A comprehensive utilization and treatment system for ginseng saponin wastewater

Abstract

The present application relates to biochemical wastewater treatment and automation control technical field, specifically to a kind of comprehensive utilization and processing system of turmeric saponin wastewater;It contains data acquisition, prediction, quantification, risk hedging and control execution module;System passes through the acquisition of water fingerprint vector and reactor state, uses nonlinear model to estimate key enzyme activity;Its core is to determine system safety distance according to enzyme activity and bacterial flora health degree, and dynamically adjust the control weight of three goals of collapse risk, effluent compliance and resource recovery;When safety distance changes, the system executes compound control through multi-objective optimization algorithm;The present application realizes the forward-looking early warning to core biochemical state, and overcomes the defect of response lag of traditional method.

Description

Technical Field

[0001] This invention relates to the field of biochemical wastewater treatment and automated control technology, specifically a comprehensive utilization and treatment system for turmeric saponin wastewater. Background Technology

[0002] In the biochemical treatment system of turmeric saponin wastewater, the influent water quality is often affected by unknown inhibitors and other factors. Traditional solutions struggle to measure key internal states such as enzyme activity in real time, resulting in a lack of effective proactive early warning mechanisms. Furthermore, existing control strategies often prioritize efficiency and lack dynamic risk management capabilities, making it difficult to adjust in time to handle shock loads and potentially leading to irreversible biochemical collapse. Therefore, accurately predicting key biochemical indicators, quantifying system safety margins, and dynamically switching between efficiency-optimal and survival-priority control strategies have become critical technical challenges for improving system robustness. Summary of the Invention

[0003] To solve the above-mentioned technical problems, the present invention provides a comprehensive utilization and treatment system for turmeric saponin wastewater. Specifically, the technical solution of the present invention includes:

[0004] The data acquisition module is used to collect multidimensional feature fingerprint vectors of the incoming water and state vectors inside the reactor in real time.

[0005] The first prediction module is used to estimate the comprehensive activity index of key enzymes in real time in response to multidimensional feature fingerprint vectors and state vectors using a pre-trained nonlinear prediction model.

[0006] The second quantification module is used to calculate the overall health index of the microbial community and, based on the comprehensive activity index of key enzymes and the overall health index of the microbial community, to calculate the system safety distance.

[0007] Among them, the comprehensive health index of the microbial community is calculated by weighting the preset baseline respiration rate and preset baseline diversity based on the real-time microbial respiration rate and real-time microbial community diversity index.

[0008] Among them, the system safety distance is the product of the comprehensive activity index of the key enzyme and the comprehensive health index of the microbial community, and the difference between it and the preset collapse threshold.

[0009] The risk hedging module is used to dynamically adjust the control weights of the three core objectives of crash risk, effluent compliance, and resource recovery in response to the system safety distance.

[0010] The control execution module is used to determine and execute the composite control action vector based on dynamically adjusted control weights and a multi-objective optimization algorithm.

[0011] Preferably, the risk hedging module includes:

[0012] The performance prediction unit is used to predict the effluent pollutant concentration and resource recovery rate under specific control actions by employing a pre-trained effluent pollutant concentration prediction model and a pre-trained resource recovery rate prediction model.

[0013] The dynamic weight adjustment unit is used to dynamically adjust the crash risk weight, water discharge compliance weight, and resource recycling weight in response to the system safety distance.

[0014] The optimization solution unit is used to determine the composite control action vector by minimizing the multi-objective cost function based on the effluent pollutant concentration, resource recovery rate, and control weights.

[0015] Preferably, the dynamic weight adjustment unit is used for:

[0016] When the system safety distance exceeds a preset warning threshold, a preset first weight is applied;

[0017] When the system safety distance is less than or equal to the warning threshold and greater than the preset sacrifice threshold, the crash risk weight is increased;

[0018] When the system safety distance is less than or equal to the sacrifice threshold, the collapse risk weight is made overwhelmingly higher than the effluent compliance weight and resource recovery weight, so as to force the optimization solution unit to choose to sacrifice effluent compliance and resource recovery.

[0019] Preferred options also include:

[0020] The observation back-calculation module is used to continuously monitor the comprehensive activity index of key enzymes after the control execution module executes the composite control action vector, and to deduce the observed enzyme activity based on the sensor-measured concentrations of characteristic pollutants in the influent and the effluent.

[0021] Among them, the enzyme activity was observed based on a simplified biochemical reactor material balance model, calculated from the influent characteristic pollutant concentration, the effluent characteristic pollutant concentration, the calibrated reaction rate constant, and the hydraulic residence time.

[0022] The parameter correction module is used to calculate the prediction deviation between the observed enzyme activity and the comprehensive activity index of the key enzyme, and in response to the prediction deviation being continuously non-zero, to correct the weight parameter set of the pre-trained nonlinear prediction model in the first prediction module online.

[0023] Preferably, the parameter correction module uses an online gradient descent algorithm to update the weight parameter set.

[0024] Preferably, the state vectors acquired in the data acquisition module include dissolved oxygen, pH, and redox potential.

[0025] Preferably, the comprehensive activity index of the key enzyme is used as an early warning signal for the system.

[0026] Preferably, the system collapse threshold is determined by gradually reducing the comprehensive activity index of key enzymes and the comprehensive health index of microbial community in the test bench, and finding the product corresponding to the critical point that causes the effluent water quality to begin to continuously and irreversibly exceed the standard.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. This system uses a nonlinear prediction model to map easily measurable multidimensional fingerprint features and reactor states to key enzyme activities that are difficult to measure online, thus achieving a forward-looking early warning of the core biochemical state of the system and overcoming the shortcomings of traditional methods, such as lack of early warning mechanism and delayed response.

[0029] 2. This system establishes a dynamic hedging relationship between the system safety distance and the control strategy; when the system safety distance is lower than the preset sacrifice threshold, the system can automatically and overwhelmingly allocate the control weight to the collapse risk target, forcing the control strategy to switch from the conventional efficiency-optimal to survival-first, effectively avoiding irreversible collapse of the system under extreme shocks;

[0030] 3. This system comprehensively evaluates and predicts the activity of key enzymes and the overall health index of the microbial community, and compares them with the calibrated system collapse threshold. It quantifies the abstract system collapse risk into a single, operable system safety distance indicator, providing a clear and unified quantitative basis for subsequent dynamic risk decisions.

[0031] 4. This system has online adaptive correction capability; by introducing an observation back-calculation module and a parameter correction module, the system can use the measured effluent and influent concentrations to back-calculate the observed enzyme activity and compare it with the model prediction value. When a prediction deviation occurs, the parameters of the prediction model can be corrected online, ensuring the long-term accuracy and robustness of the prediction model under dynamic changing environments. Attached Figure Description

[0032] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0033] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0035] Example 1:

[0036] Please see Figure 1A comprehensive utilization and treatment system for turmeric saponin wastewater includes:

[0037] The data acquisition module is used to collect multidimensional feature fingerprint vectors of the incoming water and state vectors inside the reactor in real time.

[0038] The first prediction module is used to estimate the comprehensive activity index of key enzymes in real time in response to multidimensional feature fingerprint vectors and state vectors using a pre-trained nonlinear prediction model.

[0039] The second quantification module is used to calculate the overall health index of the microbial community and, based on the comprehensive activity index of key enzymes and the overall health index of the microbial community, to calculate the system safety distance.

[0040] Among them, the comprehensive health index of the microbial community is calculated by weighting the preset baseline respiration rate and preset baseline diversity based on the real-time microbial respiration rate and real-time microbial community diversity index.

[0041] Among them, the system safety distance is the product of the comprehensive activity index of the key enzyme and the comprehensive health index of the microbial community, and the difference between it and the preset collapse threshold.

[0042] The risk hedging module is used to dynamically adjust the control weights of the three core objectives of crash risk, effluent compliance, and resource recovery in response to the system safety distance.

[0043] The control execution module is used to determine and execute the composite control action vector based on dynamically adjusted control weights and a multi-objective optimization algorithm.

[0044] The system in this embodiment includes a data acquisition module, a first prediction module, a second quantification module, a risk hedging module, and a control execution module;

[0045] The data acquisition module aims to provide real-time, high-dimensional input data for subsequent nonlinear prediction and risk assessment. In this embodiment, the data acquisition module includes, but is not limited to, a spectral sensor array for real-time acquisition of the multidimensional feature fingerprint vector of the influent curcumin saponin wastewater. Multidimensional feature fingerprint vector This refers to a high-dimensional data vector, sourced from sources such as UV-Vis spectroscopy or fluorescence spectroscopy sensor arrays, which reflects comprehensive fingerprint information of organic components, turbidity, and potential unknown inhibitors in the influent; a condition sensor is used to collect the state vector inside the reactor. State vector This refers to a vector of macroscopic parameters characterizing the current biochemical reaction environment. Its source is sensors deployed in the reactor, such as dissolved oxygen (DO), pH, and oxidation-reduction potential (ORP) sensors.

[0046] The first prediction module aims to address the difficulty of directly measuring the activity of key enzymes required for the degradation of steroid rings online, using a data-driven model for indirect estimation. This first prediction module responds to the multidimensional feature fingerprint vector acquired by the data acquisition module. With state vector A pre-trained nonlinear prediction model is used. Real-time estimation of the comprehensive activity indicators of key enzymes ;

[0047] To infer this crucial hidden state from measurable external features, this embodiment defines the following core prediction model:

[0048] ;

[0049] in, The comprehensive activity index of the key enzymes predicted is a dimensionless normalized scalar calculated by this module. Its physical meaning is to characterize the comprehensive catalytic ability of microorganisms to degrade core pollutants such as steroidal rings under the current operating conditions. This index will be transmitted to the second quantification module as an early warning signal of the system.

[0050] Pre-trained nonlinear prediction models, such as deep neural networks, support vector regression, or gradient boosting tree models, with their model structure and initial set of weight parameters. Determined through offline training; in this embodiment, the nonlinear prediction model Specifically, it can be a feedforward neural network with two hidden layers. The first hidden layer contains 16 nodes, and the second hidden layer contains 8 nodes, both of which use the ReLU activation function.

[0051] The real-time acquisition of multidimensional feature vectors of the incoming water is provided by the spectral sensor array of the data acquisition module;

[0052] The internal state vector of the reactor is provided by sensors such as dissolved oxygen, pH, and ORP in the data acquisition module.

[0053] The initial weight parameter set of the model is a high-dimensional parameter vector obtained through offline training; specifically, the calibration process involves collecting a set of historical calibration datasets containing different operating conditions. ,in and It is a historical working condition The input vector, This is the actual steroid degradation rate measured by high-precision offline analysis experiments under this operating condition; parameters are optimized using methods such as least squares or gradient descent. To minimize the output of the prediction model Compared with the actual degradation rate The prediction error function between;

[0054] Through this model The system can input multi-dimensional, heterogeneous sensors. and Mapped to a single, standardized comprehensive activity index of key enzymes This indirect prediction method overcomes the shortcomings of traditional methods in obtaining core biochemical indicators in real time, and provides core and forward-looking input parameters for subsequent system resilience quantification.

[0055] The second quantification module aims to quantify the abstract risk of system collapse into a single, actionable scalar, providing a unique and unambiguous basis for risk hedging decisions. This second quantification module performs two core calculations:

[0056] Calculate the overall health index of the gut microbiota : Overall health index of gut microbiota This refers to a comprehensive indicator used to characterize the long-term survival ability and robustness of special degrading microbial communities; in this embodiment, it is based on real-time microbial respiration rate. With real-time microbial diversity indicators , for the preset baseline respiratory rate Diversity with preset benchmarks It is obtained by weighted average calculation; the calculation formula is defined as follows:

[0057] ;

[0058] in,

[0059] : Overall health index of gut microbiota, a dimensionless scalar;

[0060] Real-time microbial respiration rate, in physical terms, reflects the immediate metabolic activity of the microbial community, and is obtained, for example, by measuring the online oxygen consumption rate (OUR) using an online oxygen consumption rate monitor.

[0061] The preset baseline respiration rate, in physical terms, is the standard respiration rate of the bacterial community under optimal operating conditions. It is derived from a set of data collected by continuously running the system under preset optimal operating conditions for one calibration cycle. respiratory rate measurement It is obtained by statistical averaging, for example ;

[0062] Real-time microbial diversity indicators, in physical terms, reflect the long-term potential of the microbial community to resist shock loads. They are obtained, for example, through online flow cytometry or periodic metagenomic analysis.

[0063] Preset benchmark diversity, calibration method is the same That is, a set of conditions based on optimal operating conditions Each diversity index measurement value The average value is determined;

[0064] Weighting coefficients, dimensionless, satisfying Its origin is based on operational experience; for example, in this invention, priority can be given to ensuring the long-term robustness of the system, thus endowing it with... Higher diversity is weighted, such as 0.6-0.7;

[0065] This formula uses instantaneous performance breathing rate With long-term potential diversity This combination provides a comprehensive assessment of the microbial community status; this embodiment uses a linear weighted average as a simplified and effective assessment method; in other embodiments, nonlinear functions, such as product forms, can also be used. Or other more complex functions To capture the more complex synergistic or antagonistic relationships between immediate metabolic activity and long-term microbial potential;

[0066] Calculate the safe distance of the system System safety distance This refers to the core margin indicator that quantifies the distance between the current system state and the boundary of permanent collapse; in this embodiment, it is the comprehensive activity indicator of key enzymes. From the first prediction module and the comprehensive health index of the microbiome The product calculated in the previous step, and the preset system crash threshold. The difference between them;

[0067] The product Defined by this invention as the comprehensive degradation capability of the system, that is, the combination of short-term enzyme activity and long-term microbial health;

[0068] Its calculation formula is defined as follows:

[0069] ;

[0070] in,

[0071] System safety distance, a dimensionless scalar;

[0072] The predicted key enzyme activities are calculated by the first prediction module;

[0073] The overall health index of the gut microbiota is calculated from the previous steps in this module.

[0074] : Preset system crash threshold, a dimensionless scalar; this threshold represents the minimum overall degradation capacity required for the system to maintain stable operation;

[0075] The risk hedging module aims to achieve the core innovation of this invention, namely, a dynamic and intelligent shift from traditional efficiency-optimized control strategies to collapse probability-minimizing control strategies; this risk hedging module is used to respond to system safety distances. The control weights of the three core objectives—collapse risk, effluent compliance, and resource recovery—are dynamically adjusted based on calculations obtained from the second quantitative module. The core logic is: when… When system security is high, modules are assigned higher weights to water compliance and resource recycling;

[0076] when When the risk decreases, the module will automatically and non-linearly increase the weight of the crash risk target, forcing the subsequent control execution module to prioritize system survival rather than short-term economic benefits.

[0077] The control execution module aims to calculate and execute the optimal physical control actions based on the strategy priorities dynamically adjusted by the risk hedging module. This module determines the composite control action vector using a multi-objective optimization algorithm, based on the dynamically adjusted control weights from the risk hedging module. And execute the vector; composite control action vector This refers to a vector containing one or more physical control parameters, derived from the solution of a multi-objective optimization algorithm, and used as an actuator applied to the reactor, for example: [Influent flow rate, aeration rate, nutrient dosage, sludge return ratio], etc.; This module finds an optimal balance control strategy determined by weights under the current risk level by solving a multi-objective cost function. ;

[0078] This embodiment solves the technical problem of traditional curcumin saponin wastewater treatment systems being prone to irreversible collapse when faced with unknown influent disturbances such as unreported inhibitors, due to a lack of forward-looking early warning and dynamic risk management, by constructing a complete technical closed loop from a state prediction first prediction module to a risk quantification second quantification module, then to a dynamic decision-making risk hedging module and a closed-loop execution control module. This system can estimate key internal states in real time. Quantifying system resilience and safety margin and based on The dynamic shift from efficiency-optimal to survival-first approach significantly improves the robustness and long-term operational stability of the biochemical treatment system.

[0079] Example 2:

[0080] The risk hedging module includes:

[0081] The performance prediction unit is used to predict the effluent pollutant concentration and resource recovery rate under specific control actions by employing a pre-trained effluent pollutant concentration prediction model and a pre-trained resource recovery rate prediction model.

[0082] The dynamic weight adjustment unit is used to dynamically adjust the crash risk weight, water discharge compliance weight, and resource recycling weight in response to the system safety distance.

[0083] The optimization solution unit is used to determine the composite control action vector by minimizing the multi-objective cost function based on the effluent pollutant concentration, resource recovery rate, and control weights.

[0084] Based on Example 1, this embodiment specifies the internal structure and working principle of the risk hedging module, transforming it from a conceptual module into an engineering-implementable computational framework; the module includes a performance prediction unit, a dynamic weight adjustment unit, and an optimization solution unit;

[0085] The performance prediction unit aims to provide what-if analysis capabilities for multi-objective optimization, that is, to predict different candidate control actions under the current state. This will result in short-term performance outcomes; the performance prediction unit is used to employ a pre-trained effluent pollutant concentration prediction model. Compared with pre-trained resource recovery rate prediction models Predicting specific control actions The concentration of pollutants in the effluent With resource recycling rate ;

[0086] These two prediction models and The structure and model of the first prediction module Similarly, they will control the current action. Water inlet status Reactor status and predicted enzyme activity As a common input, its prediction formula is as follows:

[0087]

[0088] ;

[0089] in,

[0090] The predicted concentration of contaminants in the effluent, such as COD or specific steroids, is used to assess the cost of effluent compliance. ;

[0091] The predicted resource recovery rate, such as the percentage of a specific intermediate product recovered, is used to assess the cost of resource recovery. ;

[0092] The control action vector to be evaluated is iteratively input by the optimization unit during the optimization process;

[0093] The current influent characteristics, reactor status, and predicted enzyme activity are provided by the preceding module.

[0094] : A pre-trained prediction model whose parameters are determined through offline training; and the model similar, and In this embodiment, it can also be a feedforward neural network, for example, each containing a hidden layer with 10 nodes and a ReLU activation function;

[0095] Through these two models, the optimization unit can evaluate any control before actual execution. Impact on both compliance and recycling objectives;

[0096] The purpose of the dynamic weight adjustment unit is to execute the core logic of the risk hedging strategy, namely, based on the safety distance. Adjusting decision priorities; this dynamic weighting unit is used in response to system safety distance. The collapse risk weight is dynamically adjusted based on the calculations of the second quantification module. Water discharge compliance weight and resource recycling weight ; The value responds to Changes, such as by setting different thresholds, when When decreasing, The value of is increased, thereby overwhelmingly increasing the priority of avoiding crashes in the optimization solution;

[0097] The optimization solution unit aims to calculate the optimal control action that minimizes the current weighted total cost; this optimization solution unit is used based on the effluent pollutant concentration predicted by the performance prediction unit. Right now Resource recycling rate and the control weights determined by the dynamic weight adjustment unit By minimizing the multi-objective cost function, the composite control action vector is determined. In this embodiment, the optimization unit specifically employs a particle swarm optimization algorithm to find the minimum value of the multi-objective cost function, thereby determining... ;

[0098] The specific form of the multi-objective cost function in this embodiment is as follows:

[0099] ;

[0100] in,

[0101] The final determined composite control action vector, i.e., the solution to the optimization problem;

[0102] Provided by the dynamic weight adjustment unit, and following Changing dynamic weights;

[0103] : Crash risk function, for A function used to quantify the cost of collapse risk, its specific form can be defined as follows: ;

[0104] The predicted effluent pollutant concentration cost function is based on... The normalized cost of the calculation, for example ,in, It is the standard for compliance of effluent;

[0105] The predicted resource recovery rate cost function, whose value is based on The normalized cost of the transformation, for example ,in, The target recovery rate is to transform it into a minimization problem;

[0106] To ensure dimensional consistency, , , All were normalized into dimensionless cost items. All weights are dimensionless, therefore all terms summed within the objective function are dimensionless; the optimization unit automatically balances the three conflicting objectives by solving this formula, and the output... Under the current risk level, The optimal solution is reflected;

[0107] This embodiment quantifies and makes feasible the risk hedging concept abstracted in Embodiment 1 by introducing a performance prediction unit and a multi-objective optimization framework. It not only defines how to dynamically adjust the weights of the dynamic weight adjustment unit, but also provides a specific mathematical path for calculating the optimal control action optimization solution unit, enabling the system to truly achieve intelligent decision-making that sacrifices short-term indicators for long-term survival, significantly enhancing the operability and accuracy of risk hedging.

[0108] Example 3:

[0109] The dynamic weight adjustment unit is used for:

[0110] When the system safety distance exceeds a preset warning threshold, a preset first weight is applied;

[0111] When the system safety distance is less than or equal to the warning threshold and greater than the preset sacrifice threshold, the crash risk weight is increased;

[0112] When the system safety distance is less than or equal to the sacrifice threshold, the collapse risk weight is made overwhelmingly higher than the effluent compliance weight and resource recovery weight, so as to force the optimization solution unit to choose to sacrifice effluent compliance and resource recovery.

[0113] Based on Example 2, this embodiment further defines the specific segmented adjustment logic of the dynamic weight adjustment unit. This logic is implemented by setting two key thresholds to achieve a non-linear, step-like strategy transition.

[0114] Warning threshold This refers to a relatively high safe distance threshold, when When the value falls below this level, the system enters a state of alert.

[0115] Sacrifice threshold This refers to a lower safe distance threshold that is close to the collapse boundary. ,when When the value falls below this level, the system must take sacrificial measures.

[0116] These two thresholds are key engineering settings, derived from the analysis of historical operational data. Value and corresponding effluent water quality The statistical relationships, combined with operational experience, are used to classify the system into three operating states: safe zone, warning zone, and danger zone.

[0117] The specific logic of this dynamic weight adjustment unit is as follows:

[0118] Response to system safety distance Greater than the preset warning threshold When the system is in the safe zone, the preset first weight is used;

[0119] In this first weighting, the water discharge compliance weighting and resource recycling weight Assigned a higher value, while the collapse risk weight Lower, for example This allows the optimization unit to prioritize efficiency and economic optimization, i.e., minimize... and It meets the operating conditions required for safe operation;

[0120] Response to system safety distance Less than or equal to the warning threshold And greater than the preset sacrifice threshold The system immediately enters the warning zone, increasing the risk of crash. ;

[0121] At this point, the system begins to hedge against the risk; Increasing this factor can cause the optimization objective to deviate from pure efficiency optimization, and instead partially consider system stability, for example... This is a gradual risk aversion strategy;

[0122] Response to system safety distance Less than or equal to the sacrifice threshold When the system enters the danger zone, the crash risk weight is adjusted. Overwhelmingly higher than the weighting for effluent compliance and resource recycling weight ;

[0123] This is the core sacrifice logic of this invention; The overwhelming proportion, for example This will force the optimization unit to choose to sacrifice effluent compliance and resource recovery, meaning it will no longer care about these aspects. and The value, and only select those that minimize the risk of collapse. Minimize is maximize Control actions For example, by urgently stopping the water intake or increasing aeration, the system can be ensured to survive in extreme conditions.

[0124] This embodiment defines a three-stage nonlinear weight adjustment strategy, providing a clear and executable switching logic for dynamic risk hedging. This threshold-based division of the safety-warning-danger zones makes the shift in system control strategy from efficiency to survival no longer ambiguous, but automatic, rapid and decisive, greatly improving the system's emergency response capability and survival probability when facing serious threats.

[0125] Example 4:

[0126] It also includes: an observation back-calculation module, which is used to continuously monitor the comprehensive activity index of key enzymes after the control execution module executes the composite control action vector, and back-calculate the observed enzyme activity based on the sensor-measured concentrations of characteristic pollutants in the influent and the measured concentrations of characteristic pollutants in the effluent;

[0127] Among them, the enzyme activity was observed based on a simplified biochemical reactor material balance model, calculated from the influent characteristic pollutant concentration, the effluent characteristic pollutant concentration, the calibrated reaction rate constant, and the hydraulic residence time.

[0128] The parameter correction module is used to calculate the prediction deviation between the observed enzyme activity and the comprehensive activity index of the key enzyme, and in response to the prediction deviation being continuously non-zero, to correct the weight parameter set of the pre-trained nonlinear prediction model in the first prediction module online.

[0129] This embodiment, based on Embodiment 1, adds a crucial adaptive correction closed loop to address the model issues in the first prediction module. and in Example 2 The prediction mismatch may be caused by environmental changes, such as the emergence of novel inhibitors not seen in the training set; the closed loop includes an observation back-calculation module and a parameter correction module.

[0130] The observation back-calculation module aims to obtain an observation value that can be used to compare with model predictions. The actual observations for comparison; due to Enzyme activity itself cannot be directly measured. This module uses an inverse model to deduce this value from measurable macroscopic indicators.

[0131] This observation back-calculation module is used to execute composite control action vectors in the control execution module. Subsequently, the predicted values ​​of the comprehensive activity indicators of key enzymes were continuously monitored. Right now Based on the sensor-measured concentration of characteristic pollutants in the influent... and the measured concentration of characteristic pollutants in the effluent By inversely inferring the observed enzyme activity ;

[0132] Observe enzyme activity This refers to the enzyme activity estimate calculated based on a simplified biochemical reactor material balance model; the inverse formula is derived based on a material balance model assuming a completely mixed reactor (CSTR) and first-order kinetics; under steady-state conditions, the material balance is... ,in, It is the reaction rate; assuming the reaction rate is related to the concentration inside the reactor. and enzyme activity Proportional, that is ;

[0133] By combining the equations, we can obtain: ;

[0134] After sorting, we get The calculation formula is as follows:

[0135] ;

[0136] in,

[0137] : The enzyme activity observed by reverse calculation is dimensionless;

[0138] The concentration of characteristic pollutants in the influent measured by the sensor, such as COD in the influent;

[0139] The concentration of characteristic pollutants in the effluent measured by the sensor, such as COD in the effluent;

[0140] k: The calibrated apparent reaction rate constant, whose dimension is the reciprocal of time, for example... The source is obtained by collecting multiple groups of enzymes with known activity under controlled experimental conditions. Different lengths of stay and influent concentration Effluent concentration under the conditions Data, through analysis Determined by regression fitting or statistical averaging;

[0141] : Reactor hydraulic retention time;

[0142] This module provides a crucial feedback signal. It represents the system's actual degradation capacity;

[0143] The parameter correction module aims to utilize... and The deviation between them is used to update the model parameters of the first prediction module online. This allows it to adapt to new operating conditions; the parameter correction module is used to calculate and observe enzyme activity. The comprehensive activity index of key enzymes, i.e., the predicted value. Prediction deviation between ; and in response to prediction bias When the value remains non-zero, for example, when the moving average of the deviation exceeds a certain threshold, it indicates that the model... The prediction model has become inaccurate; online correction is needed for the pre-trained nonlinear prediction model in the first prediction module. Fixed weight parameter set ;

[0144] This embodiment introduces online learning and adaptive capabilities into the system by adding an observation back-calculation module and a parameter correction module; it solves the problem of data-driven models such as The fundamental problem is the performance degradation when faced with operating conditions not seen in the training set, such as novel inhibitors; when the model's predictions are inaccurate. This closed loop can automatically correct the model parameters. This ensures the accuracy of the predictions, thereby guaranteeing the reliability of the entire risk assessment. and control decision dependence Its long-term effectiveness and robustness.

[0145] Example 5:

[0146] The parameter correction module uses an online gradient descent algorithm to update the weight parameter set.

[0147] Based on Example 4, this embodiment specifically defines the online correction algorithm used by the parameter correction module;

[0148] The parameter correction module uses an online gradient descent algorithm to update the weight parameter set. ;

[0149] Online gradient descent algorithms aim to minimize prediction bias. The resulting loss function For example, squared loss By calculating the loss For parameters gradient And update the parameters in the opposite direction of the gradient to achieve online model correction;

[0150] According to the chain rule, ;

[0151] The standard gradient descent update rule is ;

[0152] After substituting the values, the parameter update formula for this embodiment is as follows:

[0153] ;

[0154] in,

[0155] The revised parameter set vector of the new model will be used for the next prediction;

[0156] The vector of model parameters at the current moment;

[0157] The learning rate is a key adjustable hyperparameter, a dimensionless scalar used to control the step size of the correction. The choice of value depends on the frequency of changes in the influent water quality; the faster the changes, the better. The larger the value, the better;

[0158] Prediction bias, i.e. A dimensionless scalar, calculated by the observation back-calculation module;

[0159] Predictive Model The output of the parameters The gradient is a vector whose dimension is... The specific calculation method depends on the model. The type, for example, for neural networks, can be calculated using the backpropagation algorithm;

[0160] When prediction deviation When it appears, the formula will drive the parameters. From the gradient in the direction that reduces the deviation Instructions to make fine adjustments Control the amplitude to make the model Predicted output Approaching the true observed values ​​again ;

[0161] This embodiment explicitly employs an online gradient descent algorithm, providing a concrete, efficient, and computationally feasible implementation path for parameter correction in Embodiment 4. This algorithm eliminates the need for costly and time-consuming offline retraining of the entire model. Instead, it enables lightweight parameter fine-tuning at each time step, allowing the system to quickly adapt to dynamically changing wastewater composition and significantly improving the real-time adaptability of the prediction model.

[0162] Example 6:

[0163] The state vectors collected in the data acquisition module include dissolved oxygen, pH, and redox potential.

[0164] This embodiment, based on embodiment 1, focuses on the state vector collected by the data acquisition module. The specific content is limited;

[0165] The state vector collected in the data acquisition module This includes: dissolved oxygen (DO), pH, and oxidation-reduction potential (ORP).

[0166] This embodiment ensures the accuracy of the state vector input to the first prediction module by explicitly specifying the acquisition of three core parameters: DO, pH, and ORP. It contains the most characterizing information about the biochemical state; these three parameters are the most critical indicators characterizing the macroscopic environment inside the bioreactor, and they directly affect microbial activity and enzymatic reaction rates, thus incorporating them into the model. The input can significantly improve the overall activity indicators of key enzymes. The accuracy and reliability of the predictions.

[0167] Example 7:

[0168] The comprehensive activity indicators of key enzymes are used as early warning signals for the system.

[0169] This embodiment, based on Embodiment 1, further clarifies the comprehensive activity index of the key enzyme calculated by the first prediction module. One of its important uses;

[0170] Comprehensive activity indicators of key enzymes Used as an early warning signal for the system;

[0171] Specifically, due to It is a direct estimate of the key enzyme activities required to degrade core pollutants such as steroidal rings. It provides a more forward-looking reflection of fundamental changes in the system's treatment capacity than traditional macroscopic indicators such as COD or BOD; when inhibitors are introduced... During rapid decline, the system can determine the quality of the effluent. Issuing a warning before the limit is exceeded and triggering a response from the risk hedging module, for example, making... Falling below the warning threshold ;

[0172] This embodiment clarifies Its early warning function makes it not only an intermediate parameter for risk quantification, but also an independent forward-looking indicator. This enables the system to shift from a traditional, post-event response based on effluent water quality to a pre-event early warning based on internal biochemical status, giving operators valuable time to take intervention measures or for the system to automatically enter risk hedging mode. It is the first line of defense against system collapse.

[0173] Example 8:

[0174] The system crash threshold is determined by gradually reducing the comprehensive activity index of key enzymes and the comprehensive health index of the microbial community in the test bench, and finding the product corresponding to the critical point that causes the effluent water quality to begin to continuously and irreversibly exceed the standard.

[0175] Based on embodiments 1 and 4, this embodiment specifically defines the key parameter in the second quantization module—the system crash threshold. —The method for determining this parameter is to ensure that its setting is scientifically based and feasible;

[0176] System crash threshold It is calibrated through controlled experiments conducted on a test bench; specifically, the calibration steps are as follows:

[0177] By using controlled methods, such as gradually adding known inhibitors in Steady-state experimental points The overall degradation capacity of the system was measured separately. and the corresponding effluent water quality Find out The critical experimental point at which continuous irreversible exceedances begin ,but identified as ;

[0178] Therefore, it represents the minimum overall degradation capacity required for the system to maintain stable operation;

[0179] This embodiment is This core parameter provides a specific, scientific, and reproducible calibration method; this solves the problem of arbitrary or unfounded threshold settings, ensuring the system's safe distance. The calculations have clear physical meaning and engineering guiding value; an accurate This is a prerequisite for the accurate triggering of subsequent risk hedging implementation 2 and sacrifice logic implementation 3, which greatly improves the accuracy and credibility of the entire risk assessment framework.

[0180] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A comprehensive utilization and treatment system for turmeric saponin wastewater, characterized in that, include: The data acquisition module is used to collect multidimensional feature fingerprint vectors of the incoming water and state vectors inside the reactor in real time. The first prediction module is used to estimate the comprehensive activity index of key enzymes in real time in response to multidimensional feature fingerprint vectors and state vectors using a pre-trained nonlinear prediction model. The second quantification module is used to calculate the overall health index of the microbial community and, based on the comprehensive activity index of key enzymes and the overall health index of the microbial community, to calculate the system safety distance. Among them, the comprehensive health index of the microbial community is calculated by weighting the preset baseline respiration rate and preset baseline diversity based on the real-time microbial respiration rate and real-time microbial community diversity index. Among them, the system safety distance is the product of the comprehensive activity index of the key enzyme and the comprehensive health index of the microbial community, and the difference between it and the preset collapse threshold. The risk hedging module is used to dynamically adjust the control weights of the three core objectives of crash risk, effluent compliance, and resource recovery in response to the system safety distance. The control execution module is used to determine and execute the composite control action vector based on dynamically adjusted control weights and a multi-objective optimization algorithm.

2. The comprehensive utilization and treatment system for turmeric saponin wastewater according to claim 1, characterized in that, The risk hedging module includes: The performance prediction unit is used to predict the effluent pollutant concentration and resource recovery rate under specific control actions by employing a pre-trained effluent pollutant concentration prediction model and a pre-trained resource recovery rate prediction model. The dynamic weight adjustment unit is used to dynamically adjust the crash risk weight, water discharge compliance weight, and resource recycling weight in response to the system safety distance. The optimization solution unit is used to determine the composite control action vector by minimizing the multi-objective cost function based on the effluent pollutant concentration, resource recovery rate, and control weights.

3. The comprehensive utilization and treatment system for turmeric saponin wastewater according to claim 2, characterized in that, The dynamic weight adjustment unit is used for: When the system safety distance exceeds a preset warning threshold, a preset first weight is applied; When the system safety distance is less than or equal to the warning threshold and greater than the preset sacrifice threshold, the crash risk weight is increased; When the system safety distance is less than or equal to the sacrifice threshold, the collapse risk weight is made overwhelmingly higher than the effluent compliance weight and resource recovery weight, so as to force the optimization solution unit to choose to sacrifice effluent compliance and resource recovery.

4. The comprehensive utilization and treatment system for turmeric saponin wastewater according to claim 1, characterized in that, Also includes: The observation back-calculation module is used to continuously monitor the comprehensive activity index of key enzymes after the control execution module executes the composite control action vector, and to deduce the observed enzyme activity based on the sensor-measured concentrations of characteristic pollutants in the influent and the effluent. Among them, the enzyme activity was observed based on a simplified biochemical reactor material balance model, calculated from the influent characteristic pollutant concentration, the effluent characteristic pollutant concentration, the calibrated reaction rate constant, and the hydraulic residence time. The parameter correction module is used to calculate the prediction deviation between the observed enzyme activity and the comprehensive activity index of the key enzyme, and in response to the prediction deviation being continuously non-zero, to correct the weight parameter set of the pre-trained nonlinear prediction model in the first prediction module online.

5. The comprehensive utilization and treatment system for turmeric saponin wastewater according to claim 4, characterized in that, The parameter correction module uses an online gradient descent algorithm to update the weight parameter set.

6. The comprehensive utilization and treatment system for turmeric saponin wastewater according to claim 1, characterized in that, The state vectors acquired by the data acquisition module include dissolved oxygen, pH, and redox potential.

7. The comprehensive utilization and treatment system for turmeric saponin wastewater according to claim 1, characterized in that, The comprehensive activity index of the key enzymes is used as an early warning signal for the system.

8. The comprehensive utilization and treatment system for turmeric saponin wastewater according to claim 1, characterized in that, The system collapse threshold is determined by gradually reducing the comprehensive activity index of key enzymes and the comprehensive health index of the microbial community in the test bench, and finding the product corresponding to the critical point that causes the effluent water quality to begin to continuously and irreversibly exceed the standard.

Images