In-situ treatment method for initial rainwater based on COD prediction in a regulating reservoir

Abstract

The application discloses a method for in-situ treatment of initial rainwater based on COD prediction in a regulating reservoir, comprising the following steps: firstly, pre-training an online learning adaptive neural network model by using historical working data of an existing regulating reservoir for initial rainwater; then, building a high-efficiency biochemical tank capable of controlling sludge discharge and aeration quantity at the regulating reservoir to be measured which needs to perform COD prediction; collecting data and inputting the data into the online learning adaptive neural network model while operating the high-efficiency biochemical tank to treat water in the regulating reservoir to be measured, controlling the operation of the high-efficiency biochemical tank according to the obtained feedback and performing online incremental learning on the online learning adaptive neural network model; and ending the training until the prediction result of the online learning adaptive neural network model after the online incremental learning meets the requirements. The application can well solve the characteristics that the water quality of initial rainwater changes over time and the difficulty that pollutants in rainwater are difficult to accurately predict.

Description

Technical Field

[0001] This invention relates to the field of rainwater treatment technology, and in particular to an in-situ treatment method for initial rainwater based on COD prediction in a stormwater storage tank. Background Technology

[0002] With the rapid development of compact cities and the frequent occurrence of extreme rainfall due to global climate change, many cities are experiencing severe urban flooding. Simultaneously, rainwater carries pollutants from the atmosphere, surface water, and drainage networks, posing a serious threat to river water quality after being discharged into rivers, representing a significant source of urban non-point source pollution. Furthermore, the impact loads and overflow pollution from combined sewer systems increase the difficulty of urban water environment management, while separate sewer systems also suffer from rainwater discharge pollution during the rainy season caused by initial rainwater runoff or mixed sewage. Therefore, polluted rainwater has increasingly become a significant factor affecting the water quality of urban rivers, demanding sufficient attention and consideration.

[0003] Rainwater storage tanks are important engineering facilities for mitigating urban flooding and controlling initial rainwater pollution both domestically and internationally. By storing initial rainwater, they can effectively reduce pollutants discharged into rivers. Drawing on the engineering experience of rainwater storage projects already implemented at home and abroad, statistics show that as of December 2022, Shanghai had successively built 18 rainwater storage tanks with a total volume of 189,700 m³. 3 However, this model has the following problems when operating in Shanghai:

[0004] (1) Heavy rainfall and serious overflow pollution: Shanghai has a subtropical monsoon climate with abundant rainfall. If the rainwater in the storage tank is not treated in time, it will overflow. According to statistics, in 2023, more than 35 million tons of sewage overflowed and were directly discharged into the Yangtze River.

[0005] (2) The content of organic matter carried by rainwater is high: the suspended matter in the initial rainwater is about 40-160 mg / L and the chemical oxygen demand is about 80-200 mg / L. Moreover, the organic matter in the rainwater has the characteristic of changing with time and environmental factors.

[0006] (3) Lack of efficient in-situ treatment technology for rainwater storage tanks: Wastewater treatment plants in Shanghai are relatively concentrated, and polluted rainwater can only enter the wastewater treatment plants after long-distance transportation. Long-distance transportation consumes a lot of energy and will increase carbon emissions.

[0007] To effectively address the aforementioned issues, it is imperative to develop a key technology for in-situ treatment of urban initial rainwater that can automatically regulate rainfall and organic matter in rainwater.

[0008] By using a highly efficient biological treatment method, the sludge age and aeration rate of the high-efficiency biological treatment tank can be adjusted according to the organic load of the influent, achieving an efficient removal of 80% of organic matter from rainwater.

[0009] To address the difficulty in predicting the organic load of influent to high-efficiency biological treatment ponds, according to local regulations in some cities, when rainfall meets the initial rainwater standard, it can be assumed that there is a corresponding relationship between the COD concentration in the rainfall and the rainfall amount, that is, there is a corresponding relationship between the rainwater storage capacity in the storage pond and the COD concentration in the pond.

[0010] However, this correspondence is a complex nonlinear relationship, which traditional mathematical models struggle to predict accurately.

[0011] In recent years, artificial intelligence algorithms such as machine learning and deep learning have been increasingly used in the industrial field. Thanks to their powerful ability to process complex data, artificial intelligence algorithms are often used for water quality prediction.

[0012] Among them, online learning adaptive neural networks are a type of neural network that can automatically update its parameters and structure based on real-time input data. They can effectively address concept drift and solve the problem of changes in organic matter in rainwater over time and with environmental factors.

[0013] However, there is currently no big data model for predicting COD in initial rainwater or an efficient in-situ COD treatment process.

[0014] Therefore, how to consider the impact of rainwater volume in the storage tank on pollutant content and how to adjust the operating parameters of the in-situ rainwater treatment process in the storage tank have become technical problems that urgently need to be solved by those skilled in the art. Summary of the Invention

[0015] In view of the aforementioned deficiencies in the prior art, this invention provides a method for treating initial rainwater runoff and an online intelligent control system. The aim is to employ an online learning model, enabling the adaptive neural network to adjust itself promptly as data distribution and the environment change—for example, by increasing or decreasing the number of neurons or layers—to adapt to new trends and patterns and maintain high predictive performance. In predicting pollutant concentrations in initial rainwater runoff, the online learning adaptive neural network effectively addresses the concept drift problem, allowing the prediction model to identify and adapt to these changes. This, in turn, enables the intelligent control system to more accurately predict the aeration rate of the high-efficiency biological treatment tank, achieving efficient treatment of initial rainwater runoff.

[0016] To achieve the above objectives, this invention discloses an in-situ treatment method for initial rainwater based on COD prediction in a stormwater storage tank, comprising the following steps:

[0017] Step 1: Pre-train an online learning adaptive neural network model using existing historical working data from stormwater storage tanks for initial rainfall.

[0018] Step 2: Construct a high-efficiency biological treatment tank next to the storage tank where COD prediction is required, which can control the amount of sludge discharged and the amount of aeration.

[0019] The high-efficiency biological treatment tank is inoculated with aerobic activated sludge and equipped with a device that can monitor dissolved oxygen concentration in real time. Water is drawn from the bottom of the storage tank to be tested by an inlet pump.

[0020] The storage tank to be tested is equipped with a level gauge;

[0021] Step 3: Run the high-efficiency biochemical pool, specifically as follows:

[0022] The liquid level data of the storage tank to be tested, obtained by the liquid level gauge, is converted into a percentage value of the storage water level to be tested that matches the maximum volume of the storage tank to be tested, and the water pump is controlled to deliver the water to the high-efficiency biological tank according to the percentage value of the storage water level to be tested.

[0023] By controlling the amount of sludge discharged from the high-efficiency biological treatment tank, the sludge age of the aerobic activated sludge is kept below 2 days, thereby achieving an 80% removal rate of organic matter in the initial rainwater.

[0024] Step 4: Input the liquid level data of the regulating reservoir after rainfall into the online learning adaptive neural network model, control the operation of the high-efficiency biological treatment tank based on the model's prediction results, and perform online incremental learning on the online learning adaptive neural network model. Specifically:

[0025] Step 4.1: Obtain the liquid level data of the storage tank under test through the liquid level gauge, and normalize the liquid level data and input it into the pre-trained online learning adaptive neural network model.

[0026] Step 4.2: Calculate the influent organic load of the high-efficiency biological treatment tank based on the predicted COD concentration fed back by the online learning adaptive neural network model and the flow rate of the influent pump.

[0027] Step 4.3: Calculate the aeration rate required for the high-efficiency biological system in the high-efficiency biological treatment tank based on the organic load of the influent. Under the premise of ensuring that the dissolved oxygen concentration in the high-efficiency biological treatment tank is not less than 2 mg / L, feedforward control is performed on the aeration rate to ensure that there is sufficient dissolved oxygen in the high-efficiency biological treatment tank for the aerobic activated sludge to remove organic matter.

[0028] Step 4.4: Perform online incremental learning on the online learning adaptive neural network model based on the actual measured COD concentration obtained from the storage tank under test, and save the updated model.

[0029] Step 5: After collecting the initial rainwater of the next rainfall in the storage tank, repeat steps 3 and 4 to enable the online learning adaptive neural network model to learn the changes in the correspondence between rainfall and COD in a timely manner, thereby improving the prediction accuracy of the model.

[0030] Preferably, in step 1, the water storage in the historical working data is converted into a percentage value that matches the maximum volume of the corresponding regulating reservoir, and all the percentage values ​​are used as pre-training inputs for pre-training the online learning adaptive neural network model;

[0031] The COD concentration in the historical working data is used as the pre-training output of the online learning adaptive neural network model.

[0032] Preferably, it also includes a PLC and an online intelligent control system including the PLC;

[0033] In step 3, the online intelligent control system converts the liquid level data of the storage tank to be tested obtained from the liquid level gauge into a percentage value of the storage tank water level that matches the maximum volume of the storage tank, and controls the water pump to deliver the water to the high-efficiency biological tank according to the percentage value of the storage tank water level.

[0034] Preferably, the online intelligent control system controls the operating frequency of the water inlet pump to be proportional to the water storage capacity of the storage tank under test.

[0035] Preferably, it also includes a PLC and an online intelligent control system including the PLC;

[0036] In step 4.2, the online intelligent control system calculates the influent organic load of the high-efficiency biological treatment tank based on the predicted COD concentration fed back by the online learning adaptive neural network model and the flow rate of the influent pump.

[0037] Preferably, it also includes a PLC and an online intelligent control system including the PLC;

[0038] In step 4.3, the online intelligent control system calculates the aeration rate required for the high-efficiency biological system in the high-efficiency biological tank based on the organic load of the influent.

[0039] Preferably, the device capable of real-time monitoring of dissolved oxygen concentration is an online dissolved oxygen electrode; the high-efficiency biological treatment tank is aerated by setting up an aeration blower; the high-efficiency biological treatment tank is sludge discharged by setting up a sludge discharge pump to control the sludge age of activated sludge to be less than 2 days.

[0040] The beneficial effects of this invention are:

[0041] This invention utilizes an online learning adaptive neural network model to predict the COD concentration in a stormwater storage tank, and adjusts the aeration intensity of a high-efficiency biological method based on the predicted COD concentration to achieve in-situ efficient removal of COD from initial rainwater.

[0042] The online learning intelligent model in this invention can be pre-trained using historical data and then incrementally learned based on real-time data. This effectively addresses the characteristic of the water quality of initial rainwater changing over time and solves the problem of accurately predicting pollutants in rainwater.

[0043] This invention introduces an online learning adaptive neural network model, which, compared to traditional neural network models, can receive real-time data and update model parameters accordingly. This allows the model to better adapt to environmental changes and address the issue of pollutant variations in initial rainwater over time. After pre-training with historical data from stormwater storage tanks, the online learning adaptive neural network model can quickly predict the COD content in the tanks based on the water volume after rainfall. Compared to the lag in manual measurements, the model's predictions are more timely and allow for more timely adjustments to the initial rainwater treatment process. Furthermore, after providing prediction results, the model continues to learn online based on manually measured water quality data, automatically adjusting its structural parameters and combining current and historical data to learn the relationship between water volume and COD content in the storage tanks. Compared to traditional neural network models, the online learning adaptive neural network model saves computational resources and continuously improves the model's predictive accuracy.

[0044] This invention employs a highly efficient biological method to achieve efficient COD removal from initial rainwater, and utilizes a feedforward feedback aeration control strategy to manage the aeration of this method. The online intelligent control system predicts the influent organic load of the highly efficient biological process based on the influent COD concentration using an online learning adaptive neural network, thereby determining the required aeration rate. To avoid incomplete COD removal due to insufficient aeration at low influent organic loads, dissolved oxygen feedback aeration control is implemented. When the dissolved oxygen in the biological treatment tank falls below 2 mg / L, the online intelligent control system increases the aeration rate to ensure that the dissolved oxygen level in the tank does not fall below 2 mg / L, thus ensuring effective COD removal while avoiding energy waste caused by over-aeration.

[0045] This invention has significant technical advantages such as high accuracy, strong timeliness, strong adaptability and high efficiency. It can effectively solve the problems of in-situ treatment of rainwater and difficulty in predicting rainwater quality in the current stage of initial rainwater treatment, and provides a new solution for the efficient treatment of initial rainwater.

[0046] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0047] Figure 1 This diagram illustrates the connection relationships between the components of an embodiment of the present invention. Detailed Implementation

[0048] Example 1

[0049] like Figure 1 As shown, the in-situ treatment method for initial rainwater based on COD prediction in a stormwater storage tank includes the following steps:

[0050] Step 1: Pre-train an online learning adaptive neural network model using existing historical working data from stormwater storage tanks for initial rainfall.

[0051] Step 2: Construct a high-efficiency biological treatment tank 2 next to the storage tank 1 where COD prediction is required; this tank can control the amount of sludge discharged and the amount of aeration.

[0052] The high-efficiency biological treatment tank 2 is inoculated with aerobic activated sludge and equipped with a device 5 that can monitor dissolved oxygen concentration in real time. Water is drawn from the bottom of the storage tank 1 to be tested through the inlet pump 3.

[0053] The storage tank 1 to be tested is equipped with a level gauge 6;

[0054] Step 3: Run the high-efficiency biological tank 2, specifically as follows:

[0055] The liquid level data of the storage tank 1 obtained from the level gauge 6 is converted into a percentage value of the storage water level that matches the maximum volume of the storage tank 1. The water pump 3 is then controlled to deliver water to the high-efficiency biological treatment tank 2 according to the percentage value of the storage water level. Through the above technical means, the inflow rate of the high-efficiency biological treatment tank 2 changes with the amount of water stored in the storage tank 1.

[0056] By controlling the amount of sludge discharged from the high-efficiency biological treatment tank 2, the sludge age of the aerobic activated sludge is kept below 2 days, thereby achieving an 80% removal rate of organic matter in the initial rainwater.

[0057] Step 4: Input the liquid level data of the regulating reservoir after rainfall into the online learning adaptive neural network model, control the operation of the high-efficiency biological treatment tank 2 based on the model's prediction results, and perform online incremental learning on the online learning adaptive neural network model:

[0058] Step 4.1: Obtain the liquid level data of the storage tank 1 to be tested through the liquid level gauge 6, and normalize the liquid level data and input it into the pre-trained online learning adaptive neural network model.

[0059] Step 4.2: Calculate the influent organic load of the high-efficiency biological treatment tank 2 based on the predicted COD concentration fed back by the online learning adaptive neural network model and the flow rate of the influent pump 3.

[0060] Step 4.3: Calculate the aeration rate required for the high-efficiency biological system in the high-efficiency biological treatment tank 2 based on the organic load of the influent. Under the premise of ensuring that the dissolved oxygen concentration in the high-efficiency biological treatment tank 2 is not less than 2 mg / L, feedforward control is carried out on the aeration rate to ensure that there is sufficient dissolved oxygen in the high-efficiency biological treatment tank 2 for the aerobic activated sludge to remove organic matter.

[0061] Step 4.4: The online learning adaptive neural network model is incrementally learned online based on the COD concentration obtained from the actual measurement of the storage tank 1.

[0062] In this process, the online learning adaptive neural network model calculates the loss gradient based on its own prediction results and the COD concentration obtained from the actual measurement of the storage tank 1, and uses the loss gradient to update the weights and biases of the existing model, thereby updating the model parameters to reduce the loss and realizing the online learning of the model.

[0063] Step 5: After the model is trained online, save the model for the next COD concentration prediction. Then, repeat steps 3 and 4 based on the liquid level data in the storage tank after the next rainfall to update the model parameters in a timely manner, adapt to the time changes in rainwater quality, solve the concept drift problem, and achieve accurate prediction and removal of COD in the initial rainwater.

[0064] This invention utilizes a highly efficient biological method to achieve rapid and effective removal of COD from initial rainwater. Simultaneously, it employs an online intelligent control system to predict the COD concentration in the storage tank and adjust the process parameters of the highly efficient biological method accordingly, thereby avoiding substandard effluent and rainwater overflow from the storage tank during rainy days.

[0065] The efficient biological method adopted in this invention can effectively treat initial rainwater in situ, reducing the cost of long-distance transportation.

[0066] Moreover, the online intelligent control system can predict the COD concentration and adjust the influent flow rate of the influent pump based on the water volume in the storage tank, thereby achieving real-time control of the organic load of the influent and enabling precise aeration of the efficient biological process, providing a new solution for the efficient treatment of initial rainwater.

[0067] In this invention, while the online learning adaptive neural network model predicts the COD concentration in the storage tank based on real-time input data, it also performs incremental learning based on the received actual measured COD concentration, which can continuously improve the model's adaptability and enhance its predictive ability.

[0068] In some embodiments, in step 1, the water storage in the historical working data is converted into a percentage value that matches the maximum volume of the corresponding storage tank, and all percentage values ​​are used as pre-training inputs for a pre-trained online learning adaptive neural network model.

[0069] The COD concentration in historical working data was used as the pre-training output of the pre-trained online learning adaptive neural network model.

[0070] In some embodiments, the system also includes a PLC8 and an online intelligent control system 9 including the PLC8;

[0071] In step 3, the online intelligent control system 9 converts the liquid level data of the storage tank 1 obtained from the liquid level gauge 6 into a percentage value of the storage water level that matches the maximum volume of the storage tank 1 through the PLC8, and controls the inlet pump 3 to deliver water to the high-efficiency biological tank 2 according to the percentage value of the storage water level.

[0072] In some embodiments, the online intelligent control system 9 controls the operating frequency of the inlet pump 3 to be proportional to the water storage capacity of the storage tank 1 under test.

[0073] In some embodiments, the system also includes a PLC8 and an online intelligent control system 9 including the PLC8;

[0074] In step 4.2, the online intelligent control system 9 calculates the influent organic load of the high-efficiency biological treatment tank 2 based on the predicted COD concentration fed back by the online learning adaptive neural network model and the flow rate of the influent pump 3.

[0075] In some embodiments, the system also includes a PLC8 and an online intelligent control system 9 including the PLC8;

[0076] In step 4.3, the online intelligent control system 9 calculates the aeration volume required for the high-efficiency biological system in the high-efficiency biological tank 2 based on the organic load of the influent.

[0077] In some embodiments, the device 5 that can monitor dissolved oxygen concentration in real time is an online dissolved oxygen electrode; the high-efficiency biological treatment tank 2 is aerated by setting an aeration blower 4 and sludge is discharged by setting a sludge pump 7 to control the sludge age to be less than 2 days.

[0078] Example 2

[0079] This embodiment provides a highly efficient biological method for treating initial rainwater using an online learning intelligent model for control. A schematic diagram of the process is shown below. Figure 1 As shown.

[0080] The process mainly includes a stormwater storage tank 1, a high-efficiency biological treatment tank 2, and an online intelligent control system 9. The high-efficiency biological treatment tank 2 is built next to the stormwater storage tank 1 to enable in-situ treatment of initial rainwater runoff.

[0081] The storage tank 1 to be tested is equipped with a level gauge 6 to enable the monitoring of the water volume in the storage tank 1 and the transmission of data.

[0082] Aerobic activated sludge is inoculated into the high-efficiency biological treatment tank 2, and the activated sludge in the tank is controlled within a short sludge age by the sludge discharge pump 7.

[0083] The online intelligent control system 9 includes a control platform containing a PLC8 and a client platform. The client programming control platform uses the MATLAB platform, and the PLC8 uses the Siemens S7-1200 platform. Communication between the client and the PLC8 is via the TCP / IP protocol. The online intelligent control system 9 utilizes... Figure 1 The connection structure shown controls the operation of the aeration blower 4.

[0084] First, historical data from 2020 to 2022 was preprocessed, including converting water level data into water volume percentages, dividing the input data into different datasets and normalizing them, then pre-training and optimizing the model, and finally using the optimal online learning adaptive neural network model from the pre-training to predict COD concentration in real time.

[0085] In this embodiment, the total volume of the storage tank 1 to be tested is 3500 m³. 3 The initial rainwater volume collected in the controlled storage tank 1 after a rainfall event was 3337 m³. 3 This accounts for 95% of the total volume of the storage tank 1 under test. At this time, the control system adjusts the influent frequency of the influent pump of the high-efficiency biological treatment tank 2 to 95%, that is, the influent flow rate is 141 m³ / h. 3 The online learning adaptive neural network predicted the initial rainwater COD concentration to be 65.47 mg / L based on the water volume in the storage tank 1. The intelligent control system then calculated the influent organic load for the efficient biological process based on the influent flow rate and COD concentration, and further estimated the required aeration rate to be 9.66 m³ / h. 3 / h, thereby regulating the operating frequency of the aeration blower 4.

[0086] After measuring the initial rainwater quality in the storage tank 1, the COD concentration was found to be 50.4 mg / L. The COD concentration measurement and the water volume in storage tank 1 were used as new data sets and input into the online learning model for incremental learning, resulting in a loss gradient of 227.352. This gradient was then used to adjust the weights and biases of the existing model, enabling online learning and updating of the model. The updated model was then saved for the next COD concentration prediction.

[0087] Example 3

[0088] The apparatus and implementation method in this embodiment are the same as those in the previous embodiments, and will not be repeated here. In this embodiment, the online adaptive neural network model has undergone three iterations. During a subsequent rainfall event, the initial rainwater volume collected in the storage tank 1 was 3617 m³. 3This accounts for 103% of the total volume of the storage tank 1 under test. At this time, the control system adjusts the influent frequency of the influent pump of the high-efficiency biological treatment tank 2 to 100%, that is, the influent flow rate is 150 m³ / h. 3 / d; The online learning adaptive neural network predicted the initial rainwater COD concentration to be 66.81 mg / L based on the water volume in the storage tank 1. The intelligent control system then calculated the influent organic load for the efficient biological process based on the influent flow rate and COD concentration, and further estimated the required aeration rate to be 10.32 m³ / L. 3 / h, thereby regulating the operating frequency of the aeration blower 4.

[0089] After measuring the initial rainwater quality in the storage tank 1, the COD concentration was found to be 63 mg / L. The measured COD concentration and the water volume in storage tank 1 were then input into the online learning model for incremental learning. The resulting loss gradient was 14.578. Compared to previous predictions, the model's loss gradient was significantly lower, and the error between the predicted and measured values ​​was significantly reduced. This indicates that the model has learned the recent trend of pollutant changes in rainwater and can accurately predict the COD concentration in the initial rainwater. The online learning algorithm then uses the gradient to adjust the weights and biases of the existing prediction model, continuing the online learning and updating of the model. The updated model is then saved for the next COD concentration prediction.

[0090] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for in-situ treatment of initial rainwater based on COD prediction in a stormwater storage tank; characterized in that, Includes the following steps: Step 1: Pre-train an online learning adaptive neural network model using existing historical working data from stormwater storage tanks. Step 2: Construct a high-efficiency biological treatment tank (2) next to the storage tank (1) where COD prediction is required, which can control the amount of sludge discharged and the amount of aeration. The high-efficiency biochemical tank (2) is inoculated with aerobic activated sludge and equipped with a device (5) that can monitor dissolved oxygen concentration in real time. Water is drawn from the bottom of the storage tank (1) to be tested by the inlet pump (3). The storage tank (1) to be tested is equipped with a level gauge (6); Step 3: Run the high-efficiency biochemical pool (2), specifically as follows: The liquid level data of the storage tank (1) to be tested obtained from the liquid level gauge (6) is converted into a percentage value of the storage water level to be tested that matches the maximum volume of the storage tank (1), and the water pump (3) is controlled to deliver the water to the high-efficiency biological tank (2) according to the percentage value of the storage water level to be tested. By controlling the sludge discharge rate of the high-efficiency biological treatment tank (2), the sludge age of the aerobic activated sludge can meet the requirement that the COD removal rate in the initial rainwater reaches 80%. Step 4: Collect data and input it into the online learning adaptive neural network model. Based on the feedback obtained, control the operation of the high-efficiency biochemical pool (2) and perform online incremental learning on the online learning adaptive neural network model. Specifically: Step 4.1: Obtain the liquid level data of the storage tank (1) to be tested through the liquid level gauge (6), and normalize the liquid level data and input it into the pre-trained online learning adaptive neural network model. Step 4.2: Calculate the influent organic load of the high-efficiency biological treatment tank (2) based on the predicted COD concentration fed back by the online learning adaptive neural network model and the flow rate of the influent pump (3); Step 4.3: Calculate the aeration volume required for the high-efficiency biological system in the high-efficiency biological tank (2) based on the organic load of the influent. Under the premise of ensuring that the dissolved oxygen concentration in the high-efficiency biological tank (2) is not less than 2 mg / L, feedforward control is carried out on the aeration volume to ensure that there is sufficient dissolved oxygen in the high-efficiency biological tank (2) for the aerobic activated sludge to remove organic matter. Step 4.4: The online learning adaptive neural network model is incrementally learned online based on the COD concentration obtained from the actual measurement of the storage tank (1) to be tested; Step 5: Repeat steps 3 and 4 until the prediction results of the online learning adaptive neural network model of the online incremental learning meet the requirements; It also includes a PLC (8) and an online intelligent control system (9) including the PLC (8); In step 3, the online intelligent control system (9) converts the liquid level data of the storage tank (1) obtained from the liquid level gauge (6) into a percentage value of the storage water level that matches the maximum volume of the storage tank (1) through the PLC (8), and controls the water pump (3) to deliver the water to the high-efficiency biological tank (2) according to the percentage value of the storage water level. The online intelligent control system (9) controls the operating frequency of the water inlet pump (3) to be proportional to the water storage capacity of the storage tank (1) to be tested; In step 4.2, the online intelligent control system (9) calculates the influent organic load of the high-efficiency biological tank (2) based on the predicted COD concentration fed back by the online learning adaptive neural network model and the flow rate of the influent pump (3); In step 4.3, the online intelligent control system (9) calculates the aeration volume required for the high-efficiency biological system in the high-efficiency biological tank (2) based on the organic load of the influent.

2. The method for in-situ treatment of initial rainwater based on COD prediction in a stormwater storage tank according to claim 1, characterized in that, In step 1, the water storage in the historical working data is converted into a percentage value that matches the maximum volume of the corresponding regulating reservoir, and all the percentage values ​​are used as pre-training inputs for pre-training the online learning adaptive neural network model. The COD concentration in the historical working data is used as the pre-training output of the online learning adaptive neural network model.

3. The method for in-situ treatment of initial rainwater based on COD prediction in a stormwater storage tank according to claim 1, characterized in that, The device (5) that can monitor dissolved oxygen concentration in real time is an online dissolved oxygen electrode; the high-efficiency biochemical tank (2) is aerated by setting an aeration blower (4) and sludge is discharged by setting a sludge discharge pump (7).

Images