentrations in the influent and effluent of each purification unit, the formula is used to determine the concentrations. The removal efficiency of each pollutant was calculated. This example focuses on four major pollutant indicators: ammonia nitrogen, COD, total nitrogen, and total phosphorus.

[0061] Basic weights of each pollutant Pre-set adjustment coefficients based on emission standards and pollutant hazard levels: ammonia nitrogen 0.35, COD 0.25, total nitrogen 0.25, total phosphorus 0.15. The value is set to 1.0. Based on the deviation of the measured concentration of each pollutant from the emission standard limit at the current moment, the formula is used... Calculate dynamic weights. The more severe the exceedance of a certain pollutant concentration, the greater its weight, and the system prioritizes the removal effect of that pollutant.

[0062] The removal efficiency is weighted and summed with dynamic weights, using the formula... The overall purification efficiency of each purification unit was calculated. At the same time, a design purification efficiency reference value is set. 85%, design maintenance cycle The period is 30 days, with weighting coefficients α=0.5, β=0.3, and γ=0.2, calculated using the formula... The health index of the purification unit was calculated. .

[0063] Step S3: Prediction of Water Quality Change Trends The merged water quality feature vectors from the most recent 12 periods (corresponding to 1 hour) are arranged in chronological order to form a feature vector sequence. As input to the LSTM model. In this embodiment, the LSTM network contains two layers of LSTM units, each with 64 hidden units, followed by a fully connected output layer.

[0064] For each time step of the input sequence, the LSTM unit performs the following computation: first, through the forget gate... Determine how much information to forget from the previous memory state; then, through the input gate... and candidate memories Decide how much new information to add to the memory state; then through Update the memory state; finally, pass through the output gate. and Calculate the current hidden state output.

[0065] The hidden state output at the last time step is linearly transformed through a fully connected output layer to map the predicted ammonia nitrogen and COD values ​​for the next 6 periods (corresponding to 30 minutes). When the forecast results show that ammonia nitrogen or COD in the water will exceed the warning threshold (set at 80% of the emission standard), the preventive control mode is triggered.

[0066] The LSTM model training method in this embodiment is as follows: Historical running data is used as the training set, with a minimum of 30 days of continuous running data, corresponding to at least 8640 sample points (calculated based on a 5-minute acquisition cycle). The mean squared error (MSE) is used as the loss function during training, and the Adam optimizer is used for parameter updates. The initial learning rate is set to 0.001, the batch size to 32, and the number of training epochs to 100. The model weight parameters are initialized using the Xavier initialization method. After training, the model performance is evaluated on the validation set. The model training is considered complete when the root mean squared error (RMSE) on the validation set is lower than a preset threshold.

[0067] Step S4: Adaptive Assessment of Saline-Alkaline Water Environment The current conductivity (EC) is collected by a conductivity sensor, and the salinity of the water is obtained based on the conversion relationship between conductivity and salinity. In this embodiment, the optimal salinity for microorganisms is set. The optimal pH value is 10‰. The optimal parameter is 8.0. This parameter was determined based on the water quality conditions corresponding to the highest purification efficiency of the biological filter in actual operation of the aquaculture farm. Salinity inhibition coefficient. The pH inhibition coefficient is 0.015. The inhibition coefficient is 0.8. This coefficient was obtained by fitting a Gaussian function to measured data on ammonia nitrogen removal efficiency in biological filters under different salinity and pH conditions. Those skilled in the art can determine the appropriate parameter range through experiments based on specific aquaculture water quality conditions and microbial population characteristics. Generally speaking, the salinity inhibition coefficient... The value ranges from 0.01 to 0.03, and the pH inhibition coefficient is... The value ranges from 0.5 to 1.5. The microbial activity coefficient under optimal conditions... A value of 1 indicates that the microbial activity reaches its maximum value under optimal conditions.

[0068] Bioactivity inhibition model based on salinity-pH coupling effect The bioactivity coefficient was calculated. The model shows that microbial activity will decline exponentially when salinity or pH deviates from its optimum.

[0069] The calculated bioactivity coefficient Compare with the preset suppression interval threshold: when When the condition is determined to be in a mild inhibition zone, maintain normal operating parameters and increase the aeration rate to 115% of the normal value; when When the condition was determined to be in the moderate inhibition range, the influent flow rate of the biological purification unit was reduced to 78% of the normal value, a 22% bypass diversion was activated, and the aeration rate was increased to 130%. When the condition is determined to be in a severe inhibition zone, a 55% bypass diversion is initiated, the influent flow rate is reduced to 45%, and reflux dilution is simultaneously activated; when or When the water level is >20‰ or the pH value is >9.0, it is determined to be in extreme condition protection mode, the water intake of the biological purification unit is suspended, all water flows are bypassed to the third-level ecological stabilization unit, and an alarm signal is issued.

[0070] Step S5: Collaborative Optimization and Regulation Define control variable vector In this embodiment, the total influent flow rate under normal operating conditions is... The flow rate is approximately 35 cubic meters per hour, with the upper and lower limits of the flow rate for each purification unit set according to the equipment capacity.

[0071] Establish the optimization objective function The weighting coefficient , , Target effluent concentration vector According to the emission standards, the following parameters are set: ammonia nitrogen ≤ 3 mg / L, COD ≤ 50 mg / L, total nitrogen ≤ 15 mg / L, and total phosphorus ≤ 1 mg / L.

[0072] Set a set of constraints, including hydraulic balance constraints. Flow upper and lower limits constraints Water quality compliance constraints and control variable rate of change constraints cubic meters per hour per cycle. The constraints are dynamically adjusted according to the adjustment strategy level determined in step S4.

[0073] A sequential quadratic programming algorithm is used to iteratively optimize the objective function under constraints. The control period is set to 5 minutes. Each iteration involves constructing a quadratic programming subproblem for local optimization until the convergence condition is met (the change in the objective function is less than 0.001 or the number of iterations reaches 50) to obtain the optimal control variable vector.

[0074] Step S6: Output control command The flow rate values ​​in the optimal control variable vector are converted into the corresponding electric butterfly valve opening setpoints and variable frequency pump frequency setpoints. A 4-20mA analog control signal is output via the PLC controller. The control signal uses a ramp-type gradual output, limiting the valve opening change rate to within 10% per minute to avoid the impact of hydraulic shock caused by rapid valve action on system stability.

[0075] Step S7: Validation of Results and Model Update In the next data collection cycle, actual water quality data is acquired, and the actual water quality change value is compared with the predicted value in step S3 to calculate the prediction deviation. The prediction deviation is compared with a preset deviation threshold (set to 15% of the predicted value). When the prediction deviation exceeds the threshold, the Adam optimizer is used to fine-tune the parameters of the LSTM model online, with the learning rate set to 0.0001 and the number of iterations per update being 10.

[0076] The daily operational data is uploaded to the cloud server via the MQTT protocol at 2:00 AM every day. The cloud server performs offline retraining on the accumulated data once a week. After the training is completed, the updated model parameters are packaged and sent to the edge computing gateway to replace the original model.

[0077] Example 2: Inlet shock load condition When the influent ammonia nitrogen concentration suddenly increases after feeding or water change, for example, from 5 mg / L to 15 mg / L, the LSTM model predicts that the effluent ammonia nitrogen will exceed the standard. The system automatically takes the following control measures: MPC optimization results show that the total influent flow rate needs to be reduced to 70% of the normal value (approximately 24.5 cubic meters / hour), the aeration rate of the biological filter needs to be increased to 90% of the maximum value, 20% bypass diversion (approximately 7 cubic meters / hour) needs to be initiated to the tertiary ecological pond, and a portion of the tertiary effluent (approximately 5 cubic meters / hour) needs to be returned to the primary sedimentation tank for dilution. Control commands are output to each actuator via PLC, and the valve opening is gradually adjusted in a ramp manner. After about 30 minutes of adjustment, the system effluent ammonia nitrogen concentration stabilizes at 2.8 mg / L, which is lower than the discharge standard limit of 3 mg / L.

[0078] Example 3: High Salinity Conditions When the salinity of the water increases to 18‰ due to evaporation or other reasons, exceeding the normal range of 8‰ to 15‰, the bioactivity coefficient A is calculated using a bioactivity inhibition model. m =1×exp[-0.015×(18-10)²]×exp[-0.8×(8.2-8.0)²]=0.49, which is in the moderate inhibition range. The system automatically takes the following control measures: reducing the influent flow rate of the secondary biological filter to 75% of the normal value (about 20 cubic meters / hour), starting a 25% bypass diversion (about 8.75 cubic meters / hour) to directly enter the tertiary ecological stabilization unit, increasing the opening of the influent valve of the constructed wetland area to extend the hydraulic retention time to 10 hours, and increasing the aeration rate to 130% of the normal value. The system maintains this operating mode until the salinity returns to normal, continuously monitoring the effluent water quality to ensure that the effluent ammonia nitrogen, COD and other indicators stably meet the standards.

[0079] The operational results of this embodiment show that under normal operating conditions, the system's effluent ammonia nitrogen concentration is stable at 2.0 to 2.8 mg / L, COD concentration at 35 to 45 mg / L, total nitrogen concentration at 10 to 13 mg / L, and total phosphorus concentration at 0.6 to 0.9 mg / L, all of which are better than the emission standard limits. Under influent shock load conditions, the system can recover stability within 30 minutes, and the effluent quality does not exceed the standards. Under high salinity conditions, through a graded adaptive adjustment strategy, the system can effectively cope with the inhibitory effect of salinity fluctuations on the biological purification unit.

[0080] like Figure 2 As shown in the figure, this embodiment also provides an intelligent control system for tailwater treatment of a land-based recirculating aquaculture system based on multi-level purification linkage, including a data acquisition module, a purification analysis module, a regulation and early warning module, a purification strategy module, a treatment decision module, and a system control module.

[0081] The data acquisition module is used to collect raw water quality data through the intelligent water quality monitoring stations set at the inlet and outlet of each purification unit. The raw water quality data is preprocessed to obtain preprocessed water quality data. The principal component analysis method is used to extract the principal components whose cumulative contribution rate reaches a preset threshold from the preprocessed water quality data, and feature dimensionality reduction is performed to obtain the fused water quality feature vector.

[0082] The purification analysis module is used to calculate the removal efficiency of each purification unit for each pollutant based on the concentration of pollutants in the influent and effluent of each purification unit. Based on the current water quality status, it calculates the dynamic weight of each pollutant index using an adaptive adjustment method based on deviation. The removal efficiency and dynamic weight are weighted and summed to obtain the comprehensive purification efficiency of each purification unit. Based on the comprehensive purification efficiency, the rate of change of purification efficiency, and the continuous operating time, the health index of the purification unit is calculated using the health assessment formula.

[0083] The regulation and early warning module takes the fused water quality feature vector sequence of the past n time steps as input, performs temporal feature learning through a long short-term memory network model, and outputs the predicted value sequence of key water quality parameters for the next m time steps. When the prediction results show that the key water quality parameters will exceed the early warning threshold, the preventive regulation mode is triggered.

[0084] The purification strategy module is used to obtain the salinity of the water body based on the current conductivity, obtain the current pH value based on the pH sensor, calculate the bioactivity coefficient through the bioactivity inhibition model of salinity-pH coupling effect, determine the inhibition interval based on the comparison between the bioactivity coefficient and the preset inhibition interval threshold, and determine the corresponding graded adaptive adjustment strategy based on the judgment result.

[0085] The processing decision module uses the purification unit health index, predicted water quality parameters, and regulation strategy level as initial conditions for the model predictive control optimization problem. It establishes an optimization objective function with the goal of maximizing purification efficiency and minimizing energy consumption, and sets a set of constraints including hydraulic balance constraints, flow upper and lower limit constraints, water quality compliance constraints, and control variable change rate constraints. The optimal control variable vector is obtained by solving the problem using a sequential quadratic programming algorithm.

[0086] The system control module is used to convert the optimal control variable vector into the opening setpoint of each electric valve and the frequency setpoint of each water pump. The control signal is output to the actuator through the PLC controller. The control signal adopts a ramp-type gradual output mode to avoid hydraulic impact.

[0087] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0088] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart control method for wastewater treatment in a land-based recirculating aquaculture system for saline-alkali water based on multi-stage purification linkage, characterized in that, The method includes the following steps: Step S1: Collect raw water quality data through the intelligent water quality monitoring stations set at the inlet and outlet of each purification unit, preprocess the raw water quality data to obtain preprocessed water quality data, and use principal component analysis to extract principal components with a cumulative contribution rate of more than a preset threshold to perform feature dimensionality reduction on the preprocessed water quality data to obtain the fused water quality feature vector. Step S2: Calculate the removal efficiency of each purification unit for each pollutant based on the concentration of pollutants in the influent and effluent of each purification unit. Calculate the dynamic weight of each pollutant index using an adaptive adjustment method based on deviation degree according to the current water quality status. Weight the removal efficiency and the dynamic weight to obtain the comprehensive purification efficiency of each purification unit. Calculate the purification unit health index using the health assessment formula based on the comprehensive purification efficiency, the rate of change of purification efficiency, and the continuous operating time. Step S3: The fused water quality feature vector sequence of the past n time steps is used as input. Temporal feature learning is performed through a long short-term memory network model to output the predicted value sequence of key water quality parameters for the next m time steps. When the prediction results show that the key water quality parameters will exceed the warning threshold, the preventive control mode is triggered. Step S4: Obtain the water salinity based on the current conductivity, obtain the current pH value based on the pH sensor, calculate the bioactivity coefficient through the bioactivity inhibition model of salinity-pH coupling effect, determine the inhibition interval based on the comparison between the bioactivity coefficient and the preset inhibition interval threshold, and determine the corresponding graded adaptive adjustment strategy based on the determination result. Step S5: The health index of the purification unit, the predicted value of water quality parameters, and the level of adjustment strategy are used as the initial conditions for the model predictive control optimization problem. An optimization objective function is established with the goal of maximizing purification efficiency and minimizing energy consumption. A set of constraints is set, including hydraulic balance constraints, upper and lower limits of flow, water quality compliance constraints, and control variable change rate constraints. The optimal control variable vector is obtained by solving the problem using a sequential quadratic programming algorithm. Step S6: Convert the optimal control variable vector into the opening setpoint of each electric valve and the frequency setpoint of each water pump. Output control signals to the actuator through the PLC controller. The control signals adopt a ramp-type gradual output mode to avoid hydraulic impact.

2. The intelligent control method for wastewater treatment of a land-based recirculating aquaculture system in saline-alkali water based on multi-stage purification linkage according to claim 1, characterized in that, The process of extracting principal components with a cumulative contribution rate exceeding a preset threshold from the pretreated water quality data using principal component analysis to obtain a fused water quality feature vector includes: The intelligent water quality monitoring station uses a multi-parameter water quality sensor array to collect raw data vectors according to a preset collection cycle. The raw data vector includes temperature, pH, dissolved oxygen, conductivity, turbidity, ammonia nitrogen, nitrite, nitrate, total nitrogen, total phosphorus, and chemical oxygen demand; Each parameter value in the original data vector is compared with the mean and standard deviation of that parameter within a sliding window. When the deviation of the parameter value from the mean exceeds 3 times the standard deviation, the data point is marked as an outlier. For data points marked as outliers or missing values, a Kalman filter is used for state estimation and data imputation. The state equation of the Kalman filter is: ； The observation equation is: ； in, Here is the state transition matrix. To control the input matrix, For control vectors, For process noise, For the observation matrix, To observe the noise, the optimal state estimate is obtained through iterative calculation of the prediction and update steps of the Kalman filter and used as the fill value; A covariance matrix is ​​constructed from the preprocessed multi-source water quality data and eigenvalue decomposition is performed. The eigenvalues ​​are sorted from largest to smallest, and the top principal components with a cumulative contribution rate of over 95% are selected. The original data vector is projected onto the direction of the selected principal components to obtain the fused water quality feature vector.

3. The intelligent control method for wastewater treatment of a land-based recirculating aquaculture system in saline-alkali water based on multi-stage purification linkage according to claim 2, characterized in that, Step S2 includes: Based on the pollutant concentration data collected by the monitoring stations at the inlet and outlet of each purification unit, the formula is used: ； Calculate the first The purification unit for the first Removal efficiency of various pollutants ,in, The concentration of the influent is... This refers to the concentration of the effluent. Obtain the basic weights and emission standard limits for each pollutant. Based on the deviation between the measured concentration of each pollutant and the emission standard limit at the current moment, use the formula: ； Calculate dynamic weights ,in, For the first The basic weights of pollutants, For adjustment coefficients, For the current moment, the first The measured concentrations of the pollutants, The emission standard limits; The removal efficiency is weighted and summed with dynamic weights, using the formula: ； Calculate the first Overall purification efficiency of each purification unit ; Obtain the design purification efficiency reference value, design maintenance cycle, and current continuous operating time of the purification unit. Calculate the difference between the current overall purification efficiency and the overall purification efficiency of adjacent data collection cycles to obtain the purification efficiency change rate, using the formula: ； The health index of the purification unit was calculated. ,in To design a reference value for purification efficiency, The rate of change in purification efficiency. For continuous running time, To design maintenance cycles, , , The weighting coefficients are and satisfy the following conditions: .

4. The intelligent control method for wastewater treatment of a land-based recirculating aquaculture system in saline-alkali water based on multi-stage purification linkage according to claim 3, characterized in that, Step S3 includes: Construct time-series input data, and arrange the fused water quality feature vectors from the past n time steps in chronological order to form a feature vector sequence. As input to the Long Short-Term Memory (LSTM) network model, The value ranges from 12 to 24, corresponding to 1 to 2 hours of historical data. For each time step in the input sequence, the current input... Compared to the previous hidden state After concatenation, input the data into the forget gate, and then use the formula: Calculate the output of the forget gate ,in It is the Sigmoid activation function. Here is the forget gate weight matrix. It is the bias vector; The concatenated vector is input into the input gate, using the formula: The input gate output is calculated. At the same time, through the formula: Calculate the candidate memory state ; Based on the output of the forget gate, selective forgetting of the previous memory state is performed; based on the output of the input gate, selective memorization of candidate memory states is performed, using the formula: Update to get the current memory state ; The concatenated vector is input to the output gate, using the formula: The output of the output gate is calculated. The output of the output gate is compared with the current memory state. Transformation multiplication, using the formula: Calculate the current hidden state ; The hidden state output at the last time step is linearly transformed through a fully connected output layer to map the predicted sequence of key water quality parameters for the next m time steps. ,in The value ranges from 6 to 12, corresponding to a prediction duration of 30 minutes to 1 hour.

5. The intelligent control method for wastewater treatment of a land-based recirculating aquaculture system in saline-alkali water based on multi-stage purification linkage according to claim 4, characterized in that, Step S4 includes: The current conductivity is collected using a conductivity sensor. The water salinity is obtained based on the conversion relationship between conductivity and salinity; the current pH value is collected using a pH sensor. ; Obtain the preset optimal salinity for microorganisms Optimal pH value Salinity inhibition coefficient and pH inhibition coefficient The bioactivity coefficient was calculated using a bioactivity inhibition model based on the salinity-pH coupling effect. The calculation formula for the bioactivity inhibition model is: ； in, The coefficient of microbial activity under optimal conditions; The calculated bioactivity coefficient Compared with the preset suppression interval threshold, when When it is determined to be a mild inhibition region, when When it is determined to be a moderate inhibition region, when When it is determined to be a severely inhibited region, when It is determined to be in extreme condition protection mode.

6. The intelligent control method for wastewater treatment of a land-based recirculating aquaculture system in saline-alkali water based on multi-stage purification linkage according to claim 5, characterized in that, The specific adjustment measures of the hierarchical adaptive adjustment strategy are as follows: If it is in the mild inhibition range, maintain the normal operating parameters and increase the aeration rate of the biological filter to 110% to 120% of the normal value to compensate for the impact of decreased microbial activity, without making any hydraulic distribution adjustments; If the inhibition level is moderate, reduce the influent flow rate of the biological purification unit to 70% to 85% of the normal flow rate to prolong the hydraulic retention time, activate the bypass diversion to directly introduce 15% to 30% of the water flow into the tertiary ecological stabilization unit, and increase the aeration rate to 125% to 135% of the normal value; If it is a severely suppressed area, more than 50% of the water flow will be diverted to the third-level ecological stabilization unit through the bypass. The return system will be activated to return part of the third-level effluent to the first-level sedimentation tank for dilution to reduce the salinity and alkalinity entering the biological purification unit. The influent flow rate of the biological purification unit will be reduced to less than 50% of the normal value. In extreme condition protection mode, water intake to the biological purification unit is suspended, the inlet valve of the biological purification unit is closed, and all water flows through the bypass to the third-level ecological stabilization unit for emergency purification. At the same time, an alarm signal is sent to the monitoring system to prompt maintenance personnel to intervene.

7. The intelligent control method for wastewater treatment of a land-based recirculating aquaculture system in saline-alkali water based on multi-stage purification linkage according to claim 6, characterized in that, The optimal control variable vector obtained by solving the sequential quadratic programming algorithm includes: Define control variable vector ,in These refer to the flow rates into the primary physical purification unit, the secondary biological purification unit, and the tertiary ecological stabilization unit, respectively. To divert traffic via bypass, For return flow; Establish the optimization objective function: ； in, For the first The system effluent pollutant concentration vector at each predicted time point Let the target effluent concentration vector be... To control the amount of change in variables, For system energy consumption, , , These are the weighting coefficients; Set a set of constraints, including hydraulic balance constraints. Flow upper and lower limits constraints Water quality compliance constraints and control variable rate of change constraints ; The constraints are dynamically adjusted according to the level of the regulation strategy, and in the moderate suppression range, the constraints are adjusted accordingly. Reduced to 70% to 85% of normal values, in the severe inhibition range Reduced to below 50% of normal value, in extreme condition protection mode Set to zero; The objective function is solved iteratively under constraints using a sequential quadratic programming algorithm. Each iteration involves constructing a quadratic programming subproblem for local optimization until the convergence condition is met and the optimal control variable vector is obtained.

8. The intelligent control method for wastewater treatment of a land-based recirculating aquaculture system in saline-alkali water based on multi-stage purification linkage according to claim 7, characterized in that, It also includes the steps of effect verification and model update: In the next collection cycle, actual water quality data is acquired, and the actual water quality change value is compared with the predicted value to calculate the prediction deviation. The prediction deviation is compared with a preset deviation threshold. When the prediction deviation exceeds the deviation threshold, the weight parameters of the long short-term memory network model are fine-tuned and updated using an online learning algorithm. At a preset time each day, the daily running data is uploaded to the cloud server. The cloud server periodically retrains the long short-term memory network model offline based on the accumulated running data. After training is completed, the updated model parameters are sent to the edge computing device to replace the original model parameters.

9. An intelligent control system for wastewater treatment of a land-based recirculating aquaculture system in saline-alkali water based on multi-stage purification linkage, characterized in that: The system is used to implement the intelligent control method of any one of claims 1 to 8, and includes a data acquisition module, a purification analysis module, a regulation and early warning module, a purification strategy module, a processing decision module, and a system control module; The data acquisition module is used to collect raw water quality data through the intelligent water quality monitoring stations set at the inlet and outlet of each purification unit. The raw water quality data is preprocessed to obtain preprocessed water quality data. The principal component analysis method is used to extract the principal components whose cumulative contribution rate reaches a preset threshold from the preprocessed water quality data, and feature dimensionality reduction is performed to obtain the fused water quality feature vector. The purification analysis module is used to calculate the removal efficiency of each purification unit for each pollutant based on the concentration of pollutants in the influent and effluent of each purification unit. Based on the current water quality status, it calculates the dynamic weight of each pollutant index using an adaptive adjustment method based on deviation. The removal efficiency and dynamic weight are weighted and summed to obtain the comprehensive purification efficiency of each purification unit. Based on the comprehensive purification efficiency, the rate of change of purification efficiency, and the continuous operating time, the health index of the purification unit is calculated using the health assessment formula. The regulation and early warning module takes the fused water quality feature vector sequence of the past n time steps as input, performs temporal feature learning through a long short-term memory network model, and outputs the predicted value sequence of key water quality parameters for the next m time steps. When the prediction results show that the key water quality parameters will exceed the early warning threshold, the preventive regulation mode is triggered. The purification strategy module is used to obtain the salinity of the water body based on the current conductivity, obtain the current pH value based on the pH sensor, calculate the bioactivity coefficient through the bioactivity inhibition model of salinity-pH coupling effect, determine the inhibition interval based on the comparison between the bioactivity coefficient and the preset inhibition interval threshold, and determine the corresponding graded adaptive adjustment strategy based on the judgment result. The processing decision module uses the health index of the purification unit, the predicted value of water quality parameters, and the level of regulation strategy as the initial conditions for the model predictive control optimization problem. It establishes an optimization objective function with the goal of maximizing purification efficiency and minimizing energy consumption, and sets a set of constraints including hydraulic balance constraints, upper and lower flow limits constraints, water quality compliance constraints, and control variable change rate constraints. The optimal control variable vector is obtained by solving the problem through a sequential quadratic programming algorithm. The system control module is used to convert the optimal control variable vector into the opening setpoint of each electric valve and the frequency setpoint of each water pump. The control signal is output to the actuator through the PLC controller. The control signal adopts a ramp-type gradual output mode to avoid hydraulic impact.

Images