Intelligent dosing system for processing wastewater of nuts based on online monitoring of water quality

Abstract

The present application relates to wastewater treatment technical field, and disclose a kind of based on water quality online monitoring's nut processing wastewater treatment intelligent dosing system, including water quality monitoring module, reagent dosing module, intelligent control module, man-machine interaction module, model self-learning module and database, the water quality monitoring module connects relevant key parameter detection module, sampling unit module, data analysis and calibration module, the present application scheme passes through multi-parameter real-time online monitoring network, system forms panorama, real-time sensing to influent water quality, has mastered control basis from source, feedforward control according to real-time influent load early control effect, quickly offset main disturbance, feedback control according to key process index fine-tuning, eliminate cumulative error, this combination ensures that system can quickly respond and keep the stability of processing effect when facing the typical intermittent, impact load of nut processing wastewater, significantly reduce the risk of effluent exceeding standard.

Description

Technical Field

[0001] This invention belongs to the field of wastewater treatment technology, specifically an intelligent dosing system for treating nut processing wastewater based on online water quality monitoring. Background Technology

[0002] In the nut processing industry, a large amount of complex industrial wastewater with fluctuating loads is generated during production. This type of wastewater is typically characterized by high concentrations of organic matter, high levels of oil and fat, high levels of suspended solids, and significant fluctuations in water quality and quantity over time. If discharged directly without proper treatment, it will cause serious pollution to the aquatic environment.

[0003] Currently, the main process route for treating this type of wastewater is "pretreatment + biological treatment". In the pretreatment stage, chemical agents (such as acid-base neutralizers, coagulants, flocculants, demulsifiers, etc.) are added for physicochemical treatment to remove grease, suspended solids and partially degrade organic matter. This is a key link to ensure the stable operation of the subsequent biological treatment unit and achieve compliant discharge. However, the existing dosing control methods have significant defects and are difficult to adapt to the characteristics of nut processing wastewater, resulting in unstable treatment effects and high operating costs. Many small and medium-sized processing enterprises still use manual timed and quantitative dosing. Operators estimate the dosage based on experience and cannot respond to the dynamic changes in influent water quality and quantity in real time. This often results in "insufficient dosing" leading to effluent exceeding standards, or "excessive dosing" causing waste of agents, increased sludge, and possible inhibition of the subsequent biological system. To address this, we propose an intelligent dosing system for nut processing wastewater treatment based on online water quality monitoring. Summary of the Invention

[0004] In order to overcome the shortcomings of the prior art, this invention provides an intelligent dosing system for nut processing wastewater treatment based on online water quality monitoring, which effectively solves the above problems.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent dosing system for treating nut processing wastewater based on online water quality monitoring, characterized in that it includes a water quality monitoring module, a reagent dosing module, an intelligent control module, a human-computer interaction module, a model self-learning module, and a database;

[0006] The water quality monitoring module is connected to a key parameter detection module, a sampling unit module, and a data analysis and calibration module.

[0007] The intelligent control module is connected to a feedforward receiving module and a drug output module;

[0008] The human-computer interaction module is connected to a monitoring and data management module and a remote operation and maintenance and early warning module.

[0009] The key parameter detection module is used to monitor pH, ORP, turbidity / suspended solids, and conductivity in nut processing wastewater. For organic matter, oil, and protein rich in nut processing wastewater, COD or TOC, UV254 (characterizing aromatic organic matter), and oil content should be monitored. Optional monitoring parameters include ammonia nitrogen and total phosphorus.

[0010] The sampling unit module is used to construct sampling points for sampling work. The sampling points need to be set up at the place where the wastewater is evenly mixed and equipped with automatic samplers with timed and flow rate ratio sampling functions to sample and test the water source near the sampling point, and then test the water source in each process of wastewater treatment to see if the treatment is qualified.

[0011] The data analysis and calibration module uses a built-in algorithm to identify and remove abnormal and drifting values ​​from sensors, and regularly (e.g., daily / weekly) compares and calibrates the data with manual laboratory testing data. Using historical data, it establishes a water quality fluctuation model to predict water quality change trends during peak production periods and processing of different nut varieties (e.g., high-oil walnuts and high-protein almonds), providing a basis for pre-dosing.

[0012] The feedforward receiving module receives the influent flow rate, the real-time values ​​of raw water quality (COD, turbidity), and the quality of treated effluent from the water quality monitoring module.

[0013] The reagent output module is used to calculate and add the basic dosage of the reagent in advance according to the preset model. Based on the quality of the treated effluent, the feedforward dosage is fine-tuned to form a closed loop, eliminate disturbances, and introduce parameter tuning of the model self-learning module to cope with the nonlinearity and time-varying nature of wastewater quality, with the optimization goal of achieving effluent compliance and the lowest cost in the future.

[0014] The self-learning module enables the system to continuously optimize itself using operational data, adapting to water quality fluctuations and equipment aging. The system continuously records complete data pairs of influent water quality, chemical dosage, process parameters, and effluent water quality. Using a unified timestamp as a benchmark, the system automatically associates and stores massive amounts of time-series data from different sources, including: sensing data, execution data, operating condition data, environmental data, data cleaning and labeling, and advanced feature engineering. The model optimization program is initiated periodically or triggered, employing a model-free Bayesian optimization algorithm. This algorithm treats the system as a "black box," aiming for the highest overall effluent quality compliance rate and the lowest chemical consumption per unit volume of water. By actively and incrementally trying new parameter combinations and observing system feedback, it efficiently finds the globally optimal or better parameter settings, automatically correcting coefficients in the feedforward model to ensure more accurate chemical dosage predictions achieve the desired treatment effect, minimizing chemical consumption over long-term operation. A comprehensive model evaluation is initiated periodically, allowing the system to... Using accumulated data, various alternative model structures are tested, and their predictive performance is evaluated through cross-validation. If the prediction error of the new model structure is found to be significantly lower than that of the existing model and passes safety verification, the model update process is triggered to replace the old model with a more accurate model, thereby achieving the "evolution" of the control strategy. This ensures that all model optimization and update operations are absolutely safe and prevents water quality from exceeding standards or equipment damage due to model errors. The system has a built-in simplified digital twin model of the process. Any new parameters or models from the optimization engine must first undergo long-term simulation testing in the digital twin sandbox to simulate various typical and extreme water intake conditions, verifying their control effect and stability. A series of insurmountable safety rules are preset, and the optimization algorithm must find the best under these hard constraints. After the model is updated, it can first be run in "shadow mode" in a parallel section or a specific time period (and its instructions are compared and analyzed with the actual control instructions). After confirming safety and effectiveness, it can then be smoothly switched to the main control loop.

[0015] The monitoring and data management module provides a graphical user interface, displaying all process parameters, equipment status, trend curves, and alarm information in real time. It automatically generates daily and monthly reports, recording chemical dosage, power consumption, water quality compliance rate, and operating costs. Data can be stored and exported, supporting environmental compliance reports.

[0016] The remote operation and maintenance and early warning module supports 4G / 5G / WIFI networking, enabling remote monitoring via mobile APP or web page, setting multi-level alarms, and timely notification of management personnel for special events;

[0017] The database is used to store the data generated by the system.

[0018] Preferably, the sampling unit module is connected to a preprocessing submodule. The preprocessing submodule is used to receive processing information from the multi-level physical preprocessing unit, and to perform synchronous and continuous multi-dimensional water quality parameter measurements on the preprocessed water sample. It adopts a multi-functional composite electrode probe, which integrates pH, ORP, conductivity / TDS, and temperature sensors. It is responsible for the aggregation, primary processing, and quality assurance of sensor data. Based on sliding window statistics (and physical quantity change rate thresholds), it identifies and removes abnormal readings caused by bubbles and instantaneous particle interference in real time. It adopts an exponential weighted moving average algorithm to effectively suppress random noise while ensuring real-time performance. It uses the theoretical correlation between parameters to perform rationality checks, marks and alarms for obviously contradictory data, and uploads the high-quality real-time data stream after preprocessing to the core control module via industrial Ethernet or wireless network.

[0019] Preferably, the model self-learning module is connected to a calibration and maintenance management sub-module. This sub-module is used to ensure the long-term measurement accuracy of the sensing module, reduce the maintenance burden, automatically generate calibration task reminders based on sensor operating time, data drift degree, or preset cycle, and can connect to the laboratory information management system to synchronize calibration plans and records. It analyzes sensor self-test signals, cleaning frequency, reading stability, etc., and provides early warnings of potential faults such as filter blockage, light source attenuation, and electrode aging. It performs real-time scoring and remaining life prediction for the "health status" of each online sensor, continuously tracks the deviation and trend of each sensor's reading from the laboratory reference method, and uses the CUSUM (cumulative sum) control chart algorithm to sensitively detect whether the sensor has begun to slowly degrade in performance. It analyzes the sensor's built-in self-test signals and uses them as direct indicators of health status. It monitors the time consumption of sensor cleaning actions, backwash pressure, calibration frequency, etc. Abnormal behavior is often a precursor to blockage or wear. Combining the above multi-dimensional data, it uses machine learning algorithms such as random survival forest to calculate the probability distribution of the "remaining useful life" of each key sensor.

[0020] Preferably, the intelligent control module is connected to a dynamic water quality feature extraction module. The module analyzes multi-parameter data streams in real time, identifies the "water quality mode" or "operating condition" of the current wastewater, and extracts the statistical features of each parameter within a sliding time window to form a "feature vector" describing the current water quality state. It presets multiple typical operating condition modes based on historical data learning, and quickly matches the real-time "feature vector" with the "operating condition mode library" to identify the most likely water quality type. This provides context for selecting the most suitable dosing control model. Based on the identified operating conditions and real-time data, it dynamically calculates the optimal instantaneous dosing rate at each dosing point.

[0021] Preferably, the reagent dosing module is connected to a safety interlock protection module. The safety interlock protection module ensures that all control actions are absolutely safe, prevents misoperation from causing process collapse or equipment damage, and sets soft start / stop logic, maximum / minimum stroke protection, and cumulative running time balancing logic for all metering pumps and valves. When an extreme anomaly is detected, the system automatically triggers the preset safety plan, such as switching the relevant dosing pump to a safe flow rate or stopping it completely, and issuing the highest level alarm.

[0022] Preferably, the database connection has a backup module. The backup module stores the data generated by the system in a third-party storage, while monitoring data changes in real time. When abnormal data changes occur, the third-party data backup is immediately blocked and the staff is notified. At the same time, the abnormal data is isolated and backed up again.

[0023] Preferably, the remote operation and maintenance and early warning module is connected to a cloud-based collaborative maintenance and knowledge base sub-module. This sub-module leverages a cloud platform to enable remote expert support, maintenance experience accumulation, and global optimization. Maintenance engineers can remotely access the device's maintenance interface through a secure VPN channel to view detailed diagnostic data, execute advanced diagnostic procedures, and even perform remote calibration, significantly reducing on-site dispatch. The system structurally records each fault phenomenon, handling measures, and root cause in the cloud-based knowledge base. When similar fault symptoms appear at a new site, the system can automatically push historical cases and solutions to assist on-site personnel in quickly troubleshooting.

[0024] Compared with the prior art, the beneficial effects of the present invention are:

[0025] 1. Through a multi-parameter real-time online monitoring network, the system achieves a panoramic and real-time perception of the influent water quality, grasping the control basis from the source. Feedforward control applies control in advance based on the real-time influent load to quickly offset the main disturbances, while feedback control makes fine adjustments based on key process indicators to eliminate cumulative errors. This combination ensures that the system can respond quickly and maintain stable treatment effect when facing the typical intermittent and shock loads of nut processing wastewater, significantly reducing the risk of effluent exceeding standards.

[0026] 2. Through the establishment of a precise control model, the system can dynamically calculate the optimal dosage according to actual needs, avoid waste of reagents, and make the cost of treating ton of water clearly controllable. The system automatically records various consumptions, providing data support for cost optimization. The system has built-in fuzzy control, PID self-tuning or advanced predictive control algorithms, which can automatically adapt to the nonlinear and time-varying characteristics of wastewater quality, reducing the dependence on the experience of operators.

[0027] 3. The model self-learning and optimization engine enables the system to continuously improve itself during operation, maintain the optimal control state in the long term, and has the ability to continuously evolve. The intelligent operation and maintenance and health management module has predictive maintenance function, which can give early warning of faults such as sensor failure and pipeline blockage, turning "passive maintenance" into "proactive maintenance" and reducing unplanned downtime.

[0028] 4. Automatic calibration and remote maintenance support functions significantly reduce on-site maintenance workload and professional requirements. Through SCADA / HMI interface and mobile remote monitoring, managers can grasp the overall system operation, water quality data and alarm information anytime and anywhere. The automatically generated reports and data analysis provide powerful and reliable data support for process optimization, environmental compliance reports and management decisions. Attached Figure Description

[0029] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0030] In the attached diagram:

[0031] Figure 1 This is a block diagram of the intelligent dosing system for treating nut processing wastewater based on online water quality monitoring, as per the present invention. Detailed Implementation

[0032] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0033] In order to solve the problems in the background art, please refer to the following: Figure 1 A smart dosing system for treating nut processing wastewater based on online water quality monitoring includes a water quality monitoring module, a reagent dosing module, an intelligent control module, a human-computer interaction module, a model self-learning module, and a database.

[0034] The water quality monitoring module is connected to a key parameter detection module, a sampling unit module, and a data analysis and calibration module;

[0035] The intelligent control module is connected to a feedforward receiving module and a drug output module;

[0036] The human-computer interaction module is connected to a monitoring and data management module and a remote operation and maintenance and early warning module;

[0037] The key parameter detection module is used to monitor pH, ORP (oxidation-reduction potential), turbidity / suspended solids (SS), and conductivity (TDS) in nut processing wastewater. This is the basis for chemical dosing (such as neutralization and coagulation) control. For nut processing wastewater rich in organic matter, oil, and protein, it is necessary to focus on monitoring COD (chemical oxygen demand) or TOC (total organic carbon), UV254 (characterizing aromatic organic matter), and oil content. Ammonia nitrogen and total phosphorus can be monitored to assess the load of subsequent biochemical treatment. Industrial online sensors with self-cleaning function should be selected. For COD / ammonia nitrogen, UV-Vis multi-parameter analyzers or electrode sensors are preferred to balance response speed, maintenance frequency, and cost.

[0038] The sampling unit module is used to construct sampling points for sampling work. The sampling points need to be set up at the place where the wastewater is uniformly mixed and equipped with automatic samplers with timed and flow ratio sampling functions to sample and test the water source near the sampling point, and then test the water source in each process of wastewater treatment to check whether the treatment is qualified. For wastewater with high suspended solids and high oil content, multi-stage pretreatment must be equipped, such as automatic filters (to remove large particles), ultrasonic homogenizers or heating devices (for demulsification and homogenization), and precision filtration units to ensure that the water sample entering the analyzer is representative and does not clog or contaminate the sensor.

[0039] The data analysis and calibration module uses the built-in algorithm of the system to identify and remove abnormal and drift values ​​of the sensors, and regularly (such as daily / weekly) compares and calibrates with manual laboratory test data. Using historical data, it establishes a water quality fluctuation model to predict the water quality change trend during peak production periods and the processing of different nut varieties (such as high-oil walnuts and high-protein almonds), providing a basis for pre-dosing.

[0040] The feedforward receiving module receives the influent flow rate, raw water quality (COD, turbidity) real-time values, and treated effluent quality from the water quality monitoring module.

[0041] The reagent output module is used to calculate and add the basic dosage of the reagent in advance according to the preset model. Based on the quality of the treated effluent (such as turbidity and pH after flocculation), the feedforward dosage is fine-tuned to form a closed loop, eliminate disturbances, and introduce parameter tuning of the model self-learning module to cope with the nonlinearity and time-varying nature of wastewater quality, with the optimization goal of achieving effluent compliance and the lowest cost in the future.

[0042] The model self-learning module enables the system to continuously optimize itself using operational data, adapting to water quality fluctuations and equipment aging. The system continuously records complete data pairs of influent water quality, chemical dosage, process parameters, and effluent water quality. Using a unified timestamp as a benchmark, the system automatically associates and stores massive amounts of time-series data from different sources, including: sensing data, execution data, operating condition data, environmental data, data cleaning and labeling, and advanced feature engineering. The model optimization program is initiated periodically (e.g., weekly) or triggered (when effluent water quality continuously deviates from the target), employing a model-free Bayesian optimization algorithm. This algorithm treats the system as a "black box," using the "comprehensive effluent water quality" as the basis for optimization. With the optimization objective of "maximizing the compliance rate and minimizing the chemical consumption per unit volume of water," the system actively and incrementally tries new parameter combinations and observes system feedback to efficiently find the globally optimal or even better parameter settings. Compared to gradient descent, it is better suited to handling noisy industrial processes with high evaluation costs (each trial can take several hours), and it avoids getting trapped in local optima. It automatically corrects the coefficients in the feedforward model, ensuring that the model's predicted chemical dosage more accurately achieves the desired treatment effect, minimizing chemical consumption over long-term operation. A comprehensive model evaluation is initiated periodically (e.g., quarterly), and the system utilizes accumulated data to try various alternative model combinations. Models (such as linear, polynomial, and decision tree-based ensemble models) are evaluated for predictive performance through cross-validation. If the prediction error of the new model structure is found to be significantly lower than that of the existing model, and it passes safety verification (e.g., simulation tests do not cause drastic fluctuations in dosage), then the model update process is triggered, replacing the old model with a more accurate one. This achieves the "evolution" of the control strategy, ensuring that all model optimization and update operations are absolutely safe and preventing effluent exceeding standards or equipment damage due to model errors. The system has a built-in simplified digital twin model of the process. Any new parameters or models from the optimization engine must first be tested in the digital twin sandbox for an extended period of time. The simulation operation test simulates various typical and extreme influent conditions to verify its control effect and stability. A series of inviolable safety rules are preset, such as "the change in the dosage of a single dosing shall not exceed ±30%" and "pH adjustment shall not cause the pH of the reaction tank to momentarily exceed the range of 5-9". The optimization algorithm must find the best under these hard constraints. After the model is updated, it can first run in "shadow mode" (i.e., receive real-time data and calculate dosing instructions, but do not actually output control) in a parallel section or a specific time period. The instructions are compared and analyzed with the actual control instructions. After confirming safety and effectiveness, it can be smoothly switched to the main control loop.

[0043] Sensing data: influent water quality with multiple parameters, and process water quality in key sections (such as neutralization tank and flocculation tank).

[0044] Execution data: Real-time dosing flow rate and cumulative dosing amount of each metering pump.

[0045] Operating data: Production line status (e.g., processing walnuts / almonds), water inlet pump flow rate, batch number.

[0046] Environmental data: reaction tank temperature, agitator status.

[0047] Data cleaning and labeling: An unsupervised learning algorithm based on variational autoencoders is used to automatically identify and label anomalous segments in the data (such as sensor failures, process start-ups and shutdowns), and isolate them from the training set. Spatiotemporal kriging interpolation is used to accurately repair missing data.

[0048] Advanced feature engineering: It not only uses raw data but also automatically generates deep-level features, such as:

[0049] Derived features: Calculate organic load (flow rate * COD) and salinity shock (rate of change in conductivity).

[0050] Statistical characteristics: mean, variance, skewness, and autocorrelation coefficient within the sliding window.

[0051] Frequency domain characteristics: Periodic components of water quality fluctuations are extracted using fast Fourier transform.

[0052] Sequence features: Constructing the change trajectory of key parameters over a past period as model input;

[0053] The monitoring and data management module provides a graphical user interface that displays all process parameters, equipment status, trend curves, and alarm information in real time. It automatically generates daily and monthly reports, recording chemical dosage, power consumption, water quality compliance rate, and operating costs. Data can be stored and exported, and environmental compliance reports are supported.

[0054] The remote operation and maintenance and early warning module supports 4G / 5G / WIFI networking, enabling remote monitoring via mobile APP or web page, setting up multi-level alarms (audio and visual, SMS, WeChat push), and promptly notifying management personnel of events such as "low reagent level", "excessive water quality", and "equipment failure".

[0055] Databases are used to store data generated by the system.

[0056] The sampling unit module is connected to a pretreatment submodule. This submodule receives processing information from multiple physical pretreatment units and performs synchronous, continuous, multi-dimensional water quality parameter measurements on the pretreated water sample, forming a real-time water quality "fingerprint." This includes features such as a series-connected automatic backwash filter (accuracy ~100μm) to remove large particulate impurities and protect subsequent units; an ultrasonic homogenizing reactor or a temperature-controlled heated stirring tank to break down emulsified oil droplets and colloids in the wastewater through ultrasonic cavitation or moderate heating (e.g., 40-60℃), uniformly dispersing oils and fine particles and promoting the precipitation of some dissolved organic matter, greatly improving the representativeness and accuracy of subsequent monitoring; and an automatic scraper-type precision filter (accuracy ~10μm) or a micro-cyclone separator to remove fine particles remaining after homogenization, producing clear and stable analytical water samples. The system also includes a constant flow pump and a back pressure valve. The system includes a buffer flow path to provide stable water sample flow with pressure and flow rate for each online sensor. It employs a multi-functional composite electrode probe that integrates pH, ORP, conductivity / TDS, and temperature sensors to achieve in-situ real-time measurement with a short response time (seconds). It is responsible for the aggregation, primary processing, and quality assurance of sensor data. Based on sliding window statistics (such as the 3σ criterion) and physical quantity change rate thresholds, it identifies and eliminates abnormal readings caused by bubbles and transient particle interference in real time. It uses an exponentially weighted moving average algorithm to effectively suppress random noise while ensuring real-time performance. It uses theoretical correlations between parameters (such as the correlation between conductivity and TDS, and UV254 and COD) for rationality checks, marks and alarms for obviously contradictory data, and uploads the pre-processed high-quality real-time data stream to the core control module via industrial Ethernet or wireless network.The model self-learning module is connected to a calibration and maintenance management submodule. This submodule ensures the long-term measurement accuracy of the sensing module, reduces maintenance burden, and automatically generates calibration task reminders based on sensor operating time, data drift, or preset cycles. It can also connect to a laboratory information management system to synchronize calibration plans and records. Authorized personnel can remotely calibrate analyzers (such as COD spectrometers) using standard samples and update models via a secure network. The module analyzes sensor self-test signals, cleaning frequency, and reading stability, providing early warnings of potential faults such as filter clogging, light source attenuation, and electrode aging. It provides real-time scoring of the "health status" of each online sensor and predicts its remaining lifespan, continuously tracking the readings of each sensor against laboratory reference methods. By analyzing the deviations and trends between the results, the CUSUM (cumulative sum) control chart algorithm is used to sensitively detect whether the sensor has begun to slowly degrade in performance. The self-test signals built into the sensor (such as light source intensity, reference voltage, and internal temperature) are analyzed and used as direct indicators of health status. The time taken for sensor cleaning, backwash pressure, and calibration frequency are monitored. Abnormal behavior (such as increasingly longer cleaning time) is often a precursor to blockage or wear. Based on the above multi-dimensional data, machine learning algorithms such as random survival forest are used to calculate the probability distribution of the "remaining useful life" of each key sensor (such as COD analyzer and pH electrode), realizing the transformation from "periodic replacement" to "precise predictive replacement", and changing "periodic maintenance" to "on-demand maintenance".

[0057] The intelligent control module is connected to a dynamic water quality feature extraction module, which analyzes multi-parameter data streams in real time, identifies the current wastewater's "water quality mode" or "operating condition," and extracts the statistical characteristics (mean, variance, and slope of change trend) of each parameter (flow rate, pH, COD, turbidity, oil content, etc.) within a sliding time window (e.g., 5 minutes) to form a "feature vector" describing the current water quality state. It presets multiple typical operating condition modes based on historical data, such as: "Walnut Processing - High Oil Mode" (features: high oil content, high COD, near-neutral pH), "Almond Blanching - High Temperature High COD Mode," and "Production Intermittent - Low Load Mode." The module quickly matches the real-time "feature vector" with the "operating condition mode library" (e.g., based on Euclidean distance or cosine similarity) to identify the most likely water quality type, providing context for selecting the most suitable dosing control model. Based on the identified operating conditions and real-time data, it dynamically calculates the optimal instantaneous dosing rate at each dosing point.

[0058] The chemical dosing module is connected to a safety interlock protection module, which ensures that all control actions are absolutely safe and prevents process failure or equipment damage caused by misoperation. For example, "starting the coagulant dosing pump" must be interlocked under the conditions that "the mixer is running" and "the influent flow rate is greater than the minimum limit"; "dosing the oxidant" must be interlocked under the condition that "the pH is within the permissible range". Soft start / stop logic, maximum / minimum stroke protection, and cumulative running time balancing logic (for multi-pump parallel systems) are set for all metering pumps and valves to extend equipment life. When extreme anomalies are detected (such as pH exceeding the safe range or critical instrument failure), the system automatically triggers the preset safety plan, such as switching the relevant dosing pump to a safe flow rate or stopping it completely, and issuing the highest level alarm.

[0059] The database connection has a backup module that stores the system-generated data in a third-party database and monitors data changes in real time. When abnormal data changes occur, the third-party data backup is immediately blocked and staff are alerted. At the same time, the abnormal data is isolated and backed up again.

[0060] The remote operation and maintenance and early warning module is connected to a cloud-based collaborative maintenance and knowledge base sub-module. This sub-module leverages the cloud platform to provide remote expert support, accumulate maintenance experience, and optimize the overall system. Maintenance engineers can remotely access the device's maintenance interface via a secure VPN channel to view detailed diagnostic data, execute advanced diagnostic procedures, and even perform remote calibration, significantly reducing on-site dispatch. The system systematically records each fault phenomenon, handling measures, and root cause in the cloud knowledge base. When similar fault symptoms appear at a new site, the system can automatically push historical cases and solutions to assist on-site personnel in quickly troubleshooting. With user authorization and anonymization, the cloud can aggregate operational data from multiple similar systems. Through big data analysis, it can discover better universal control parameters or maintenance strategies and periodically push "policy update packages" to various edge systems, ensuring that all networked systems continuously benefit from collective wisdom.

Claims

1. A smart dosing system for treating nut processing wastewater based on online water quality monitoring, characterized in that: It includes a water quality monitoring module, a reagent dosing module, an intelligent control module, a human-computer interaction module, a model self-learning module, and a database; The water quality monitoring module is connected to a key parameter detection module, a sampling unit module, and a data analysis and calibration module. The intelligent control module is connected to a feedforward receiving module and a drug output module; The human-computer interaction module is connected to a monitoring and data management module and a remote operation and maintenance and early warning module. The key parameter detection module is used to monitor pH, ORP, turbidity / suspended solids, and conductivity in nut processing wastewater. For organic matter, oil, and protein rich in nut processing wastewater, COD or TOC, UV254 (characterizing aromatic organic matter), and oil content should be monitored. Optional monitoring parameters include ammonia nitrogen and total phosphorus. The sampling unit module is used to construct sampling points for sampling work. The sampling points need to be set up at the place where the wastewater is evenly mixed and equipped with automatic samplers with timed and flow rate ratio sampling functions to sample and test the water source near the sampling point, and then test the water source in each process of wastewater treatment to see if the treatment is qualified. The data analysis and calibration module uses a built-in algorithm to identify and remove abnormal and drifting values ​​from sensors, and regularly (e.g., daily / weekly) compares and calibrates the data with manual laboratory testing data. Using historical data, it establishes a water quality fluctuation model to predict water quality change trends during peak production periods and processing of different nut varieties (e.g., high-oil walnuts and high-protein almonds), providing a basis for pre-dosing. The feedforward receiving module receives the influent flow rate, the real-time values ​​of raw water quality (COD, turbidity), and the quality of treated effluent from the water quality monitoring module. The reagent output module is used to calculate and add the basic dosage of the reagent in advance according to the preset model. Based on the quality of the treated effluent, the feedforward dosage is fine-tuned to form a closed loop, eliminate disturbances, and introduce parameter tuning of the model self-learning module to cope with the nonlinearity and time-varying nature of wastewater quality, with the optimization goal of achieving effluent compliance and the lowest cost in the future. The self-learning module enables the system to continuously optimize itself using operational data, adapting to water quality fluctuations and equipment aging. The system continuously records complete data pairs of influent water quality, chemical dosage, process parameters, and effluent water quality. Using a unified timestamp as a benchmark, the system automatically associates and stores massive amounts of time-series data from different sources, including: sensing data, execution data, operating condition data, environmental data, data cleaning and labeling, and advanced feature engineering. The model optimization program is initiated periodically or triggered, employing a model-free Bayesian optimization algorithm. This algorithm treats the system as a "black box," aiming for "the highest overall effluent quality compliance rate and the lowest chemical consumption per unit volume of water." By actively and incrementally trying new parameter combinations and observing system feedback, it efficiently finds the globally optimal or better parameter settings, automatically correcting coefficients in the feedforward model to ensure more accurate chemical dosage predictions achieve the desired treatment effect, minimizing chemical consumption over long-term operation. A comprehensive model evaluation is initiated periodically, allowing the system to... Using accumulated data, various alternative model structures are tested, and their predictive performance is evaluated through cross-validation. If the prediction error of the new model structure is found to be significantly lower than that of the existing model and passes safety verification, the model update process is triggered to replace the old model with a more accurate model, thereby achieving the "evolution" of the control strategy. This ensures that all model optimization and update operations are absolutely safe and prevents water quality from exceeding standards or equipment damage due to model errors. The system has a built-in simplified digital twin model of the process. Any new parameters or models from the optimization engine must first undergo long-term simulation testing in the digital twin sandbox to simulate various typical and extreme water intake conditions, verifying their control effect and stability. A series of insurmountable safety rules are preset, and the optimization algorithm must find the best under these hard constraints. After the model is updated, it can first be run in "shadow mode" in a parallel section or a specific time period (and its instructions are compared and analyzed with the actual control instructions). After confirming safety and effectiveness, it can then be smoothly switched to the main control loop. The monitoring and data management module provides a graphical user interface, displaying all process parameters, equipment status, trend curves, and alarm information in real time. It automatically generates daily and monthly reports, recording chemical dosage, power consumption, water quality compliance rate, and operating costs. Data can be stored and exported, supporting environmental compliance reports. The remote operation and maintenance and early warning module supports 4G / 5G / WIFI networking, enabling remote monitoring via mobile APP or web page, setting multi-level alarms, and timely notification of management personnel for special events; The database is used to store the data generated by the system.

2. The intelligent dosing system for treating nut processing wastewater based on online water quality monitoring according to claim 1, characterized in that: The sampling unit module is connected to a preprocessing submodule, which receives processing information from multiple physical preprocessing units and performs synchronous and continuous multi-dimensional water quality parameter measurements on the preprocessed water sample. It employs a multi-functional composite electrode probe, integrating pH, ORP, conductivity / TDS, and temperature sensors, and is responsible for data aggregation, primary processing, and quality assurance. Based on sliding window statistics and physical quantity change rate thresholds, it identifies and eliminates abnormal readings caused by bubbles and transient particle interference in real time. An exponentially weighted moving average algorithm is used to effectively suppress random noise while ensuring real-time performance. Theoretical correlations between parameters are used for rationality checks, and clearly contradictory data is marked and alarmed. The preprocessed high-quality real-time data stream is then uploaded to the core control module via industrial Ethernet or a wireless network.

3. The intelligent dosing system for treating nut processing wastewater based on online water quality monitoring according to claim 1, characterized in that: The model self-learning module is connected to a calibration and maintenance management sub-module. This sub-module ensures the long-term measurement accuracy of the sensing module and reduces maintenance burden. Based on sensor operating time, data drift, or a preset cycle, it automatically generates calibration task reminders and can connect to a laboratory information management system to synchronize calibration plans and records. It analyzes sensor self-test signals, cleaning frequency, and reading stability to provide early warnings of potential faults such as filter blockage, light source attenuation, and electrode aging. It performs real-time scoring and predicts the remaining lifespan of each online sensor, continuously tracking the deviation and trend between each sensor's reading and the laboratory reference method results. Using the CUSUM (cumulative sum) control chart algorithm, it sensitively detects whether a sensor has begun to slowly degrade in performance. It analyzes the sensor's built-in self-test signals as a direct indicator of health status, monitoring sensor cleaning time, backwash pressure, and calibration frequency. Abnormal behavior is often a precursor to blockage or wear. Combining this multi-dimensional data, it uses machine learning algorithms such as random survival forest to calculate the probability distribution of the remaining useful lifespan for each key sensor.

4. The intelligent dosing system for treating nut processing wastewater based on online water quality monitoring according to claim 1, characterized in that: The intelligent control module is connected to a dynamic water quality feature extraction module. It analyzes multi-parameter data streams in real time, identifies the "water quality mode" or "operating condition" of the current wastewater, and extracts the statistical features of each parameter within a sliding time window to form a "feature vector" describing the current water quality state. It presets multiple typical operating condition modes based on historical data learning, and quickly matches the real-time "feature vector" with the "operating condition mode library" to identify the most likely water quality type. This provides context for selecting the most suitable dosing control model. Based on the identified operating condition and real-time data, it dynamically calculates the optimal instantaneous dosing rate at each dosing point.

5. The intelligent dosing system for treating nut processing wastewater based on online water quality monitoring according to claim 1, characterized in that: The reagent dosing module is connected to a safety interlock protection module, which ensures that all control actions are absolutely safe and prevents process failure or equipment damage caused by misoperation. Soft start / stop logic, maximum / minimum stroke protection, and cumulative running time balancing logic are set for all metering pumps and valves. When an extreme anomaly is detected, the system automatically triggers a preset safety plan, such as switching the relevant dosing pump to a safe flow rate or stopping it completely, and issuing the highest level alarm.

6. The intelligent dosing system for nut processing wastewater treatment based on online water quality monitoring according to claim 1, characterized in that: The database connection has a backup module, which stores the data generated by the system in a third-party storage, while monitoring data changes in real time. When abnormal data changes occur, the third-party data backup is immediately blocked and staff are alerted. At the same time, the abnormal data is isolated and backed up again.

7. The intelligent dosing system for nut processing wastewater treatment based on online water quality monitoring according to claim 1, characterized in that: The remote operation and maintenance and early warning module is connected to a cloud-based collaborative maintenance and knowledge base sub-module. This sub-module leverages the cloud platform to enable remote expert support, maintenance experience accumulation, and global optimization. Maintenance engineers can remotely access the equipment's maintenance interface through a secure VPN channel to view detailed diagnostic data, execute advanced diagnostic procedures, and even perform remote calibration, significantly reducing on-site dispatch. The system systematically records each fault phenomenon, handling measures, and root cause in the cloud-based knowledge base. When similar fault symptoms appear at a new site, the system can automatically push historical cases and solutions to assist on-site personnel in quickly troubleshooting.

Images