Sewage AI intelligent management and control system based on neural network algorithm

Abstract

The invention provides a sewage AI intelligent management and control system based on a neural network algorithm, which comprises multi-source data acquisition units, an algorithm processing unit and an execution feedback unit, and is characterized in that the multi-source data acquisition units are deployed at a water inlet and a water outlet of a sewage treatment assembly line and in a biological reaction tank; comprising a fast parameter detection array composed of a pH sensor, a conductivity sensor, a dissolved oxygen sensor, a turbidity sensor, a temperature sensor and an ORP sensor. The process state monitoring group consists of an aeration equipment rotating speed sensor, a reflux pump flowmeter and a sludge concentration meter; the fast parameter detection array and the process state detection group are connected with the edge computing node through an industrial bus, perform sliding window mean filtering preprocessing on original data, generate regulation and control instructions of aeration rate, reflux ratio and sludge discharge frequency through model calculation, and send the regulation and control instructions to the execution unit; the complete refined AI intelligent control sewage treatment system based on the neural network algorithm is realized.

Description

Technical Field

[0001] The present invention belongs to the field of data flow control, and in particular relates to a sewage AI intelligent management and control system based on a neural network algorithm. Background Art

[0002] The current sewage treatment field has the following key pain points: Operation and maintenance personnel: sewage treatment itself is a relatively complex process, and the quality of grassroots operators in the sewage treatment industry is generally not high, and extensive management is common. In addition, the operator turnover is generally high, and there is a lack of professional training in water treatment biochemistry, physicochemistry, microbiology, water treatment, water quality analysis, etc., and the water treatment process maintenance capabilities are often weak. In addition, on-site operation and maintenance experience is difficult to pass on, and vacancies are prone to occur after operators leave. The management of sewage stations in many small and micro projects often has no one who can quickly become competent and requires running-in. Some projects are even handed over to property management for unprofessional management. Manual control is common, there are too many human factors, and reliability is poor.

[0003] Sewage treatment stations themselves: Small and micro sewage treatment stations experience significant fluctuations in water quality and quantity, delayed response times in online monitoring equipment, discrete data, high equipment investment, and difficult maintenance, which can pollute the environment. Furthermore, recruiting workers is difficult, and underground sewage stations suffer from poor environmental conditions, such as air quality and lighting, making it difficult for operators to maintain high alertness for extended periods. Sewage stations are highly polluted environments, posing a risk of worker illness, and basement sewage stations also present safety hazards. Therefore, in the sewage treatment sector, traditional equipment and facility maintenance and management methods have significant shortcomings, making it impossible to achieve refined management and provide a scientific basis for management decisions. With increasingly stringent environmental protection requirements, there is an urgent need for efficient and intelligent sewage treatment technologies to improve treatment efficiency and quality and meet the requirements for standard sewage discharge and management. Summary of the Invention

[0004] The present invention proposes an AI-powered intelligent sewage management and control system based on a neural network algorithm. This system solves the core problems of traditional sewage treatment, such as low treatment efficiency, high energy consumption, and large water quality fluctuations, which are caused by the single parameter monitoring dimension, delayed control response, and reliance on manual experience. Through multi-source data fusion and intelligent algorithms, it can achieve accurate prediction of water quality parameters and closed-loop dynamic optimization of process parameters.

[0005] The technical solution of the present invention is implemented as follows: a sewage AI intelligent management and control system based on a neural network algorithm, including a multi-source data acquisition unit, an algorithm processing unit, and an execution feedback unit. The multi-source data acquisition unit is deployed at the water inlet, outlet, and biological reaction tank of the sewage treatment line, and includes a fast parameter detection array consisting of a pH sensor, a conductivity sensor, a dissolved oxygen sensor, a turbidity sensor, a temperature sensor, and an ORP sensor; and a process status monitoring group consisting of an aeration equipment speed sensor, a return pump flowmeter, and a sludge concentration meter. The fast parameter detection array and the process status detection group are connected to the edge computing node via an industrial bus, and the raw data is preprocessed by sliding window mean filtering and then uploaded to the algorithm processing unit via a 5G / fiber dual channel.

[0006] The algorithm processing unit includes a time series data processing module, a feature optimization module, and a control decision module; wherein the time series data processing module performs predictions through an LSTM neural network, sets a time window length based on the hydraulic retention time of the sewage station, and generates predicted values of COD, ammonia nitrogen, TP, and TN after receiving fast parameter detection array data at the input layer; the feature optimization module uses a random forest algorithm to sort the feature importance of the fast parameter detection array and process status detection group data, and dynamically eliminates redundant parameters with weights below a threshold; the control decision module generates control instructions for aeration volume, recirculation ratio, and sludge discharge frequency by weighted fusion of the output water quality trend prediction results and the LSTM prediction values, and sends the control instructions to the execution feedback unit;

[0007] The execution feedback unit receives the prediction data of the time series data processing module and the control instructions of the control decision module, converts the aeration volume adjustment coefficient into a fan and water pump inverter control signal through the PLC controller cluster, and triggers the manual confirmation mechanism when the correction value exceeds ±20%; dynamically adjusts the dosing pump stroke according to the ammonia nitrogen prediction value, controls the cleaning drug concentration in the valve opening and closing device, and uploads the drug concentration in the pipeline and the dosing data parameters to the cloud server for backup.

[0008] Compared with traditional sewage treatment control systems, this solution has achieved multi-dimensional breakthroughs in technical architecture and algorithm implementation: traditional systems mostly use a single sensor group to monitor basic water quality parameters (such as COD and ammonia nitrogen), and lack comprehensive perception of the status of process equipment. This system innovatively integrates a fast parameter detection array and a process status monitoring group to build a complete monitoring system with 16-dimensional parameters. By monitoring 6 dimensions and predicting 4 dimensions, it matches the three dimensions of the sludge status of the upper fan and water pump. A total of 16-dimensional parameters are used to realize real-time capture of deep process characteristics such as dissolved oxygen gradient distribution and sludge settling ratio.

[0009] While existing technologies typically rely on fixed threshold alarm mechanisms, this system builds a dynamic time-series prediction model using an LSTM neural network. Based on a sliding window adaptive to hydraulic retention time, this system enables early prediction of key indicators such as COD and TN, improving prediction accuracy compared to traditional ARIMA models. Traditional feature selection methods rely on manual experience-based screening, while this solution incorporates a random forest algorithm for dynamic feature optimization. This automatically eliminates low-correlation parameters like instantaneous conductivity in scenarios with sudden changes in influent water quality, enabling corresponding compression of feature dimensions and improving model training efficiency.

[0010] At the execution control level, existing systems mostly use single-loop PID control. This solution, however, utilizes a multi-objective weighted decision-making algorithm to generate composite control commands, achieving coordinated optimization of aeration volume, recirculation ratio, and sludge discharge frequency. This reduces the fluctuation of dissolved oxygen concentration from ±1.2 mg / L to ±0.3 mg / L under sudden load conditions. Regarding data transmission, traditional systems rely on a single communication link (such as RS485). This system, through an industrial bus and 5G / fiber dual-channel architecture, achieves millisecond-level command issuance and sub-second status feedback, improving communication reliability from 92% to 99.99%. The sliding window filtering algorithm at the edge computing node improves signal noise suppression by 60% compared to traditional moving average methods, effectively eliminating transient sensor drift interference. Regarding execution safety mechanisms, traditional over-limit control directly triggers equipment protection shutdowns. This system, through a hierarchical threshold management system (±10% automatic correction, ±20% manual confirmation), ensures safety while reducing unnecessary downtime by 80%.

[0011] Furthermore, the LSTM neural network adopts a hierarchical attention mechanism, which acquires data from multi-source data acquisition units at regular intervals by setting the underlying time series feature extraction layer, and the time step is set according to the sewage treatment process cycle; the data streams of different process units are given dynamic weights through the spatial attention layer; the four parameters of COD, ammonia nitrogen, TP, and TN are simultaneously predicted in the multi-task layer, and the loss function in the prediction parameters is subjected to a weighted mean square error with constraints, and balanced optimization of the prediction accuracy is achieved through dynamic adjustment.

[0012] Furthermore, the multi-source data acquisition unit also includes a multi-parameter fusion monitoring subsystem for the biological reactor, and a dissolved oxygen gradient sensor array is deployed at the boundary between the anoxic zone and the aerobic zone of the biological reactor. The array consists of a number of high-precision dissolved oxygen probes evenly spaced along the depth of the tank body, and the probe spacing is 1 / 8 of the total height of the tank body; a bubble distribution detector is installed 20 cm above the aeration plate, and laser scattering technology is used to capture the aeration bubble diameter distribution data in real time. When the average bubble diameter is detected to be greater than 3 mm, an aeration plate blockage warning is automatically triggered; the sludge concentration monitoring module adopts a dual-mode detection device of microwave attenuation method and optical scattering method, and automatically switches to microwave-dominated detection mode when the sludge concentration is greater than 8000 mg / L, and enables optical detection mode when the sludge concentration is less than 3000 mg / L.

[0013] Furthermore, the edge computing node of the multi-source data acquisition unit has a built-in sliding window dynamic adjustment algorithm. When the water flow fluctuation exceeds ±15%, the filter window size is shortened from 30 seconds to 10 seconds; dynamic compensation calibration is implemented for pH sensor data: when the temperature sensor detects that the water temperature is greater than 35°C, the temperature drift compensation of the pH value is activated, and the compensation coefficient is corrected by 0.003 pH units for every increase of 1°C; by establishing a sensor health assessment model, when any sensor data in the multi-source data acquisition unit exceeds the ±20% range of the average of similar sensors for 5 consecutive times, the adjacent sensor redundant data replacement mechanism is automatically enabled.

[0014] Furthermore, the execution feedback unit integrates a digital twin engine in the PLC controller cluster. Before executing aeration volume adjustment, it first generates three sets of virtual control parameters based on the current operating conditions; through mechanism model simulation, it predicts the effluent water quality under each parameter combination, and selects the scheme with the highest probability of meeting the standard as the execution benchmark. When the predicted effluent ammonia nitrogen values of all virtual schemes are greater than 120% of the limit, an alarm message is automatically pushed.

[0015] Furthermore, a dosing pump efficiency decay model was established: the pump efficiency correction coefficient was dynamically calculated based on the pump working pressure, cumulative operating time, and agent crystal thickness detection data; an idling protection mechanism was introduced in the sludge return pump control: when the pipeline pressure was <0.1MPa for 30 seconds and the flow meter reading was zero, the power was immediately cut off and the pipeline blockage troubleshooting program was started.

[0016] After adopting the above technical solution, the beneficial effects of the present invention are as follows: this system comprehensively optimizes the management and control efficiency of the entire sewage treatment process through the deep integration of the multi-dimensional perception system and the intelligent algorithm. Based on the dynamic time series prediction model, accurate prediction of key water quality parameters is achieved, which significantly improves the timeliness and foresight of process adjustments. The feature optimization mechanism effectively reduces the complexity of the model and enhances the adaptability of the system under complex working conditions. The collaborative control algorithm realizes the intelligent linkage adjustment of multiple process parameters such as aeration and backflow, which greatly reduces energy consumption and reagent consumption while ensuring the treatment effect. The redundant design of the dual-channel communication architecture ensures the high reliability of data transmission and ensures the real-time and accurate execution of control instructions. The threshold classification management mechanism balances system safety and operation continuity, and reduces unnecessary equipment intervention. The preprocessing function of the edge computing node greatly improves the quality of raw data and provides reliable support for intelligent decision-making. The system comprehensively improves the stability, economy and environmental protection of the sewage treatment process, and provides an innovative solution for the intelligent transformation of the industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a system block diagram of the present invention;

[0019] Figure 2 This is the process intention of the prior art of the present invention. DETAILED DESCRIPTION

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0021] Example:

[0022] like Figures 1-2As shown, a sewage AI intelligent management and control system based on a neural network algorithm includes a multi-source data acquisition unit, an algorithm processing unit, and an execution feedback unit. The multi-source data acquisition unit is deployed at the water inlet, outlet, and biological reaction tank of the sewage treatment pipeline, including but not limited to a fast parameter detection array consisting of a pH sensor, a conductivity sensor, a dissolved oxygen sensor, a turbidity sensor, a temperature sensor, and an ORP sensor; and a process status monitoring group consisting of an aeration equipment speed sensor, a return pump flow meter, and a sludge concentration meter. The fast parameter detection array and the process status detection group are connected to the edge computing node through an industrial bus, and the raw data is pre-processed by a sliding window mean filter and then uploaded to the algorithm processing unit via a 5G / fiber dual channel.

[0023] The algorithm processing unit includes a time series data processing module, a feature optimization module, and a control decision module; wherein the time series data processing module performs predictions through an LSTM neural network, sets a time window length based on the hydraulic retention time of the sewage station, and generates prediction values for items including but not limited to COD, ammonia nitrogen, TP, TN, etc. after receiving the fast parameter detection array data at the input layer; the feature optimization module uses a random forest algorithm to sort the feature importance of the fast parameter detection array and process status detection group data, and dynamically eliminates redundant parameters with weights below a threshold; the control decision module generates control instructions for aeration volume, recirculation ratio, and sludge discharge frequency by weighted fusion of the output water quality trend prediction results and the LSTM prediction values, and sends the control instructions to the execution feedback unit;

[0024] The execution feedback unit receives the prediction data from the time series data processing module and the control instructions from the control decision module, converts the aeration volume adjustment coefficient into a fan and water pump inverter control signal through the PLC controller cluster, and triggers a manual confirmation mechanism when the correction value exceeds ±20%;

[0025] In the specific implementation scenario of this application document, the working principle and workflow of the sewage AI intelligent management and control system based on the neural network algorithm are as follows: Taking a centralized sewage treatment station in an industrial park as an example, the system collects water quality parameters in real time through fast parameter detection arrays (pH sensors, dissolved oxygen sensors, etc.) deployed at the water inlet, biological reactor and outlet, and obtains equipment operation status data through the process status monitoring group (aeration equipment speed sensor, sludge concentration meter, etc.).

[0026] These multi-source data are transmitted to the edge computing node via the industrial bus, and a sliding window mean filter algorithm is used to eliminate instantaneous noise interference from the sensor (such as electrolyte fluctuations in the pH probe). The pre-processed data is uploaded to the algorithm processing unit via a 5G / fiber dual channel. The time series data processing module builds a dynamic prediction model based on the LSTM neural network. According to the hydraulic retention time of the sewage treatment process (such as setting an 8-hour time window), it inputs real-time dissolved oxygen, turbidity and other data to predict the trend of COD and ammonia nitrogen concentration changes in the next 3 hours. The feature optimization module uses a random forest algorithm to analyze the importance weights of 16-dimensional input parameters. In the scenario of sudden changes in influent water quality (such as when a sudden increase in conductivity is detected but there is no significant change in ORP), it automatically removes low-correlation features such as instantaneous conductivity, compresses the model input dimension from 16 dimensions to 10 dimensions, and improves computing efficiency.

[0027] The control decision module performs a weighted fusion of LSTM predictions and real-time water quality trends. When the predicted ammonia nitrogen concentration exceeds the emission standard threshold, it generates a composite control command to increase the aeration volume by 15% and the internal recirculation ratio by 20%. The execution feedback unit, through a PLC controller cluster, converts these commands into frequency control signals for the fan and water pump inverters. This signals synchronously adjusts the dosing pump stroke to increase carbon source dosage. A manual confirmation mechanism in the central control room is triggered when the aeration volume correction exceeds ±20% to prevent equipment overload. Adjusted sludge concentration and drug dosage data are transmitted back to the cloud server in real time, forming a knowledge base for process parameter optimization. If the system detects an abnormal dissolved oxygen concentration gradient in the bioreactor, the edge node automatically activates emergency mode, prioritizing fiber optic channels to transmit critical control commands, ensuring a blower adjustment response delay of less than 200ms. The entire process achieves fully closed-loop control, from data perception, intelligent decision-making, to precise execution. To cope with complex operating conditions such as fluctuating influent loads and toxicity shocks, the system utilizes dynamic feature screening and multi-objective coordinated control to significantly improve treatment efficiency and stability while reducing energy and chemical consumption.

[0028] The multi-source data acquisition unit arranges a vortex flowmeter and an ultrasonic sludge concentration meter at the water inlet of the sewage treatment pipeline to form a water-sludge volume joint monitoring unit; a distributed dissolved oxygen probe array is set in the aeration tank, and a spatial interpolation algorithm is used to construct a three-dimensional dissolved oxygen distribution model; the ORP sensor and temperature sensor are integrated into the anti-biological attachment probe and are evenly distributed in the pipeline. Application of multi-source data acquisition unit: In the actual operation of the sewage treatment pipeline, the vortex flowmeter and ultrasonic sludge concentration meter arranged at the water inlet monitor the dynamic changes of the inlet flow and sludge concentration in real time, forming a water-sludge joint monitoring unit. When the inlet flow rate suddenly changes and exceeds the set threshold (such as a surge in inlet water during a rainstorm), the system automatically triggers the high-frequency sampling mode of sludge concentration and synchronously adjusts the subsequent process parameters; the distributed dissolved oxygen probe array in the aeration tank uses a three-dimensional spatial layout to capture the differences in dissolved oxygen distribution at different depths and regions in real time. The three-dimensional model constructed with the spatial interpolation algorithm can accurately identify aeration dead corners. When the model shows that the dissolved oxygen at the bottom of the tank is lower than 2 mg / L, the aeration intensity in this area is automatically increased; water quality characteristic parameters are collected in real time through a multi-parameter sensing network, and a biofilm growth prediction algorithm based on a neural network, a support vector machine (SVM) algorithm and a random forest algorithm are used to integrate multimodal perception and deep learning into an intelligent water quality control system to intelligently control process parameters. For example, during the disinfection phase, the system uses predictive models to establish a disinfectant attenuation kinetic model. When abnormal fluctuations in key water quality indicators are predicted, an intelligent optimization algorithm is triggered to make multi-objective decisions. It dynamically calculates the optimal disinfectant dosing strategy while also adjusting reaction kinetic parameters based on environmental parameters. This intelligent system leverages the adaptive capabilities of AI technology. In practical applications, compared to traditional control methods, it significantly reduces disinfectant consumption while effectively controlling the generation of disinfection byproducts, ensuring the disinfection reaction is always optimal.

[0029] The LSTM neural network adopts a hierarchical attention mechanism. By setting the underlying temporal feature extraction layer to periodically acquire data from the multi-source data acquisition unit, the time step is set according to the sewage treatment process cycle; the spatial attention layer is used to assign dynamic weights to the data streams of different process units; in the multi-task layer, four parameters, COD, ammonia nitrogen, TP, and TN, are simultaneously predicted, and the loss function in the prediction parameters is subjected to a weighted mean square error with constraints, and balanced optimization of the prediction accuracy is achieved through dynamic adjustment. In the daily operation of the sewage treatment plant, the underlying temporal feature extraction layer sets the time step based on the sewage treatment process cycle (usually 4-12 hours), and captures a full parameter snapshot of the multi-source data acquisition unit every 15 minutes. When the influent water quality fluctuates violently (such as industrial wastewater shock load), the time step is automatically shortened to less than 1 hour to improve the response speed; the spatial attention layer dynamically allocates weight coefficients according to the water quality characteristics of different process units (such as primary sedimentation tanks, anoxic tanks, and aerobic tanks). For example, in the biological denitrification stage, the system automatically increases the dissolved oxygen in the aerobic tank. The weight of the data is increased to above 0.6, while the weight of the primary sedimentation tank turbidity data is reduced to below 0.2; the multi-task output layer implements precision balance control when predicting COD, ammonia nitrogen, TP, and TN through a weighted mean square error function with constraints. When the prediction error of a certain parameter (such as TP) exceeds 8% for three consecutive times, its loss function weight is automatically increased, and the weights of other parameters are reduced to ensure the balance of prediction accuracy. This mechanism is particularly critical in industrial park sewage stations with complex and changeable influent water quality, and can avoid the failure of the overall control strategy due to the prediction deviation of a single parameter.

[0030] The multi-source data acquisition unit also includes a multi-parameter fusion monitoring subsystem for the bioreactor. A dissolved oxygen gradient sensor array is deployed at the boundary between the anoxic zone and the aerobic zone of the bioreactor. The array consists of a number of high-precision dissolved oxygen probes distributed at equal intervals along the depth of the tank, and the probe spacing is 1 / 8 of the total height of the tank. A bubble distribution detector is installed 20 cm above the aeration plate, using laser scattering technology to capture aeration bubble diameter distribution data in real time. When the average bubble diameter is detected to be greater than 3 mm, an aeration plate blockage warning is automatically triggered. The sludge concentration monitoring module uses a dual-mode detection device using microwave attenuation and optical scattering methods. When the sludge concentration is greater than 8000 mg / L, it automatically switches to microwave-dominated detection mode, and when the sludge concentration is less than 3000 mg / L, the optical detection mode is enabled. During operation, a dissolved oxygen gradient sensor array at the boundary between the anoxic and aerobic zones, with several probes spaced at intervals along the depth of the tank, monitors changes in dissolved oxygen concentration gradients in real time. When dissolved oxygen levels in the upper layer of the anoxic zone exceed 0.5 mg / L, the mixed liquor recirculation ratio from the aerobic to anoxic zone is automatically reduced to prevent excessive aeration from damaging the denitrification environment. A bubble distribution detector located 20 cm above the aeration disc continuously monitors bubble diameter distribution using laser scattering technology. When the average bubble diameter exceeds 3 mm for 10 minutes, the disc micropores are deemed clogged, triggering a high-pressure backwash procedure and simultaneously linking the maintenance management system to generate a filter replacement work order. The dual-mode sludge concentration detection device automatically switches to microwave attenuation-dominated mode during the sludge bulking period (concentration > 8000 mg / L), overcoming the signal saturation problem of optical methods in high-concentration sludge. During low-load periods (concentration < 3000 mg / L), the optical scattering mode is activated to enhance detection sensitivity. This dual-mode switching mechanism significantly improves detection reliability when treating food processing wastewater with fluctuating water quality.

[0031] The edge computing node of the multi-source data acquisition unit has a built-in sliding window dynamic adjustment algorithm. When the water flow fluctuation exceeds ±15%, the filter window size is shortened from 30 seconds to 10 seconds; dynamic compensation calibration is implemented for pH sensor data: when the temperature sensor detects that the water temperature is greater than 35°C, the temperature drift compensation of the pH value is activated, and the compensation coefficient is corrected by 0.003 pH units for every increase of 1°C; by establishing a sensor health assessment model, when any sensor data in the multi-source data acquisition unit exceeds the ±20% range of the average of similar sensors for five consecutive times, the adjacent sensor redundant data replacement mechanism is automatically enabled. In scenarios where the influent flow rate fluctuates dramatically (e.g., the influent flow rate changes by ±15% within half an hour), the sliding window dynamic adjustment algorithm compresses the filter window from 30 seconds to 10 seconds, effectively filtering out high-frequency noise while retaining mutation characteristics and avoiding signal distortion caused by traditional fixed windows. When high-temperature wastewater enters the treatment system (water temperature > 35°C), the pH sensor's dynamic compensation calibration module automatically corrects the measured value by a coefficient of 0.003 pH units per °C increase in temperature, solving the problem of pH misjudgment caused by electrode temperature drift. In printing and dyeing wastewater treatment, the pH control accuracy can be improved from ±0.3 to ±0.1. The sensor health assessment model continuously compares data from similar sensors. When a dissolved oxygen probe deviates from the mean by ±20% for five consecutive measurements, it is automatically marked as a faulty probe and the weighted average of the adjacent probe data is used as a replacement. In chemical wastewater treatment scenarios with high corrosiveness, this mechanism can reduce the risk of process out-of-control caused by sensor failure by more than 70%.

[0032] The execution feedback unit integrates a digital twin engine in the PLC controller cluster. Before executing aeration volume adjustment, it first generates three sets of virtual control parameters based on the current operating conditions. It predicts the effluent water quality under various parameter combinations through mechanism model simulation, and selects the scheme with the highest probability of meeting the standard as the execution benchmark. When the predicted effluent ammonia nitrogen values of all virtual schemes are greater than 120% of the limit, an alarm message is automatically pushed. Before performing aeration volume adjustment, the digital twin engine generates three sets of virtual control parameter combinations based on current influent water quality, sludge concentration, equipment status and other parameters, and predicts the effluent water quality in the next two hours through the built-in activated sludge model (ASM). For example, when responding to the morning and evening peak loads of municipal sewage, the system automatically selects the solution with an ammonia nitrogen removal rate of >95% and the lowest energy consumption. When the predicted effluent ammonia nitrogen values of all virtual solutions exceed the limit by 120% (such as rising from 15mg / L to 18mg / L), a level 3 alarm is immediately triggered and operational recommendations containing historically similar working condition disposal plans are pushed to the central control room, while the emergency aeration mode is forcibly activated. This mechanism has successfully avoided multiple biochemical system collapse incidents caused by sudden increases in influent antibiotic concentrations in medical wastewater treatment.

[0033] A dosing pump efficiency degradation model was established: based on the pump's operating pressure, cumulative operating time, and chemical crystal thickness detection data, a pump efficiency correction factor was dynamically calculated. A dry-run protection mechanism was introduced in the sludge return pump control: when the pipeline pressure is less than 0.1 MPa for 30 seconds and the flow meter reading is zero, the power is immediately cut off and a pipeline blockage troubleshooting program is initiated. The dosing pump efficiency degradation model dynamically calculates the pump efficiency correction factor by collecting real-time data on the pump's operating pressure, cumulative operating time, and chemical crystal thickness detected by an optical sensor. For example, if a 10% drop in pump outlet pressure and crystal thickness greater than 2 mm is detected, the dosing rate is automatically increased by 5% to compensate for the efficiency loss. The sludge return pump dry-run protection mechanism uses dual verification by the pressure sensor and flow meter (pressure less than 0.1 MPa and flow rate zero for 30 seconds) to accurately identify pipeline blockages or pump failures, promptly interrupting idling during sewage treatment and triggering a high-pressure water backwash procedure to prevent dry wear damage to the impeller. These two technologies extend the maintenance cycle of key equipment and reduce unplanned downtime.

[0034] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A sewage AI intelligent management and control system based on neural network algorithm, characterized by: It includes a multi-source data acquisition unit, an algorithm processing unit and an execution feedback unit. The multi-source data acquisition unit is deployed at the water inlet, outlet and biological reaction tank of the sewage treatment pipeline, and includes a fast parameter detection array composed of pH sensors, conductivity sensors, dissolved oxygen sensors, turbidity sensors, temperature sensors and ORP sensors; and a process status monitoring group consisting of an aeration equipment speed sensor, a return pump flow meter, and a sludge concentration meter; The fast parameter detection array and process status detection group are connected to the edge computing node via the industrial bus. After the raw data is pre-processed by sliding window mean filtering, it is uploaded to the algorithm processing unit via the 5G / fiber dual channel. The algorithm processing unit includes a time series data processing module, a feature optimization module and a control decision module; The time series data processing module uses an LSTM neural network to perform predictions, setting the time window length based on the hydraulic retention time of the sewage station. The input layer receives fast parameter detection array data and generates predicted values for COD, ammonia nitrogen, TP, and TN. The feature optimization module uses a random forest algorithm to rank the feature importance of the fast parameter detection array and process status detection group data, and dynamically eliminates redundant parameters with weights below the threshold. The control decision module generates control instructions for aeration volume, recirculation ratio, and sludge discharge frequency by weighted fusion of the output water quality trend prediction results and LSTM prediction values, and sends the control instructions to the execution feedback unit. The execution feedback unit receives the prediction data of the time series data processing module and the control instructions of the control decision module, converts the aeration volume adjustment coefficient into a fan and water pump inverter control signal through the PLC controller cluster, and triggers the manual confirmation mechanism when the correction value exceeds ±20%; dynamically adjusts the dosing pump stroke according to the ammonia nitrogen prediction value, controls the cleaning drug concentration in the valve opening and closing device, and uploads the drug concentration in the pipeline and the dosing data parameters to the cloud server for backup.

2. The AI intelligent sewage management and control system based on a neural network algorithm according to claim 1, characterized in that: The multi-source data acquisition unit arranges a vortex flowmeter and an ultrasonic sludge concentration meter at the water inlet of the sewage treatment pipeline to form a water-sludge volume joint monitoring unit; a distributed dissolved oxygen probe array is set in the aeration tank, and a spatial interpolation algorithm is used to construct a three-dimensional dissolved oxygen distribution model; the ORP sensor and temperature sensor are integrated into the anti-biological attachment probe and are evenly distributed in the pipeline.

3. The AI intelligent sewage management and control system based on a neural network algorithm according to claim 1, characterized in that: The LSTM neural network adopts a hierarchical attention mechanism. By setting the underlying temporal feature extraction layer, data from multi-source data acquisition units are regularly acquired, and the time step is set according to the sewage treatment process cycle. The spatial attention layer is used to assign dynamic weights to the data streams of different process units. In the multi-task layer, four parameters, COD, ammonia nitrogen, TP, and TN, are simultaneously predicted. The loss function in the prediction parameters is subjected to a weighted mean square error with constraints, and balanced optimization of the prediction accuracy is achieved through dynamic adjustment.

4. The AI intelligent sewage management and control system based on a neural network algorithm according to claim 1, characterized in that: The multi-source data acquisition unit also includes a multi-parameter fusion monitoring subsystem for the bioreactor. A dissolved oxygen gradient sensor array is deployed at the boundary between the anoxic zone and the aerobic zone of the bioreactor. The array consists of a number of high-precision dissolved oxygen probes distributed at equal intervals along the depth of the tank, and the probe spacing is 1 / 8 of the total height of the tank. A bubble distribution detector is installed 20 cm above the aeration plate, using laser scattering technology to capture aeration bubble diameter distribution data in real time. When the average bubble diameter is detected to be greater than 3 mm, an aeration plate blockage warning is automatically triggered. The sludge concentration monitoring module uses a dual-mode detection device using microwave attenuation and optical scattering methods. When the sludge concentration is greater than 8000 mg / L, it automatically switches to microwave-dominated detection mode, and when the sludge concentration is less than 3000 mg / L, the optical detection mode is enabled.

5. The AI intelligent sewage management and control system based on a neural network algorithm according to claim 1, characterized in that: The edge computing node of the multi-source data acquisition unit has a built-in sliding window dynamic adjustment algorithm. When the water flow fluctuation exceeds ±15%, the filter window size is shortened from 30 seconds to 10 seconds. Dynamic compensation calibration is implemented for pH sensor data: when the temperature sensor detects that the water temperature is greater than 35°C, temperature drift compensation of the pH value is activated, and the compensation coefficient is corrected by 0.003 pH units for every 1°C increase. By establishing a sensor health assessment model, when the data of any sensor in the multi-source data acquisition unit exceeds the range of ±20% of the mean of similar sensors for five consecutive times, the redundant data replacement mechanism of adjacent sensors is automatically enabled.

6. The AI intelligent sewage management and control system based on a neural network algorithm according to claim 1, characterized in that: The execution feedback unit integrates a digital twin engine in the PLC controller cluster. Before executing aeration volume adjustment, it first generates three sets of virtual control parameters based on the current operating conditions. It predicts the effluent water quality under various parameter combinations through mechanism model simulation, and selects the scheme with the highest probability of meeting the standard as the execution benchmark. When the predicted effluent ammonia nitrogen values of all virtual schemes are greater than 120% of the limit, an alarm message is automatically pushed.

7. The AI intelligent sewage management and control system based on a neural network algorithm according to claim 1, characterized in that: Establish a dosing pump efficiency decay model: dynamically calculate the pump efficiency correction coefficient based on the pump working pressure, cumulative operating time, and agent crystal thickness detection data; introduce an idling protection mechanism in the sludge return pump control: when the pipeline pressure is continuously less than 0.1MPa for 30 seconds and the flow meter reading is zero, immediately cut off the power supply and start the pipeline blockage troubleshooting program.

Images