Sewage early warning and treatment analysis system based on deep learning

Abstract

The invention relates to the technical field of environment monitoring and automatic control, in particular to a sewage early warning and treatment analysis system based on deep learning. Comprising a data acquisition module used for acquiring multi-dimensional real-time data of an activated sludge system of an urban sewage treatment plant and constructing a time sequence characteristic data set; the first processing module is used for outputting a dynamic weight coefficient through deep learning model training and reasoning based on the time sequence characteristic data set; the second processing module is used for calculating a critical index representing the system risk in real time by combining the multi-dimensional real-time data, the dynamic weight coefficient and a preset parameter reference value; the early warning judgment module is used for comparing the critical index with a preset early warning threshold value and judging the risk level of the system; and the intervention decision module is used for responding to the risk level judged by the early warning judgment module and generating and executing a hierarchical intervention strategy. According to the system, accurate prediction of the viscous collapse critical state is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of environmental monitoring and automation control, in particular to a sewage early warning and treatment analysis system based on deep learning. BACKGROUND

[0002] With the acceleration of urbanization, the stable operation of sewage treatment plants is crucial to environmental protection and public health; as the core process of urban sewage treatment, the stability and reliability of activated sludge system directly affect the effluent quality; however, activated sludge system is a complex, nonlinear dynamic biological system, whose operating state is influenced by multiple factors, and is prone to instability due to operating condition fluctuations; Viscous collapse is a key inducement to functional disorders of activated sludge system, which further leads to deterioration of sludge settling performance, such as sludge bulking or floating, and other serious production accidents; due to the complex nonlinear coupling relationship between internal microbial community, biochemical reaction and physical environment, traditional monitoring and control methods face serious challenges; these traditional methods have the following limitations: Traditional methods mainly rely on conventional macroscopic water quality indicators for monitoring; such indicators are usually lagging, and cannot capture signs of system instability at a deeper level and earlier stage from the microscopic molecular level and mesoscopic morphological structure; in addition, they also ignore the influence of hidden physical and chemical stress factors such as aeration sound field and influent synergistic toxicity on system stability; Traditional methods mostly use static thresholds based on experience to judge a single indicator; this method cannot adapt to the dynamic characteristics of the system, cannot reveal the complex nonlinear relationship between multi-dimensional parameters, and cannot dynamically adjust the weights of each risk factor according to real-time operating conditions, resulting in untimely and inaccurate early warning; Intervention measures based on traditional monitoring methods are usually passive and lagging; when macroscopic indicators are monitored, the system may have already reached a state that is difficult to reverse, and the control measures at this time have limited effect, making it difficult to achieve proactive preventive intervention and form a closed-loop control from perception, evaluation to decision-making and execution; In recent years, deep perception technology and deep learning models based on multi-dimensional data have provided new ideas for solving the above problems; by integrating microscopic biological signals, mesoscopic structural morphology, dynamic biochemical data and physical field environment data, a more comprehensive cross-scale understanding of the state of activated sludge system can be constructed; among them, the concentration of quorum sensing signal molecules can be used as an early molecular marker of system dysfunction; the fractal dimension of sludge flocs can directly represent the stability of its physical structure; the metabolic rate ratio of extracellular polymeric substances is a direct inducement to viscous collapse; Convolutional neural networks (CNN) and long short-term memory networks (LSTM) and other deep learning models have shown strong capabilities in processing complex high-dimensional time series data; CNN can effectively capture the local combination features of different parameters within a short time window, while LSTM is good at learning the dependence relationship and evolution law of these features over a long time span; In summary, the existing sewage treatment monitoring technology relies on a single lagging macroscopic indicator and a static threshold, making it difficult to achieve early and accurate warning of the critical state of activated sludge system viscosity collapse; Currently, there is no research that combines multi-scale, multi-dimensional deep perception data with deep learning models that can reveal their inherent nonlinear, time series coupling relationship to build a complete closed-loop control system from accurate prediction to active intervention.

[0003] The above information disclosed in the background section is only used to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0004] To solve the above technical problems, the present application discloses a sewage early warning and treatment analysis system based on deep learning, specifically, the technical scheme of the present application is: A sewage early warning and treatment analysis system based on deep learning, comprising: A data acquisition module for acquiring multi-dimensional real-time data of an activated sludge system of a municipal sewage treatment plant and constructing a time series feature data set; A first processing module for training and reasoning based on the time series feature data set through a deep learning model to output dynamic weight coefficients; A second processing module for real-time calculation of a critical index representing system risk by combining multi-dimensional real-time data, dynamic weight coefficients and a pre-set parameter reference benchmark value; An early warning judgment module for comparing the critical index with a pre-set early warning threshold to determine the risk level of the system; An intervention decision module for generating and executing a graded intervention strategy in response to the risk level determined by the early warning judgment module.

[0005] Preferably, the multi-dimensional real-time data includes microscopic biological signals, mesoscopic structure morphology, time-varying dynamic biochemical data and physical field environment data; The microscopic biological signal is the concentration of quorum sensing signal molecules; The mesoscopic structure morphology is the fractal dimension of sludge flocculation; The time-varying dynamic biochemical data is the dynamic secretion-degradation rate ratio of extracellular polymeric substances; The physical field environment data includes acoustic field anomaly factors and synergistic toxicity gradients.

[0006] Preferably, the first processing module is specifically used for: The convolutional neural network and the long short-term memory network hybrid model are adopted to process the time series data of all parameters collected by the data collection module, so as to learn the nonlinear coupling relationship between the parameters and output dynamic weight coefficients.

[0007] Preferably, the parameter reference benchmark value is set based on historical data statistics of long-term stable operation of the sewage treatment plant, and the early warning threshold value is set according to data performance before the sludge floating accident occurs.

[0008] Preferably, the early warning determination module is specifically used for: comparing the critical index with the first early warning threshold value and the second early warning threshold value; when the critical index is less than or equal to the first early warning threshold value, determining that the system is in a safe state; when the critical index is greater than the first early warning threshold value and less than or equal to the second early warning threshold value, determining that the system is in a first early warning state; when the critical index is greater than the second early warning threshold value, determining that the system is in a second early warning state.

[0009] Preferably, the intervention decision module is further used for: when the system enters the first early warning state or the second early warning state, calculating the aeration disturbance effect value based on the current critical index calculated by the second processing module; the aeration disturbance effect value is a function of the relative adjustment rate of the aeration amount; by solving the maximum value of the aeration disturbance effect value, the optimal aeration adjustment rate is determined.

[0010] Preferably, the calculation of the aeration disturbance effect value is based on an effect model that establishes a mathematical relationship between the aeration adjustment amount and the system risk state index; the effect model includes: a linear term representing the basic action of aeration; a negative cross term representing the negative effect of aeration when the system is unstable; and a quadratic negative term representing the destruction effect of excessive aeration itself.

[0011] Preferably, the hierarchical intervention strategy includes a first intervention; the first intervention is used to execute the optimal aeration adjustment rate when the system enters the first early warning state, and to execute targeted suppression measures according to the main driving factors that cause the risk index to rise, which are identified according to the weight of each parameter output by the first processing module.

[0012] Preferably, the hierarchical intervention strategy includes a second intervention; the second intervention is used to execute the optimal aeration adjustment rate when the system enters the second early warning state, and to start at least one of the following measures: adding a high molecular flocculant; adjusting the sludge return ratio; reducing the influent load.

[0013] Preferably, after the hierarchical intervention strategy is executed, the data acquisition module is returned to continuously monitor the data and recalculate the critical index, and the intervention intensity is dynamically adjusted according to the recalculated critical index, forming a closed-loop negative feedback control.

[0014] Compared with the prior art, the present application has the following beneficial effects: 1、The system can identify the potential instability risk of the activated sludge system earlier and more accurately than traditional methods by fusing real-time data at multiple levels such as micro, meso and macro, and using a deep learning model to reveal the inherent nonlinear coupling relationship, and realizes accurate prediction of the critical state of viscous collapse.

[0015] 2、The system can dynamically output a set of weight coefficients according to real-time data, and calculate a comprehensive critical risk index based on multi-dimensional data, realizing adaptive evaluation of system risk changes with operating conditions, and overcoming the misjudgment and omission problems caused by the dependence of traditional methods on static threshold and single index.

[0016] 3、Based on risk level and traceability analysis, the system can automatically generate and execute a hierarchical intervention strategy including optimal aeration adjustment rate and targeted suppression measures, converting early warning signals into specific and optimal control actions, and realizing active suppression and closed-loop control of system risk.

[0017] 4、The system continuously monitors and dynamically adjusts the intervention intensity by returning to the data acquisition module after intervention, forming a complete closed-loop negative feedback control loop from perception-evaluation-decision-execution-re-perception, thereby significantly improving the stability and robustness of the wastewater treatment system and effectively avoiding production accidents. BRIEF DESCRIPTION OF DRAWINGS

[0018] The present application will be further explained in conjunction with the accompanying drawings and examples: Figure 1 is a flow chart of the system of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in conjunction with specific examples.

[0020] Example 1: Please refer to Figure 1 A wastewater early warning and treatment analysis system based on deep learning, comprising: A data acquisition module for acquiring multi-dimensional real-time data of an activated sludge system of a municipal wastewater treatment plant and constructing a time series feature data set; The first processing module is configured to output dynamic weight coefficients based on the time sequence feature dataset through deep learning model training and inference. The second processing module is configured to calculate a critical index representing system risk in real time by combining multi-dimensional real-time data, dynamic weight coefficients, and a preset parameter reference benchmark value. The early warning judgment module is configured to compare the critical index with a preset early warning threshold to determine the risk level of the system. The intervention decision module is configured to generate and execute a hierarchical intervention strategy in response to the risk level determined by the early warning judgment module. The embodiment provides a sewage early warning and processing analysis system based on deep learning. The technical goal of the system is to accurately predict and actively intervene in the viscous collapse critical state that may cause deterioration of sludge settling performance by performing multi-dimensional and multi-scale deep perception on the activated sludge system, thereby constructing a complete technical closed loop from perception, evaluation, decision-making, and execution. In a specific implementation scenario, the system is deployed in a municipal sewage treatment plant, and the specific implementation manner includes the following modules. The data acquisition module is configured to comprehensively and in real time capture underlying driving factors that can represent the state of the activated sludge system and provide high-quality data input for the subsequent deep learning model. In the embodiment, the module obtains multi-dimensional real-time data of the activated sludge system of the municipal sewage treatment plant by integrating various sensors and online analysis equipment, and integrates and constructs a time sequence feature dataset. The data is not simply a conventional water quality index, but a key feature that can reflect the health status of the system from multiple levels such as micro, meso, and macro, after being selected by a preset rule. To ensure the robustness of the system, the module further includes a data preprocessing unit for validity verification of sensor data, which can identify and process outliers or null values that are out of the physical meaning range due to sensor failure, and ensure the data quality input to the subsequent model through interpolation or rejection, to avoid misjudgment of the entire early warning system due to failure of a single data source. The first processing module is configured to autonomously learn and reveal the potential nonlinear coupling relationship between features from the collected complex and high-dimensional data, and provide dynamic and adaptive weights for the risk assessment model. In the embodiment, the module receives the time sequence feature dataset constructed by the data acquisition module, and performs training and inference through a pre-trained deep learning model, and finally outputs a set of dynamic weight coefficients. The set of weight coefficients directly reflects the contribution of each monitoring parameter to the overall risk of the system under the current working condition, and is the technical basis for realizing adaptive early warning. The second processing module aims to construct a core index capable of unifying different physical dimensions and comprehensively evaluating system risks; the module combines the multi-dimensional real-time data collected by the data collection module , the dynamic weight coefficient output by the first processing module , and the parameter reference benchmark value preset by the system , and calculates a critical index representing system risks in real time through the following risk index model : ; wherein is an S-shaped normalization function, for example, the specific form of a logistic function: , which is used to convert the ratio of each parameter to its benchmark value into a risk contribution degree between 0 and 1; the parameters and can be calibrated according to historical data to adjust the steepness and center position of the function, and the function can nonlinearly amplify the effect of deviation from the normal state; it is noted that the decrease of the fractal dimension represents an increase in risk, so it adopts an inverse proportion form ; the sound field anomaly factor and the synergistic toxicity gradient are themselves dimensionless risk representation values after preprocessing, so they are directly introduced into the model; The index unifies multiple dimensional risk factors into a single, dimensionless risk scale; The early warning judgment module aims to accurately classify the running state of the system according to the quantified risk index, and provides clear triggering basis for subsequent graded intervention; the module compares the critical index calculated by the second processing module in real time with the preset early warning threshold , so as to determine the risk level of the system, such as a safe state, a first-level warning or a second-level warning; The intervention decision module aims to convert the early warning signal into specific and optimal control actions to achieve active suppression and closed-loop control of system risks; the module automatically generates and executes corresponding graded intervention strategies in response to the risk level determined by the early warning judgment module; these strategies not only include optimization and control of core equipment such as aerators, but also include precise intervention measures targeting the root causes of risks; The embodiment builds a complete sewage treatment process early warning and closed-loop control system through the cooperation of the above modules; compared with the monitoring method relying on traditional single index and static threshold, the application can perceive the deep-seated reasons of system instability earlier and more accurately, and take the optimal preventive intervention measures through dynamic evaluation and intelligent decision, thereby significantly improving the stability and robustness of the sewage treatment system, and effectively avoiding serious production accidents caused by sludge swelling or floating.

[0021] Embodiment 2: The multi-dimensional real-time data includes micro-biological signals, mesoscopic structure morphology, time-varying dynamic biochemical data and physical field environment data; The micro-biological signal is the concentration of quorum sensing signal molecules; The mesoscopic structure morphology is the fractal dimension of sludge flocculation body; The time-varying dynamic biochemical data is the dynamic secretion-degradation rate ratio of extracellular polymeric substance; The physical field environment data includes sound field anomaly factor and synergistic toxicity gradient; In further refinement of embodiment 1, the multi-dimensional real-time data of the application in order to more comprehensively depict the complex state of the sludge system, the selection and definition embodies the cross-scale technical concept; specifically, it includes: Micro-biological signal: its purpose is to capture the abnormality of microbial population behavior from the biochemical level; in this embodiment, the micro-biological signal refers to the concentration of quorum sensing signal molecules such as AHLs ; the source of this parameter is obtained by real-time monitoring of high-frequency sampling sensors deployed at the anoxic / oxic (A / O) interface of the aeration tank; the technical consideration is that quorum sensing is the key to microbial communication and collective behavior, and the abnormal accumulation of its signal molecules is often an early molecular level indicator of system dysfunction, much earlier than the change of macroscopic index; Mesoscopic structure morphology: its purpose is to quantify the aggregation state and stability of sludge flocculation body from the physical morphology level; in this embodiment, the mesoscopic structure morphology refers to the fractal dimension of sludge flocculation body ; the source of this parameter is obtained by real-time shooting of sludge images of key areas through an online image analyzer, and calculated by combining a special image processing algorithm; fractal dimension can accurately describe the compactness and complexity of flocculation structure, and the numerical decrease is directly related to the risk of loose and easy-to-break flocculation structure, which is a direct physical morphology index representing sludge settling performance; Time-varying dynamic biochemical data: its purpose is to capture the dynamic balance change of microbial metabolic activity; in this embodiment, the time-varying dynamic biochemical data refers to the dynamic secretion-degradation rate ratio of extracellular polymeric substance (EPS) EPS is the key substance to form the skeleton of sludge floc, and its excessive secretion or insufficient degradation will lead to abnormal sludge viscosity, which is the direct cause of viscosity collapse; the source of this parameter is to obtain the rate ratio by monitoring the relevant biochemical parameters and performing model calculation, thereby dynamically reflecting the metabolic balance state of EPS; Physical field environment data: its purpose is to quantify the disturbance of the external physical and chemical environment to the system; in this embodiment, it further includes two factors: Acoustic field anomaly factor : refers to the quantitative abnormality of the specific frequency band, such as the 150-400Hz acoustic field spectrum data, formed by the aeration system underwater; the acoustic field data is collected by a hydrophone array and signal processing is performed to obtain it; its calculation result itself is a non-dimensional index reflecting the degree of deviation from the normal acoustic spectrum, ranging from [0, 1], without the need to set a reference value separately; the technical innovation point is that the present application finds that the acoustic field generated by aeration not only provides mixing power, but also provides a physical stress, such as the enhancement of specific frequency energy, which will disturb the quorum sensing of microbial populations, and is a key external disturbance source that is ignored by traditional monitoring; Synergistic toxicity gradient : refers to the quantitative index of the concentration change of trace heavy metal ions such as Cu²⁺ and Zn²⁺ in the influent pipeline; it is monitored by an online heavy metal analyzer and processed into a dimensionless gradient; the gradient value directly represents the relative strength of the toxic shock, without the need to set an additional reference value; its role is to quantify the chemical stress that has a synergistic inhibitory effect on microbial activity under the combined action of multiple low-concentration heavy metal ions. This synergistic toxicity is more hidden and more harmful than single-ion toxicity.

[0022] Embodiment 3: The first processing module is specifically configured to: adopt a hybrid model of convolutional neural network and long short-term memory network to process the time series data of all parameters collected by the data acquisition module, to learn the nonlinear coupling relationship between the parameters, and output dynamic weight coefficients; In further optimization of embodiment 1, the first processing module of the embodiment is specifically implemented as follows in order to efficiently process multi-dimensional heterogeneous time series data and accurately capture its complex internal relationship: The module is specifically configured to adopt a hybrid model of convolutional neural network (CNN) and long short-term memory network (LSTM) to process the time series data of all parameters collected by the data acquisition module; The model is specifically configured to adopt a hybrid model of convolutional neural network (CNN) and long short-term memory network (LSTM) to process the time series data of all parameters collected by the data acquisition module; For further illustration, the design of the model embodies a unique innovative consideration: the sewage treatment system is a complex nonlinear time-varying system, and its state evolution is not only related to the instantaneous values of the current indicators, but also depends on the dynamic change trends of these indicators in the past period of time and their mutual influence. Convolutional Neural Network (CNN) layers are used to process the input multi-dimensional time series data, which acts as a feature extractor to effectively capture local patterns and feature correlations formed by different parameter combinations within a short time window; Long Short-Term Memory (LSTM) layers receive the feature sequences extracted by the CNN layers, which learn the dependency relationships and evolution laws of these features over a longer time span. The design of the LSTM network effectively avoids the gradient vanishing or explosion problem, thereby accurately capturing long-term time series dynamics that have a decisive impact on the system state; Through this hybrid model architecture, the system learns the nonlinear coupling relationships between parameters and outputs dynamic weight coefficients For example, the model may learn that: under the background of continuous decrease in fractal dimension , a small fluctuation in the concentration of group sensing signal significantly amplifies the contribution weight of the system collapse.

[0023] Embodiment 4: The parameter reference benchmark value is set based on historical data statistics of long-term stable operation of the sewage treatment plant; the early warning threshold is set according to the data performance before the historical sludge floating accident; In the further implementation of embodiment 1, to ensure the applicability and accuracy of the early warning model, the setting method of the parameter reference benchmark value and the early warning threshold involved has clear technical basis and implementability: The parameter reference benchmark value is set based on historical data statistics of long-term stable operation of the sewage treatment plant; the specific determination method is that the parameter reference benchmark value refers to the statistical average or median of each monitoring parameter when the system is in a recognized, efficient and stable operation state, for example, the effluent quality has been up to standard for one month and the sludge settling ratio SVI is stable below 100 mL / g; its role is to provide a health state scale for risk calculation, and the subsequent real-time parameters will be compared with this benchmark to quantify the degree of deviation; the unique technical consideration of this setting method is that it makes the model benchmark completely personalized, closely fitting the water quality characteristics and process parameters of a specific sewage plant, avoiding the inaccuracy problem caused by using universal standards; The early warning threshold is set according to the data performance before the historical sludge floating accident; the specific determination method is to determine the threshold by retrospectively analyzing the historical trajectory of the critical index calculated by the method of the present application within 12-24 hours before the sewage plant's several times of sludge swelling or floating accidents in the past; for example, the first early warning threshold may be set to cover 95% of the normal operation state​ Value fluctuation quantile, which serves to identify statistically significant abnormal deviation, triggering preventive intervention; second early warning threshold Can be set to 6 hours before the historical multiple accidents Average level reached by the value, which serves to identify that the system has entered a critical state of high risk, and emergency control must be initiated.

[0024] Embodiment 5: The early warning determination module is specifically used for: Comparing the critical index with the first early warning threshold and the second early warning threshold; When the critical index is less than or equal to the first early warning threshold, the system is determined to be in a safe state; When the critical index is greater than the first early warning threshold and less than or equal to the second early warning threshold, the system is determined to be in a first-level early warning state; When the critical index is greater than the second early warning threshold, the system is determined to be in a second-level early warning state; In further refinement of Embodiment 1, the early warning determination module of the system for the purpose of fine management of risk, the specific determination logic is as follows: The module is specifically used for comparing the critical index calculated by the second processing module With the first early warning threshold And the second early warning threshold And outputting a clear risk level according to the comparison result; the thresholds here have been pre-set according to the method of Embodiment 4; The specific logic rules for determination are as follows: When the critical index Is less than or equal to the first early warning threshold , the system is determined to be in a safe state; in this state, all indicators of the system fluctuate within the acceptable baseline range, and the control system only executes the normal steady-state operation strategy; When the critical index Is greater than the first early warning threshold And less than or equal to the second early warning threshold , the system is determined to be in a first-level early warning state; this indicates that the system has shown a significant tendency to lose stability, although it has not yet reached the critical point of collapse, but the key underlying driving factors have begun to deteriorate; the determination at this level will trigger preventive intervention measures aimed at restoring system stability; When the critical index Is greater than the second early warning threshold , the system is determined to be in a second-level early warning state; this indicates that the system is close to or has entered a dangerous state of viscous collapse, and the sludge settling performance may deteriorate sharply in a short time; the determination at this level will trigger emergency and strong control strategies to avoid accidents.

[0025] Embodiment 6: The intervention decision module is also used for: When the system enters the first warning state or the second warning state, based on the current critical index calculated by the second processing module, the aeration disturbance effect value is calculated; The aeration disturbance effect value is a function of the relative adjustment rate of the aeration amount; By solving the maximum value of the aeration disturbance effect value, the optimal aeration adjustment rate is determined; The calculation of the aeration disturbance effect value is based on an effect model that establishes a mathematical relationship between the aeration adjustment amount and the system risk state index; The effect model includes: a linear term representing the basic effect of aeration; a negative cross term representing the negative effect of aeration when the system is unstable; and a quadratic negative term representing the destructive effect of excessive aeration itself; The hierarchical intervention strategy includes a first intervention; the first intervention is used to execute the optimal aeration adjustment rate when the system enters the first warning state, and to identify the main driving factors leading to the rise of the risk index according to the parameter weight output by the first processing module, so as to execute targeted suppression measures; The hierarchical intervention strategy includes a second intervention; the second intervention is used to execute the optimal aeration adjustment rate when the system enters the second warning state, and to start at least one of the following measures: Adding high molecular flocculants; Adjusting the sludge return ratio; Reducing the influent load; On the basis of the hierarchical warning of embodiment 5, the intervention decision module of the present embodiment is further enriched and refined in order to realize intelligent, optimized and multi-strategy intervention control; this embodiment combines the description of the several interrelated embodiments to show their synergistic effect; When the system enters the first warning state or the second warning state, a core innovation of the intervention decision module is that it does not adopt a fixed aeration adjustment strategy, but quantitatively predicts the effect of aeration adjustment; for this purpose, the module will calculate the aeration disturbance effect value based on the current critical index calculated by the second processing module ; In order to realize this calculation, the present embodiment constructs an innovative effect model, which first establishes a mathematical relationship between the aeration adjustment amount and the system risk state index; the internal logic is that the influence of aeration on the system is dual, and its final effect is highly dependent on the current stability of the system, which is characterized; the specific mathematical expression of the effect model is: ; The model is composed of the following items, and the parameters are explained as follows: ​The aeration disturbance effect value is a dimensionless index, whose purpose is to quantify the comprehensive effect of aeration adjustment on improving the system state, and a positive value represents a positive effect, while a negative value represents a negative effect; The dimensionless relative adjustment rate of aeration quantity is a control variable, and its source is the definition formula ; The current viscous collapse critical index calculated by the second processing module is used as an input parameter to reflect the current stability of the system; The system state coefficient is a dimensionless parameter; in order to clarify its calibration process, a historical event data set for calibration is defined; the data set is composed of a plurality of data points, and each data point contains a set of historical measurement values ; wherein, refers to the critical index at a certain time in the past; refers to the known aeration adjustment rate implemented at that time; and is an observed effect value calculated according to the change in the system state within a predetermined time window, for example, within 6 hours after the intervention, which quantitatively represents the actual effect of the adjustment; Its specific calculation method can be defined as the difference between the critical index before the intervention and the critical index after the intervention, i.e. ; a positive value indicates that the intervention has a positive effect; the system state coefficient is fitted by performing multiple regression analysis on the historical event data set, i.e., using mathematical optimization methods such as least squares method to find the optimal parameters to minimize the error between and the model predicted value , and imposing physical constraint conditions during the process; The linear term representing the basic effect of aeration : embodies the basic positive effect of aeration in providing dissolved oxygen and promoting microbial metabolism; The negative cross term representing the negative effect of aeration when the system is unstable : This is the core innovation of the model, which quantifies an abnormal effect, i.e., when the system itself is unstable, a higher value means that the hydraulic shear and acoustic field disturbance caused by increased aeration will exacerbate floc breakage and quorum sensing out of control, resulting in a negative effect that is proportional to the increase in aeration and the system risk; The quadratic negative term representing the inherent destructive effect of excessive aeration : embodies the fact that excessive aeration will destroy sludge flocculation due to excessive hydraulic shear, regardless of the system state, which is an inherent negative effect; The constraint, the aeration disturbance effect value It concerns the relative adjustment rate of aeration volume. It is a quadratic function opening downwards; therefore, the intervention decision module can determine the optimal aeration adjustment rate by solving for the maximum value of the aeration disturbance effect. Based on the above model, by solving... The optimal solution can be obtained directly: ; In calculation Then, the system will execute specific tiered intervention strategies: Level 1 Intervention: When the system enters the Level 1 warning state, the system executes parallel coordinated control; one of these is to execute the main control path, that is, to immediately execute the optimal aeration adjustment rate. The first method involves two steps: First, parallel activation of the auxiliary control path. The system will then operate based on the weights of the parameters output by the first processing module. Identify the risk index The main driving factors for the rise; for example, if the weight Corresponding to swarm induction signal At the current maximum time step, the system will determine that quorum sensing runaway is the primary risk. Therefore, while optimizing aeration, it will implement targeted suppression measures, such as precisely adding trace amounts of quorum sensing inhibitors to key areas such as the A / O interface. Level 2 Intervention: When the system enters Level 2 warning status, it indicates a more critical situation; at this time, the system executes the optimal aeration adjustment rate. To stabilize the fundamentals, and based on the main drivers causing the deterioration of the risk index, initiate at least one of the following most relevant emergency measures: If weight Corresponding fractal dimension The significantly higher weighting of the sludge floc structure indicates that physical damage to the sludge floc structure is the primary problem. Therefore, it is recommended to add polymeric flocculants to the influent of the secondary sedimentation tank to rapidly enhance the flocculation and settling performance of the sludge.

[0026] If weight Corresponding EPS rate ratio and weight Corresponding to swarm induction signal The fact that the main contributing factor indicates that microbial metabolic imbalance is the core cause suggests that the sludge return ratio should be adjusted first to change the microbial concentration and sludge age in the aeration tank, in an attempt to reshape the microecological balance.

[0027] If weight Corresponding to the synergistic toxicity gradient If the level remains high, it indicates that the toxic impact of the influent is the root cause of the risk. Therefore, instructions should be sent to the next higher level of treatment unit to temporarily reduce the influent load and alleviate the system's treatment pressure from the source. Robustness and safety constraints of control output: To ensure the stability and safety of the system, the calculated optimal aeration adjustment rate Boundary constraints are performed; set a maximum adjustment rate and minimum adjustment rate that conforms to the physical and engineering practice, for example, ±30%, the final executed aeration adjustment rate is determined according to the following rules: ; In addition, the system has built-in safety check logic. When the input critical index Ψ or the calculated far exceeds the range of historical data, the system will trigger an abnormal state, suspend automatic adjustment, and send an alarm to the operator for manual confirmation, to prevent the model from outputting unreasonable control instructions in extreme working conditions outside the training data range.

[0028] Embodiment 7: After implementing the hierarchical intervention strategy, return to the data acquisition module to continuously monitor the data and recalculate the critical index, and dynamically adjust the intervention intensity according to the recalculated critical index, forming a closed-loop negative feedback control; In the last loop of the system closed loop of embodiment 1, to ensure the effectiveness of the intervention measures and realize true adaptive control, the present application introduces a dynamic feedback correction mechanism; Specifically, after implementing the hierarchical intervention strategy of any level, the system's workflow does not terminate, but immediately returns to the data acquisition module, forming a continuous closed-loop negative feedback control loop as claimed by the present application, from perception-evaluation-decision-execution-re-perception; In this loop, the system will continuously monitor various multidimensional real-time data, and recalculate the critical index trend based on the latest data; the control system will closely monitor the response of value after the intervention measures take effect; If value presents a continuous downward trend after intervention and eventually returns to the first warning threshold safe interval below, the system will determine that the risk has been effectively controlled, and will gradually exit the intervention state, such as restoring the aeration adjustment rate to 0, stopping the addition of reagents, etc., to avoid unnecessary resource consumption; If value remains high after intervention, or even continues to rise, the system will determine that the current intervention intensity is insufficient or that the risk factors have changed; at this time, the system will recalculate and update the optimal aeration adjustment rate according to the new value of the recalculated critical index according to embodiment 6 and dynamically adjust the intervention intensity, for example, increase the dosage of the inhibitor in the first-level intervention, or upgrade the intervention level from the first level to the second level; In this way, the system realizes complete closed-loop negative feedback control from perception-evaluation-decision-execution-re-perception.

[0029] The above is only a preferred embodiment of the present application, and is not intended to limit the scope of protection of the present application; any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

[0030] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A wastewater early warning and treatment analysis system based on deep learning, characterized in that, include: The data acquisition module is used to acquire multi-dimensional real-time data of the activated sludge system in urban wastewater treatment plants and construct a time-series feature dataset. The first processing module is used to output dynamic weight coefficients based on the time-series feature dataset through deep learning model training and inference. The second processing module is used to combine multi-dimensional real-time data, dynamic weighting coefficients, and preset parameter reference benchmarks to calculate the critical index that characterizes the system risk in real time. The early warning determination module is used to compare the critical index with the preset early warning threshold to determine the risk level of the system. The intervention decision module is used to generate and execute tiered intervention strategies in response to the risk level determined by the early warning judgment module.

2. The wastewater early warning and treatment analysis system based on deep learning according to claim 1, characterized in that, The multi-dimensional real-time data includes microscopic biological signals, mesoscopic structural morphology, time-varying dynamic biochemical data, and physical field environment data. Microscopic biological signals are the concentrations of quorum sensing signal molecules; The mesoscopic structural morphology is the fractal dimension of sludge flocs; Time-varying dynamic biochemical data represent the dynamic secretion-degradation rate ratio of extracellular polymers; The physical field environmental data includes acoustic anomaly factors and co-toxicity gradients.

3. The wastewater early warning and treatment analysis system based on deep learning according to claim 1, characterized in that, The first processing module is specifically used for: A hybrid model combining convolutional neural networks and long short-term memory networks is used to process the time-series data of all parameters collected by the data acquisition module, in order to learn the nonlinear coupling relationship between the parameters and output dynamic weight coefficients.

4. The wastewater early warning and treatment analysis system based on deep learning according to claim 1, characterized in that, The reference values ​​for these parameters are set based on historical data statistics from the long-term stable operation of the wastewater treatment plant. The warning threshold is set based on historical data prior to sludge floating incidents.

5. The wastewater early warning and treatment analysis system based on deep learning according to claim 1, characterized in that, The early warning determination module is specifically used for: Compare the critical index with the first and second early warning thresholds; When the critical index is less than or equal to the first warning threshold, the system is determined to be in a safe state. When the critical index is greater than the first warning threshold and less than or equal to the second warning threshold, the system is determined to be in a level one warning state. When the critical index is greater than the second warning threshold, the system is determined to be in a level two warning state.

6. The wastewater early warning and treatment analysis system based on deep learning according to claim 5, characterized in that, The intervention decision module is also used for: When the system enters a Level 1 or Level 2 warning state, the aeration disturbance effect value is calculated based on the current critical index calculated by the second processing module. The aeration disturbance effect value is a function of the relative adjustment rate of aeration volume; The optimal aeration adjustment rate is determined by solving for the maximum value of the aeration disturbance effect.

7. A wastewater early warning and treatment analysis system based on deep learning according to claim 6, characterized in that, The calculation of the aeration disturbance effect value is based on an effect model that establishes a mathematical relationship between the aeration adjustment amount and the system risk state index. The effect model includes: a linear term characterizing the basic role of aeration; a negative cross term characterizing the negative effects of aeration when the system is unstable; and a quadratic negative term characterizing the destructive effect of over-aeration itself.

8. A wastewater early warning and treatment analysis system based on deep learning according to claim 6, characterized in that, The tiered intervention strategy includes a first-level intervention; the first-level intervention is used to execute the optimal aeration adjustment rate when the system enters the first-level warning state, and to identify the main driving factors that cause the risk index to rise based on the weights of each parameter output by the first processing module, so as to execute targeted suppression measures.

9. A wastewater early warning and treatment analysis system based on deep learning according to claim 6, characterized in that, The tiered intervention strategy includes a secondary intervention; the secondary intervention is used to, when the system enters a secondary warning state, execute the optimal aeration adjustment rate and initiate at least one of the following measures: Add polymeric flocculant; Adjust the sludge return ratio; Reduce the influent load.

10. A wastewater early warning and treatment analysis system based on deep learning according to claim 1, characterized in that, After implementing the tiered intervention strategy, the system returns to the data acquisition module to continuously monitor the data and recalculate the critical index. The intervention intensity is then dynamically adjusted based on the recalculated critical index, forming a closed-loop negative feedback control.

Images