A constructed wetland and denitrification filter coupled regenerative water deep denitrification system

Abstract

The present application relates to the technical field of reclaimed water reuse, and specifically discloses a reclaimed water advanced denitrification system coupled with an artificial wetland and a denitrification filter, comprising: a benchmark processing module, a monitoring and diagnosis module, a water quality monitoring module, and a differential response module, a target processing unit for reclaimed water advanced denitrification is built, a benchmark process is defined, a multi-source sensing monitoring network is built, the benchmark process is run, water quality data vectors and water temperature data are collected, a sensor health degree self-diagnosis thread is run, the health degree coefficient of the sensor is evaluated, and the failed sensor is marked; based on the water quality data vectors and the water temperature data, in combination with the health degree coefficient, a water quality anomaly monitoring thread is run, it is judged whether a real water quality impact load occurs, if so, type classification is performed, based on the type of the real water quality impact load, a response mechanism is triggered to execute, thereby realizing abnormal accurate identification based on credible data and differential intelligent control, and ensuring stable and efficient operation of the system.

Description

Technical Field

[0001] This invention relates to the field of reclaimed water reuse technology, specifically to a deep denitrification system for reclaimed water that couples an constructed wetland with a denitrifying filter. Background Technology

[0002] With the acceleration of urbanization and the increasing demand for wastewater resource utilization, reclaimed water, as a stable secondary water source, is widely used in landscape environmental replenishment and industrial reuse. However, the Class A standard of the "Discharge Standard of Pollutants for Urban Wastewater Treatment Plants" (GB18918-2002), which is generally implemented by urban wastewater treatment plants, has a total nitrogen limit of 15 mg / L, far exceeding the Class IV standard for surface water environmental quality (TN≤1.5 mg / L) and the eutrophication control threshold for lakes (TN<0.2 mg / L). Large amounts of nitrogen-rich reclaimed water are directly discharged into or replenish rivers and lakes, becoming a significant risk source for eutrophication and algal blooms. Taking Yangzhou's Slender West Lake as an example, water quality monitoring in 2026 showed that although other indicators of the wastewater treatment plant effluent reached Class II surface water quality levels, the total nitrogen concentration was still as high as 5.38 mg / L, resulting in the overall water quality of the lake area being Class V, with a transparency of only 20-30 cm, seriously affecting the landscape ecological function.

[0003] Currently, constructed wetlands and denitrification filters are the mainstream advanced treatment processes. In-depth research and practical monitoring have revealed the following core problems with existing technologies: First, process control relies on human experience, resulting in a delayed response to fluctuations in influent water quality and difficulty in accurately addressing different types of shock loads. Second, the reliability of sensor data is not guaranteed; a single sensor failure or drift can easily lead to misjudgments or missed diagnoses. Third, there is a lack of refined identification capabilities for anomalies; the distinction between global pollution and localized shocks, and between pulsed and continuous loads, is inadequate, leading to crude and insufficiently targeted control strategies. Furthermore, existing intelligent models are mostly designed based on emission standards, neglecting the specific requirements for reclaimed water reuse, and most remain at the level of simulated environmental verification, raising questions about their engineering applicability.

[0004] To address the aforementioned problems, this invention proposes a deep denitrification system for reclaimed water that couples an artificial wetland with a denitrification filter. Summary of the Invention

[0005] The purpose of this invention is to provide a deep denitrification system for reclaimed water that couples an artificial wetland with a denitrification filter, in order to solve the aforementioned background problems.

[0006] The objective of this invention can be achieved through the following technical solution: a deep denitrification system for reclaimed water coupled with an constructed wetland and a denitrification filter, comprising:

[0007] Benchmark treatment module: Construct the target treatment unit for deep denitrification of reclaimed water and define the benchmark process;

[0008] Monitoring and Diagnosis Module: Build a multi-source sensor monitoring network, run the baseline process and collect water quality data vectors and water temperature data, run the sensor health self-diagnosis thread, evaluate the sensor health coefficient, and mark failed sensors;

[0009] Water quality monitoring module: Based on water quality data vectors and water temperature data, combined with health coefficients, a water quality anomaly monitoring thread is run to determine whether a real water quality shock load has occurred, and if so, the type is classified.

[0010] Differential Response Module: Triggers the execution of a response mechanism based on the type of actual water quality shock load;

[0011] Furthermore, the baseline process includes:

[0012] The reclaimed water sequentially enters the pretreatment subunit, denitrification filter, constructed wetland and ecological pond;

[0013] The pretreatment subunit is equipped with a bar screen channel and a regulating tank. After the reclaimed water passes through the bar screen channel to intercept suspended solids, it flows by gravity into the regulating tank. After being homogenized by a submersible mixer, it is quantitatively transported to the denitrification filter by a booster pump station. The denitrification filter adopts an upflow mode and is equipped with a composite buffer packing layer and a composite filter media layer in sequence. The composite filter media layer is composed of a mixture of sulfur-based autotrophic packing and slow-release carbon source heterotrophic packing. The effluent from the denitrification filter is partially returned to the inlet of the filter according to a preset reflux ratio. The constructed wetland is composed of a vertical subsurface flow wetland and a surface flow wetland connected in series. The vertical subsurface flow wetland is filled with zeolite packing and planted with emergent plants, while the surface flow wetland is planted with emergent plants and submerged plants. The effluent from the constructed wetland enters an ecological pond, where cash crops are planted and filter-feeding fish are raised. After being collected by a collection well, the effluent meets the standards.

[0014] Furthermore, the marking method for failed sensors is as follows:

[0015] Establish a multi-source sensor monitoring network, including a core water quality monitoring unit and an environmental monitoring unit;

[0016] The core water quality monitoring unit is deployed at the outlet of the equalization tank and integrates a variety of online monitoring devices. For the key online monitoring devices, at least two sensors with different measurement principles are configured to form a heterogeneous redundant combination. The environmental monitoring unit includes a water temperature sensor deployed inside the denitrification filter.

[0017] The programmable logic controller collects real-time measurement values ​​from each sensor, integrates them to obtain water quality data vector and water temperature data, runs a sensor health self-diagnosis thread, calculates the health coefficient of each sensor, the health coefficient is obtained by weighted fusion of the sensor's consistency index, reliability index and stability index, compares the health coefficient with the preset allowable threshold, and if it is lower than the preset allowable threshold, the sensor is determined to be a failed sensor.

[0018] Furthermore, the consistency indicators are obtained as follows:

[0019] For sensors configured with heterogeneous redundancy, the real-time measurement values ​​of each sensor in the heterogeneous redundancy combination are read synchronously at the same time. The average deviation is obtained by averaging the relative deviation between the real-time measurement value of the sensor and the real-time measurement values ​​of different sensors in the heterogeneous redundancy combination. The consistency index is obtained by data processing based on the average deviation.

[0020] For sensors without heterogeneous redundancy combinations, a lightweight LSTM time series prediction model is established for the sensor to predict the expected value at the current time and the data is processed to obtain the prediction residual. The historical residual sequence is integrated and statistically analyzed, and the consistency index is calculated based on the degree of deviation of the current prediction residual from the normal fluctuation range.

[0021] Furthermore, the reliability and stability indicators are obtained as follows:

[0022] The reliability index is calculated based on the dynamic historical operation file established for the sensor, which includes the cumulative number of failures, the time since the last calibration, the preset standard calibration cycle duration, and the equipment operating time; the stability index is obtained by normalizing the coefficient of variation of the sensor's real-time measurement values ​​at various times within a preset short period of time.

[0023] Furthermore, the method for determining whether a real water quality shock load has occurred is as follows:

[0024] The water quality anomaly monitoring thread integrates a preliminary anomaly capture unit and a comprehensive judgment unit;

[0025] After filtering out failed sensor data based on the health coefficient, a high-confidence water quality data vector is obtained. The actual fusion value of each water quality indicator is obtained with a preset sampling period. For water quality indicators with heterogeneous redundancy combinations, the health coefficient is used as the weight to perform weighted fusion of the corresponding heterogeneous redundant data groups in the high-confidence water quality data vector to obtain the actual fusion value. Otherwise, the corresponding water quality indicator data in the high-confidence water quality data vector is directly used as the actual fusion value.

[0026] The initial anomaly capture unit is executed to obtain an initial water quality anomaly marker. The comprehensive judgment unit is activated to integrate the actual fusion value to obtain a multi-dimensional water quality feature vector. The Mahalanobis distance between the multi-dimensional water quality feature vector and the preset historical normal operating condition feature vector is calculated. If it exceeds the preset dynamic Mahalanobis distance threshold, it is determined that a real water quality shock load has occurred.

[0027] Furthermore, the preliminary markers of water quality anomalies are obtained as follows:

[0028] Set the capture flag value and assign it an initial value of 0. Based on the high-confidence water quality data vector, query the pre-built water temperature-water quality correlation model to obtain the normal fluctuation range of each water quality indicator corresponding to the current water temperature data. Compare the actual fusion value of each water quality indicator with the corresponding normal fluctuation range item by item. When the actual fusion value of any water quality indicator in a preset number of consecutive sampling periods exceeds the corresponding normal fluctuation range and the capture flag value is 0, set the capture flag value to 1 and record the preliminary water quality anomaly. The preliminary water quality anomaly includes the water quality indicator exceeding the standard, the amplitude of the exceedance, the time of occurrence, and the duration.

[0029] Furthermore, the pre-construction method for the water temperature-water quality correlation model is as follows:

[0030] Water temperature data under the baseline process and the actual fusion values ​​of various water quality indicators within the historical standard analysis period with the current time as the endpoint and the preset standard analysis duration are collected. The water temperature data is divided into segments according to the preset length water temperature intervals. The 5th percentile of the actual fusion value of each water quality indicator in each water temperature interval is used as the lower limit of normal fluctuation and the 95th percentile is used as the upper limit of normal fluctuation. The cubic spline interpolation method is used to fit continuous water temperature-fluctuation upper limit value curve and water temperature-fluctuation lower limit value curve based on each water temperature interval, and the water temperature-water quality correlation model is integrated.

[0031] Furthermore, the method of categorization is as follows:

[0032] The anomaly monitoring thread integrates a preliminary anomaly capture unit and a comprehensive judgment unit. The comprehensive judgment unit includes a multi-parameter coupled verification subunit and a timing anomaly capture subunit.

[0033] When a real water quality shock load is determined to occur, the grey correlation degree between various water quality indicators in the past short period of time is calculated in parallel. If the grey correlation degree is lower than the preset correlation threshold, it is determined to be a local shock anomaly; otherwise, it is determined to be a global anomaly. If it is a local shock anomaly, it is further identified as a pulse shock load or a continuous abnormal load through the time-series anomaly capture subunit.

[0034] Furthermore, the method for identifying pulse-type impact loads and continuous abnormal loads is as follows:

[0035] For localized impact anomalies, the actual fused values ​​of the water quality indicators exceeding the standard for L sampling periods prior to the current moment are read to form the indicator time series within the sliding window. This sequence is input into a pre-constructed time series prediction model to predict the expected value of the indicator at the current moment. The absolute value of the difference between the actual fused value and the expected value of the indicator is calculated as the indicator residual. If the indicator residual exceeds the preset indicator residual threshold, the current moment is recorded as the time of exceeding the limit. The difference between the indicator residual and the preset indicator residual threshold is calculated to obtain the value of exceeding the limit. Data processing is performed to obtain the cumulative impact energy and the duration of exceeding the limit within the sliding window. If the cumulative impact energy exceeds the preset energy threshold and the duration of exceeding the limit does not exceed the preset short duration upper limit, it is determined to be a pulse-type impact load. If it exceeds the preset short duration upper limit, it is determined to be a continuous abnormal load.

[0036] The beneficial effects of this invention are as follows:

[0037] 1. This invention constructs a multi-level barrier-type deep nitrogen removal process chain by coupling denitrification filters with constructed wetlands and ecological ponds. The denitrification filters use sulfur-based autotrophic and slow-release carbon source heterotrophic composite packing materials, which efficiently remove nitrogen under the synergistic effect of autotrophic and heterotrophic bacteria. The series of vertical subsurface flow and surface flow wetlands utilize plant root oxygen secretion and packing material adsorption to further remove residual nitrogen. Finally, the ecological pond plants absorb the nitrogen and filter-feeding fish feed to stabilize the water quality. This fully leverages the synergistic advantages of the high-efficiency treatment of the filter and the ecological buffering of the wetland, significantly improving the nitrogen removal efficiency of reclaimed water and the system's resistance to shock loads, while creating good landscape ecological benefits.

[0038] 2. This invention integrates a multi-source sensor monitoring network and an intelligent control system. Through the combination of heterogeneous redundant sensors and a health self-diagnosis thread, it effectively filters out faulty data, ensuring data reliability from the source. Based on a three-level anomaly diagnosis system with a dynamic baseline of water temperature and water quality, it can accurately identify global, pulsed, and continuous water quality shock loads and trigger differentiated response mechanisms such as reflux ratio adjustment, over-discharge, and early warning push. This intelligent management and control method realizes the transformation from passive response to proactive early warning, significantly reducing reagent consumption and operation and maintenance costs, and ensuring the long-term stable and compliant operation of the reclaimed water treatment system. Attached Figure Description

[0039] The invention will now be further described with reference to the accompanying drawings.

[0040] Figure 1 This is a modular architecture diagram of a deep denitrification system for reclaimed water that couples an artificial wetland with a denitrification filter, as described in an embodiment of the present invention.

[0041] Figure 2 This is a flowchart illustrating the specific steps of a deep denitrification system for reclaimed water that couples an artificial wetland with a denitrifying filter, as described in an embodiment of the present invention. Detailed Implementation

[0042] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0043] Example 1

[0044] Please see Figure 1 As shown in the embodiment of the present invention, a deep denitrification system for reclaimed water coupled with an constructed wetland and a denitrifying filter aims to solve the problems of existing deep denitrification methods for reclaimed water relying on manual experience, having delayed control, and suffering from misjudgments due to poor sensor data reliability, thus failing to accurately cope with different types of water quality shock loads. This is achieved by constructing a target treatment unit consisting of a pretreatment subunit, a denitrifying filter, an constructed wetland, and an ecological pond; building a multi-source sensor monitoring network and running a sensor health self-diagnosis thread; filtering out invalid data; and combining dynamic water temperature baseline, Mahalanobis distance, and grey relational analysis to verify the operation of a water quality anomaly monitoring thread. This accurately identifies global anomalies, pulsed shock loads, and persistent abnormal loads, triggering corresponding response mechanisms. This achieves accurate anomaly identification and differentiated intelligent control based on reliable data, ensuring stable and efficient system operation. Specifically, the system includes the following modules:

[0045] Benchmark treatment module: Construct the target treatment unit for deep denitrification of reclaimed water and define the benchmark process;

[0046] The target treatment unit includes a pretreatment subunit, a booster pump station, a denitrification filter, an artificial wetland, and an ecological pond. The baseline process is a baseline process for deep denitrification of reclaimed water based on the target treatment unit.

[0047] Specifically, in the baseline process, reclaimed water first enters the pretreatment subunit, which is equipped with a bar screen channel and a regulating tank. The spacing of the bar screen in the bar screen channel is set. After the reclaimed water is treated by the bar screen channel to intercept large-diameter suspended solids, it flows by gravity into the regulating tank. The regulating tank has a built-in submersible agitator to homogenize the water quality and buffer the fluctuation of the incoming water volume, resulting in the effluent from the regulating tank.

[0048] The effluent from the equalization tank is quantitatively pumped to the denitrification filter by the booster pump station as the filter influent. The booster pump station is equipped with a frequency converter to regulate the influent flow rate. The denitrification filter adopts an upflow mode. After the filter influent is evenly distributed across the cross-section of the denitrification filter by the water distribution system, it flows sequentially through the composite buffer packing layer and the composite filter media layer. The composite buffer packing layer is composed of modified zeolite or activated carbon and polyurethane sponge, which adsorbs ammonia nitrogen and intercepts suspended solids in the filter influent and provides an attachment carrier for microorganisms. The composite filter media layer is located downstream of the composite buffer packing layer and is obtained by mixing sulfur-based autotrophic packing and slow-release carbon source heterotrophic packing at a preset volume ratio. Under the synergistic action of autotrophic and heterotrophic denitrifying bacteria, nitrate nitrogen and ammonia nitrogen are converted into nitrogen gas to obtain the denitrification filter effluent. The hydraulic retention time (HRT) and reflux ratio of the denitrification filter are set. According to the reflux ratio, part of the denitrification filter effluent is returned as the filter influent to dilute the filter influent load.

[0049] The effluent from the denitrification filter enters the constructed wetland as wetland influent. The total hydraulic retention time of the constructed wetland is set. The constructed wetland is composed of a vertical subsurface flow wetland and a surface flow wetland connected in series. The vertical subsurface flow wetland is filled with zeolite packing material and emergent plants such as reeds and calamus are planted on the surface. The ammonia nitrogen in the wetland influent is further removed by the oxygen secretion of plant roots and the adsorption performance of zeolite packing material before entering the surface flow wetland. The surface flow wetland is planted with emergent plants and submerged plants. Submerged plants such as foxtail grass and Vallisneria natans are planted. The residual nitrogen is removed through plant absorption and microbial metabolism and the water transparency is improved to obtain the constructed wetland effluent.

[0050] The effluent from the artificial wetland eventually enters the ecological pond, where cash crops are planted and filter-feeding fish are raised. Cash crops include lotus root and water chestnut, and filter-feeding fish include silver carp and bighead carp. Through the absorption of algae by plants and the feeding of algae by filter-feeding fish, the water quality is stabilized and a landscape is created. The effluent that meets the standards is collected by a collection well set at the end of the ecological pond and reused.

[0051] It should be noted that the role of this module is to construct a biological-environmental coupled deep treatment process chain of denitrification filter, constructed wetland and ecological pond, clarify the functional positioning and hydraulic connection parameters of each unit, form a multi-level barrier to achieve deep denitrification and improve water transparency;

[0052] Monitoring and Diagnosis Module: Build a multi-source sensor monitoring network, run the baseline process and collect water quality data vectors and water temperature data, run the sensor health self-diagnosis thread, evaluate the sensor health coefficient, and mark failed sensors;

[0053] Specifically, various online monitoring devices are deployed in the target processing unit according to the preset spatial topology to form a multi-source sensor monitoring network, including a core water quality monitoring unit and an environmental monitoring unit;

[0054] The core water quality monitoring unit is deployed at the outlet of the equalization tank included in the target treatment unit. The online monitoring equipment integrated in the core water quality monitoring unit includes: pH meter, online ammonia nitrogen analyzer, online total nitrogen analyzer, turbidity meter, online COD analyzer, and online biotoxicity monitoring instrument based on luminescent bacteria. It collects water quality index data for each water quality indicator, including: pH value of reclaimed water monitored by pH meter, ammonia nitrogen concentration and total nitrogen concentration measured in real time by online ammonia nitrogen analyzer and online total nitrogen analyzer, chemical oxygen demand measured by online COD analyzer, and luminescence inhibition rate of luminescent bacteria detected by online biotoxicity monitoring instrument based on luminescent bacteria. For the key online monitoring equipment preset in the core water quality monitoring unit, at least two sensors with different measurement principles are configured to form a heterogeneous redundant combination. The water quality index data collected by each heterogeneous redundant combination at the same time is marked as a heterogeneous redundant data group. All water quality index data collected at the same time are integrated to obtain a water quality data vector.

[0055] For example, a heterogeneous redundancy combination, such as an online ammonia nitrogen analyzer, can simultaneously deploy a sensor based on the ion-selective electrode method and a sensor based on the salicylic acid spectrophotometry method. It should be noted that the power supply and communication lines of the sensors included in the heterogeneous redundancy combination also adopt independent redundancy design to ensure that the normal operation of the other sensor is not affected when one sensor fails due to power supply or communication interruption.

[0056] It should be noted that the purpose of configuring heterogeneous redundancy combinations only for the preset key online monitoring equipment is to ensure the accuracy of data acquisition from the key online monitoring equipment while controlling costs.

[0057] The online monitoring equipment of the environmental monitoring unit includes water temperature sensors installed at preset points inside the denitrification filter. The water temperature sensors are installed in the middle of the composite filter media layer to collect water temperature data.

[0058] In a multi-source sensor monitoring network, online monitoring devices all collect corresponding data based on internal sensors through programmable logic controllers and transmit the data in real time to the intelligent control system via industrial Ethernet, forming a water quality data vector with a unified spatiotemporal reference.

[0059] The intelligent control system continuously receives water quality data vectors and simultaneously starts the sensor health self-diagnosis thread to evaluate the health coefficient of all sensors built into the online monitoring equipment.

[0060] Specifically, the health coefficient H_i(t) of each sensor is calculated, where i represents the sensor number and t represents time. The health coefficient is obtained by weighted fusion of the sensor's consistency index C_i(t), reliability index R_i(t), and stability index S_i(t) based on preset weights.

[0061] The consistency index C_i(t) is obtained as follows:

[0062] For heterogeneous redundant combinations, the real-time measurement values ​​V_i(t) of each sensor in the heterogeneous redundant combination are synchronously read at the same time at a preset frequency and integrated into a redundant observation set. For the sensor with the number i, the relative deviation δ_ij(t) between the real-time measurement value of sensor i and the other real-time measurement values ​​in the redundant observation set is calculated as δ_ij(t) = |V_i(t) - V_j(t)| / max(V_i(t), V_j(t), ε), where j represents the sensor number in the heterogeneous redundant combination that is different from i, and ε is a very small positive number to avoid division by zero. The average deviation is obtained by taking the mean of all relative deviations calculated for sensor i in the redundant observation set. The real-time consistency index C_i(t) = 1 - [average deviation / preset deviation tolerance upper limit] is calculated and normalized to the interval [0,1].

[0063] For sensors without heterogeneous redundancy combinations, the consistency index C_i(t) employs a time-series-based prediction bias detection method: a lightweight LSTM time-series prediction model is established for the sensor. Based on the data collected by the sensor within a historical period of a preset duration ending at the current time, the expected value V_i,pred(t) at the current time is predicted. The prediction residual ε_i(t) = |V_i(t) - V_i,pred(t)| is calculated, and the operational data within the time period ending at the previous time and a preset statistical duration is obtained. The predicted residual sequence under the baseline process is calculated, and the mean μ and standard deviation σ of the predicted residual sequence are statistically analyzed. The mean ± n times the standard deviation is taken as the normal fluctuation range. The degree to which the current predicted residual ε_i(t) deviates from the normal fluctuation range is calculated. If ε_i(t) is within the normal fluctuation range, the consistency index C_i(t) = 1. If ε_i(t) exceeds the normal fluctuation range, then C_i(t) = max(0, 1-(|ε_i(t)-μ| / σ-n) / m), where m is the normalization coefficient.

[0064] A dynamic historical operation file is established for each sensor, including the cumulative number of failures, the time since the last calibration, the preset standard calibration cycle duration, and the device runtime. The reliability index R_i(t) is calculated based on the dynamic historical operation file of each sensor: R_i(t) = R_base × (1 - f_fail) × max(0, 1 - D_cal / D_max), where R_base is the preset base reliability coefficient, f_fail is the historical failure frequency, f_fail = min(1, λ / λ_max), where λ represents the ratio of the cumulative number of failures to the device runtime, λ_max represents the preset maximum allowable failure rate, and when λ exceeds λ_max, f_fail is 1, indicating complete loss of reliability. D_cal is the calibration cycle deviation, which is obtained by calculating the difference between the time since the last calibration and the preset standard calibration cycle duration and then comparing it with the preset standard calibration cycle duration. The minimum value of the calibration cycle deviation is 0, and D_max represents the preset maximum allowable calibration cycle deviation.

[0065] The stability index S_i(t) is obtained by normalizing the coefficient of variation of the real-time measurement values ​​of the sensor at each moment within a preset short time period with the current moment as the endpoint;

[0066] For all sensors, the calculated health coefficient H_i(t) is compared with the preset allowable threshold. If it is lower than the preset allowable threshold, the sensor is determined to be a failed sensor and a failure alarm to be calibrated is sent to the cloud operation and maintenance platform.

[0067] It should be noted that the function of this module is to collect data in real time through the core water quality monitoring unit and the environmental monitoring unit. The heterogeneous redundancy combination configuration ensures the accuracy of key indicator measurements. The sensor health coefficient assessment filters out data from failed sensors to avoid misjudgments caused by data distortion. Sensor health self-diagnosis is used as a prerequisite for data governance. The sensor status is assessed by combining heterogeneous redundancy comparison with LSTM time series prediction residuals, ensuring data reliability from the source.

[0068] Water quality monitoring module: Based on water quality data vectors and water temperature data, combined with health coefficients, a water quality anomaly monitoring thread is run to determine whether a real water quality shock load has occurred, and if so, the type is classified.

[0069] Specifically, the intelligent control system acquires water quality data vectors and, based on the health coefficient, filters out the water quality index data collected by the failed sensors in the water quality data vectors, obtaining the filtered high-confidence water quality data vectors as input, and starts the water quality anomaly monitoring thread to identify the abnormal impact load of the reclaimed water inlet at the outlet of the equalization tank. The water quality anomaly monitoring thread integrates a preliminary anomaly capture unit and a comprehensive judgment unit.

[0070] The preliminary anomaly capture unit performs preliminary capture of water quality anomalies based on the high-confidence water quality data vector. It obtains the actual fusion value of each water quality indicator in the high-confidence water quality data vector at a preset sampling period. For water quality indicators with configured heterogeneous redundancy combinations, the health coefficient is used as the weight to perform weighted fusion of the corresponding heterogeneous redundant data groups in the high-confidence water quality data vector to obtain the actual fusion value. If not configured, the corresponding water quality indicator data in the high-confidence water quality data vector is directly used as the actual fusion value.

[0071] An example table of actual fusion value data is shown below:

[0072] Table 1: Initial Water Quality Indicators of Effluent from Tangwang Wastewater Treatment Plant;

[0073] Water quality indicators Actual fusion value GB18918-2002 Class A Standard Limits Surface water environmental quality standards (GB3838-2002) - Worst category pH value 7.9 6～9 Class I Chemical oxygen demand (COD) 15mg / L ≤50mg / L Class II Ammonia nitrogen (as N) 0.138 mg / L ≤5 (8) mg / L Class I Total nitrogen (as N) 5.38 mg / L ≤15mg / L Class V (inferior)

[0074] Define a capture flag value and initialize it to 0;

[0075] A pre-constructed water temperature-water quality correlation model is used. Water temperature data under the baseline process and the actual fusion values ​​of various water quality indicators are collected within a historical standard analysis period with the current time as the endpoint and a preset standard analysis duration. The water temperature data is segmented according to a preset length of water temperature interval. The 5th percentile of the actual fusion value of each water quality indicator within each water temperature interval is used as the lower limit of normal fluctuation, and the 95th percentile is used as the upper limit of normal fluctuation. The cubic spline interpolation method is used to fit a continuous water temperature-fluctuation upper limit value curve and a water temperature-fluctuation lower limit value curve for each water temperature interval. The water temperature-water quality correlation model is then integrated. Based on the water temperature data at the current time, the water temperature-water quality correlation model is queried to obtain the lower limit and upper limit of normal fluctuation for each water quality indicator at the current time, i.e., the normal fluctuation interval of the corresponding water quality indicator.

[0076] It should be noted that the water temperature-water quality correlation model is updated using an online incremental learning method;

[0077] The actual fusion value of each water quality indicator is compared with the normal fluctuation range of the corresponding water quality indicator. When the actual fusion value of any water quality indicator exceeds the normal fluctuation range of the corresponding water quality indicator in a preset number of sampling periods and the capture flag value is 0, the capture flag value is changed to 1. The period when the capture flag value is 1 is recorded as a preliminary water quality abnormality event.

[0078] If the capture flag value is 1, record the preliminary water quality anomaly marker and start the comprehensive judgment unit to determine whether a real water quality shock load has occurred. The preliminary water quality anomaly marker includes the water quality index exceeding the standard, the exceeding amplitude, the time of occurrence, and the duration. Among them, the exceeding amplitude is the amplitude of the actual fusion value exceeding the normal fluctuation range of the water quality index, the time of occurrence is the current time, and the duration is the duration of the current corresponding preliminary water quality anomaly event.

[0079] If it is determined that no real water quality shock load has occurred, and the actual fusion values ​​of all water quality indicators in the preset number of sampling periods do not exceed the normal fluctuation range of the corresponding water quality indicators, then the capture flag value will be set to 0.

[0080] If the capture flag value is 0, the water quality is considered normal, the baseline process is maintained, and monitoring continues.

[0081] For the initial marking of water quality anomalies, a comprehensive judgment unit is activated, which includes a multi-parameter coupling verification subunit and a time-series anomaly capture subunit.

[0082] Specifically, the actual fused values ​​of each water quality indicator at the current moment are integrated to obtain a multi-dimensional water quality feature vector, which is input into the multi-parameter coupling verification subunit. The Mahalanobis distance between the current multi-dimensional water quality feature vector and the historical normal operating condition feature vector is calculated. The historical normal operating condition feature vector is the average of the actual fused values ​​between the lower limit and the upper limit of the normal fluctuation of each water quality indicator within the current water temperature range under the baseline process during the historical standard analysis period. The historical normal operating condition feature vector is obtained by integrating the values. The grey relational degree between each water quality indicator in the past short period of time is calculated in parallel. If the Mahalanobis distance exceeds the dynamic Mahalanobis distance threshold, it is determined that a real water quality shock load has occurred. Through multi-parameter coupling verification, the real water quality shock load is divided into global anomaly or local shock anomaly according to the grey relational degree. Specifically, if the grey relational degree is lower than the preset relational threshold, it is determined to be a local shock anomaly; otherwise, it is determined to be a global anomaly.

[0083] Among them, the dynamic Mahalanobis distance threshold is the 95th percentile of all Mahalanobis distances under the benchmark process in the past short period of time.

[0084] It should be noted that Mahalanobis distance is used to measure the overall deviation of the current water quality state from the baseline process, while grey relational analysis is used to analyze the synchronicity of the changing trends among various water quality indicators. The dynamic Mahalanobis distance threshold represents the maximum reasonable value of Mahalanobis distance under normal fluctuations. If the current Mahalanobis distance exceeds the dynamic Mahalanobis distance threshold, it means that the overall deviation of the current water quality has exceeded the normal fluctuation range. Further comparison with grey relational analysis, which reflects the synchronicity of the changing trends of various water quality indicators, shows that if the grey relational analysis is high, it indicates good synchronicity, which may be due to synchronous global pollution. If the grey relational analysis is lower than the preset relational threshold, it indicates that there are uncoordinated changes among different water quality indicators, which is more in line with the characteristics of specific pollutants entering the system and local pollutant impacts.

[0085] If the multi-parameter coupling verification fails, it is determined that no real water quality shock load has occurred.

[0086] If the multi-parameter coupling verification is successful and the actual water quality shock load is a local shock anomaly, the time-series anomaly capture subunit is executed. The water quality indicators exceeding the standard in the preliminary water quality anomaly label are read, and the actual fusion values ​​of the corresponding water quality indicators in the previous L sampling periods at the current time are read to form the indicator time-series sequence within the sliding window. The window length of the sliding window is L, corresponding to L sampling periods, where L is a preset positive integer. A time-series prediction model based on a Long Short-Term Memory (LSTM) network is constructed. The network structure includes an input layer, two hidden layers, and an output layer. Each hidden layer contains 64 memory units. The time-series sequence of the actual fusion values ​​of the corresponding water quality indicators under the baseline process in the historical standard analysis period is used as the training set for offline training and daily incremental learning. The indicator time-series sequence is input into the trained time-series prediction model to predict the expected value of the indicator at the current time.

[0087] The absolute value of the difference between the actual fusion value and the expected value of the corresponding index is calculated as the index residual. If the index residual exceeds the preset index residual threshold, the current moment is recorded as the over-limit moment, and the difference between the index residual and the preset index residual threshold is calculated to obtain the over-limit value. The cumulative impact energy within the sliding window is defined as the sum of the products of the over-limit values ​​of all over-limit moments within the sliding window and the sampling period duration. The number of consecutive over-limit moments within the sliding window is counted and multiplied with the sampling period duration to obtain the over-limit duration. The preset energy threshold and short duration upper limit are also defined.

[0088] If the cumulative impact energy does not exceed the preset energy threshold, it is determined that a misjudgment of the real water quality impact load based on random disturbance has occurred, that is, no real water quality impact load has occurred.

[0089] If the cumulative impact energy exceeds the preset energy threshold and the duration of the excess does not exceed the short duration limit, the local impact anomaly is determined to be a pulse-type impact load, and the cumulative impact energy is recorded.

[0090] If the cumulative impact energy exceeds the preset energy threshold and the duration of the excess exceeds the short duration limit, it is determined that a continuous abnormal load has occurred.

[0091] It should be noted that the types of actual water quality shock loads include global anomalies and local shock anomalies, and local shock anomalies include pulse-type shock loads and continuous abnormal loads.

[0092] It should be noted that the function of this module is to dynamically establish a normal fluctuation range based on the water temperature-water quality correlation model after screening out the failed data according to the health coefficient. The Mahalanobis distance and grey relational degree are coupled to verify and distinguish between global anomalies and local impact anomalies. Local impact anomalies are further identified as pulse-type or continuous anomalies by calculating the cumulative impact energy through LSTM time series prediction residuals, so as to achieve accurate diagnosis and classification of anomalies. A three-level anomaly diagnosis system of dynamic baseline initial capture, multi-dimensional coupling verification, and time series anomaly capture is constructed to accurately distinguish the nature of anomalies.

[0093] Differential Response Module: Triggers the execution of a response mechanism based on the type of actual water quality shock load;

[0094] The intelligent control system continuously runs the water quality anomaly monitoring thread. When the comprehensive judgment unit outputs the type judgment result of the actual water quality shock load, the response mechanism is immediately triggered.

[0095] Specifically, the intelligent control system has a built-in response strategy knowledge base, which stores the mapping relationship between the actual water quality shock load types and response strategies. The response strategies are encapsulated in the form of a rule engine. The input of the rule engine is the actual water quality shock load type and its characteristic parameters. The characteristic parameters include the water quality index exceeding the standard, the magnitude of the exceedance, the cumulative shock energy, and the duration of the exceedance. The output of the rule engine is a specific set of execution instructions. The instruction set is sent to the programmable logic controllers of the target treatment units, including the booster pump station, denitrification filter and constructed wetland, as well as the external carbon source dosing system and emergency overrun valve group.

[0096] When the judgment result is a global anomaly, multiple water quality indicators characterizing the reclaimed water at the outlet of the equalization tank simultaneously show an overall deviation, and the intelligent control system executes a global response command:

[0097] The flow rate of the variable frequency pumps at the booster pumping station will be reduced by a preset ratio to extend the hydraulic retention time of reclaimed water in subsequent units, thereby enhancing the system's buffering capacity. The corresponding enhanced treatment mode will be matched according to the water quality index with the largest exceedance value. The reflux ratio of the denitrification filter will be adjusted to increase the reflux ratio from the current value by a preset amount to enhance the synergistic effect of autotrophic and heterotrophic denitrifying bacteria in the filter and dilute the concentration of pollutants entering the filter. If the duration of the global anomaly exceeds the preset duration, the intelligent control system will push an orange warning message to the cloud operation and maintenance platform. The warning message includes the anomaly type, the water quality index exceeding the standard, the exceedance value, and the adjustment measures already taken. It is recommended that operation and maintenance personnel check the upstream water source.

[0098] When the judgment result is a local shock anomaly and it is further identified as a pulse-type shock load, indicating that a single or a few water quality indicators are experiencing short-term high-intensity shocks, the intelligent control system executes a pulse response command:

[0099] Extract the water quality indicators and exceedance values ​​from the preliminary markers of water quality anomalies. Determine the response intensity based on the range of exceedance values. If the exceedance value is in the preset mild range, only the reflux ratio adjustment is activated to instantly increase the reflux ratio of the denitrification filter to the maximum allowable value. The reflux water is used to quickly dilute the shock load. At the same time, the event is recorded but no alarm is triggered.

[0100] If the exceedance value is within the preset severe range, in addition to increasing the reflux ratio, the emergency bypass valve group will be activated simultaneously to directly introduce part of the effluent from the equalization tank into the constructed wetland, bypassing the denitrification filter. The bypass flow rate is calculated by the cumulative impact energy of the pulse impact load using an energy-flow conversion model. The energy-flow conversion model is obtained by fitting the correspondence between the cumulative impact energy and the required bypass flow rate based on historical operating data. The bypass duration is equal to the over-limit duration. After the bypass ends, the bypass valve group will be automatically closed, and the baseline process will be restored.

[0101] It should be noted that historical operating data refers to the time-series data archive collected and stored by the multi-source sensor monitoring network when the target processing unit is running under the baseline process. This data includes the actual fused values ​​of various water quality indicators, water temperature data, and sensor health coefficients.

[0102] During the execution of the pulse response command, the intelligent control system continuously monitors the actual fusion value of the subsequent sampling cycle. If the water quality index exceeding the standard returns to the normal fluctuation range within the preset buffer time, the reflux ratio is gradually restored to the baseline process. If the same type of pulse impact is captured again, the pulse frequency is accumulated. When the accumulated pulse frequency exceeds the preset daily tolerance limit, a pulse frequency warning is pushed to the cloud operation and maintenance platform.

[0103] When the judgment result is a local shock anomaly, and it is further identified as a continuous abnormal load, indicating that one or a few water quality indicators exceed the normal fluctuation range for a long time, the intelligent control system executes a continuous response command:

[0104] The intelligent control system retrieves the time series of water quality indicators corresponding to the continuous abnormal load within the sliding window, inputs the time series prediction model to extrapolate the expected value of the indicators for the future preset duration, and performs trend analysis. If the trend analysis shows that the continuous abnormal load will continue to rise or remain at a high level, it is determined that the current processing capacity is insufficient to cope with it, and an emergency warning reminder is immediately pushed to the cloud operation and maintenance platform to notify the operation and maintenance personnel to investigate and handle the situation.

[0105] It should be noted that during the execution of the response mechanism, the intelligent control system continuously acquires the actual fusion values ​​of each water quality indicator at a preset sampling period and compares them item by item with the normal fluctuation range of the corresponding water quality indicator. When the actual fusion values ​​of the water quality indicators that triggered the response return to the normal fluctuation range within a preset number of consecutive sampling periods, and the cumulative impact energy decays to below the preset energy threshold, the intelligent control system automatically triggers the regression benchmark process instruction. After determining the regression benchmark process, the captured flag value is set to 0 and monitoring continues.

[0106] The technical solution of this invention is as follows: A target treatment unit for deep denitrification of reclaimed water is constructed, a baseline process is defined, a multi-source sensor monitoring network is constructed, the baseline process is run and water quality data vectors and water temperature data are collected, a sensor health self-diagnosis thread is run, the health coefficient of the sensors is evaluated, failed sensors are marked, and a water quality anomaly monitoring thread is run based on the water quality data vectors and water temperature data, combined with the health coefficient, to determine whether a real water quality shock load has occurred. If it has occurred, the type is classified, and a response mechanism is triggered based on the type of the real water quality shock load.

[0107] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A deep denitrification system for reclaimed water coupled with an constructed wetland and a denitrifying filter, characterized in that: Includes the following modules: Benchmark treatment module: Construct the target treatment unit for deep denitrification of reclaimed water and define the benchmark process; Monitoring and Diagnosis Module: Build a multi-source sensor monitoring network, run the baseline process and collect water quality data vectors and water temperature data, run the sensor health self-diagnosis thread, evaluate the sensor health coefficient, and mark failed sensors; Water quality monitoring module: Based on water quality data vectors and water temperature data, combined with health coefficients, a water quality anomaly monitoring thread is run to determine whether a real water quality shock load has occurred, and if so, the type is classified. Differential Response Module: Triggers the execution of the response mechanism based on the type of actual water quality shock load.

2. The deep denitrification system for reclaimed water coupled with an constructed wetland and a denitrifying filter according to claim 1, characterized in that: The benchmark process includes: The reclaimed water sequentially enters the pretreatment subunit, denitrification filter, constructed wetland and ecological pond; The pretreatment subunit is equipped with a bar screen channel and a regulating tank. After the reclaimed water passes through the bar screen channel to intercept suspended solids, it flows by gravity into the regulating tank. After being homogenized by a submersible mixer, it is quantitatively delivered to the denitrification filter by a booster pump station. The denitrification filter adopts an upflow mode and is equipped with a composite buffer packing layer and a composite filter media layer in sequence. The composite filter media layer is composed of a mixture of sulfur-based autotrophic packing and slow-release carbon source heterotrophic packing. The effluent from the denitrification filter is partially returned to the inlet of the filter according to a preset reflux ratio. The constructed wetland is composed of a vertical subsurface flow wetland and a surface flow wetland connected in series. The vertical subsurface flow wetland is filled with zeolite packing and planted with emergent plants, while the surface flow wetland is planted with emergent plants and submerged plants. The effluent from the constructed wetland enters an ecological pond, where economic crops are planted and filter-feeding fish are stocked. After being collected by a collection well, the effluent meets the standards.

3. The deep denitrification system for reclaimed water coupled with an constructed wetland and a denitrifying filter according to claim 1, characterized in that: The marking method for failed sensors is as follows: Establish a multi-source sensor monitoring network, including a core water quality monitoring unit and an environmental monitoring unit; The core water quality monitoring unit is deployed at the outlet of the equalization tank and integrates a variety of online monitoring devices. For the key online monitoring devices, at least two sensors with different measurement principles are configured to form a heterogeneous redundant combination. The environmental monitoring unit includes a water temperature sensor deployed inside the denitrification filter. The programmable logic controller (PLC) collects real-time measurements from each sensor, integrates them to obtain water quality data vectors and water temperature data, runs a sensor health self-diagnosis thread, calculates the health coefficient of each sensor. The health coefficient is obtained by weighted fusion of the sensor's consistency index, reliability index and stability index. The health coefficient is compared with a preset allowable threshold. If it is lower than the preset allowable threshold, the sensor is determined to be a failed sensor.

4. The deep denitrification system for reclaimed water coupled with an artificial wetland and a denitrifying filter according to claim 3, characterized in that: The consistency index is obtained as follows: For sensors configured with heterogeneous redundancy, the real-time measurement values ​​of each sensor in the heterogeneous redundancy combination are read synchronously at the same time. The average deviation is obtained by averaging the relative deviation between the real-time measurement value of the sensor and the real-time measurement values ​​of different sensors in the heterogeneous redundancy combination. The consistency index is obtained by data processing based on the average deviation. For sensors without heterogeneous redundant combinations, a lightweight LSTM time-series prediction model is established to predict the expected value at the current moment and the data is processed to obtain the prediction residual. The historical residual sequence is integrated and statistically analyzed, and the consistency index is calculated based on the deviation of the current prediction residual from the normal fluctuation range.

5. The deep denitrification system for reclaimed water coupled with an artificial wetland and a denitrifying filter according to claim 4, characterized in that: The reliability and stability indicators are obtained as follows: The reliability index is calculated based on the dynamic historical operation file established for the sensor, which includes the cumulative number of failures, the time since the last calibration, the preset standard calibration cycle duration, and the device operating time. The stability index is obtained by normalizing the coefficient of variation of the sensor's real-time measurement values ​​at various times within a preset short period of time.

6. The deep denitrification system for reclaimed water coupled with an constructed wetland and a denitrifying filter according to claim 1, characterized in that: The method for determining whether a real water quality shock load has occurred is as follows: The water quality anomaly monitoring thread integrates a preliminary anomaly capture unit and a comprehensive judgment unit; After filtering out failed sensor data based on the health coefficient, a high-confidence water quality data vector is obtained. The actual fusion value of each water quality indicator is obtained with a preset sampling period. For water quality indicators with heterogeneous redundancy combinations, the health coefficient is used as the weight to perform weighted fusion of the corresponding heterogeneous redundant data groups in the high-confidence water quality data vector to obtain the actual fusion value. Otherwise, the corresponding water quality indicator data in the high-confidence water quality data vector is directly used as the actual fusion value. The initial anomaly capture unit is executed to obtain a preliminary water quality anomaly marker. The comprehensive judgment unit is then activated to integrate the actual fusion values ​​to obtain a multi-dimensional water quality feature vector. The Mahalanobis distance between the multi-dimensional water quality feature vector and the preset historical normal operating condition feature vector is calculated. If the distance exceeds the preset dynamic Mahalanobis distance threshold, it is determined that a real water quality shock load has occurred.

7. The deep denitrification system for reclaimed water coupled with an constructed wetland and a denitrifying filter according to claim 6, characterized in that: The method for obtaining preliminary markers of water quality anomalies is as follows: Set the capture flag value and assign it an initial value of 0. Based on the high-confidence water quality data vector, query the pre-constructed water temperature-water quality correlation model to obtain the normal fluctuation range of each water quality indicator corresponding to the current water temperature data. Compare the actual fusion value of each water quality indicator with the corresponding normal fluctuation range item by item. When the actual fusion value of any water quality indicator in a preset number of consecutive sampling periods exceeds the corresponding normal fluctuation range and the capture flag value is 0, set the capture flag value to 1 and record the preliminary water quality anomaly. The preliminary water quality anomaly includes the water quality indicator exceeding the standard, the magnitude of the exceedance, the time of occurrence, and the duration.

8. The deep denitrification system for reclaimed water coupled with an constructed wetland and a denitrifying filter according to claim 7, characterized in that: The pre-construction method for the water temperature-water quality correlation model is as follows: Water temperature data under the baseline process and the actual fusion values ​​of various water quality indicators within the historical standard analysis period with the current time as the endpoint and a preset standard analysis duration are collected. The water temperature data is segmented according to a preset length water temperature interval. The 5th percentile of the actual fusion value of each water quality indicator within each water temperature interval is used as the lower limit of normal fluctuation and the 95th percentile is used as the upper limit of normal fluctuation. The cubic spline interpolation method is used to fit a continuous water temperature-fluctuation upper limit value curve and a water temperature-fluctuation lower limit value curve based on each water temperature interval, and the integration is used to obtain a water temperature-water quality correlation model.

9. A deep denitrification system for reclaimed water coupled with an artificial wetland and a denitrifying filter according to claim 7, characterized in that: The method of type classification is as follows: The anomaly monitoring thread integrates a preliminary anomaly capture unit and a comprehensive judgment unit. The comprehensive judgment unit includes a multi-parameter coupled verification subunit and a timing anomaly capture subunit. When a real water quality shock load is determined to occur, the grey correlation degree between various water quality indicators in the past short period of time is calculated in parallel. If the grey correlation degree is lower than the preset correlation threshold, it is determined to be a local shock anomaly; otherwise, it is determined to be a global anomaly. If it is a local shock anomaly, it is further identified as a pulse shock load or a continuous abnormal load through the time-series anomaly capture subunit.

10. A deep denitrification system for reclaimed water coupled with an constructed wetland and a denitrifying filter according to claim 9, characterized in that: The method for distinguishing between pulse-type impact loads and continuous abnormal loads is as follows: For localized impact anomalies, the actual fused values ​​of the water quality indicators exceeding the standard for L sampling periods prior to the current moment are read to form the indicator time series within the sliding window. This sequence is input into a pre-constructed time series prediction model to predict the expected value of the indicator at the current moment. The absolute value of the difference between the actual fused value and the expected value of the indicator is calculated as the indicator residual. If the indicator residual exceeds the preset indicator residual threshold, the current moment is recorded as the time of exceeding the limit. The difference between the indicator residual and the preset indicator residual threshold is calculated to obtain the value of exceeding the limit. Data processing is performed to obtain the cumulative impact energy and the duration of exceeding the limit within the sliding window. If the cumulative impact energy exceeds the preset energy threshold and the duration of exceeding the limit does not exceed the preset short duration upper limit, it is determined to be a pulse-type impact load. If it exceeds the preset short duration upper limit, it is determined to be a continuous abnormal load.

Images