BiLSTM-based UV / H2O2 advanced oxidation device parameter control method and system

Abstract

The invention provides a UV / H2O2 advanced oxidation device parameter control method and system based on BiLSTM, and belongs to the technical field of water treatment. The method comprises the following steps: constructing an edge-cloud collaborative architecture, collecting water inlet data and water outlet data of a UV / H2O2 advanced oxidation device in real time by an edge end, and uploading the water inlet data and the water outlet data to a cloud; the cloud screens the types of the water inlet data and the water outlet data through Granger causality test; building a lightweight BiLSTM model, and training to obtain a control parameter prediction model; the cloud end issues the control parameter prediction model to the edge end, and the edge end collects water inlet data and water outlet data in real time and inputs the data into the control parameter prediction model to predict the hydrogen peroxide dosage and the ultraviolet power; the edge end drives the UV / H2O2 advanced oxidation device to adjust the hydrogen peroxide adding amount and the ultraviolet power according to the predicted hydrogen peroxide adding amount and the predicted ultraviolet power. According to the invention, through a BiLSTM machine learning algorithm, the dosage and ultraviolet light intensity in the UV / H2O2 advanced oxidation device can be accurately predicted and controlled.

Description

Technical Field

[0001] This invention relates to a parameter control method and system for a BiLSTM-based UV / H2O2 advanced oxidation device, belonging to the field of water treatment technology. Background Technology

[0002] In the field of water treatment, especially for advanced treatment to remove pollutants that are difficult to treat using conventional processes, advanced oxidation methods have become a widely used approach. With the continuous improvement of environmental protection standards and the increasing complexity of water pollution problems, how to achieve efficient, stable, and economical deep water purification has become a key challenge for the water treatment industry. UV / H2O2 advanced oxidation technology, as a highly efficient treatment method without secondary pollution, has received widespread attention and application due to its ability to generate highly oxidizing hydroxyl radicals that effectively degrade organic pollutants in water.

[0003] However, in actual operation, the performance of UV / H2O2 advanced oxidation units is affected by a variety of factors, including fluctuations in influent water quality and quantity, uneven distribution of UV light intensity, and the internal structural design of the reactor. These factors work together to make it difficult for traditionally experience-based water treatment plants to accurately match process parameters to actual water quality conditions. They often resort to conservative strategies such as overdosing on chemicals and increasing UV power to ensure effluent meets standards. This approach not only increases chemical dosing costs but also shortens the lifespan of UV lamps, raising overall operating costs.

[0004] Currently, intelligent control systems for water treatment processes are mainly divided into two categories: one is based on automatic control systems, which are constrained by the nonlinearity, uncertainty, time delay, and numerous variables in the wastewater treatment process, making it difficult to achieve efficient and stable control; the other is based on intelligent algorithms, which can consider the influence of more variables on the dosage of chemicals, realizing the transformation from automation to intelligence, but the amount of parameters collected is large, the algorithm prediction is slow, and it is difficult to guarantee the timeliness of control. Summary of the Invention

[0005] The purpose of this invention is to provide a parameter control method and system for a UV / H2O2 advanced oxidation device based on BiLSTM, which realizes precise control of the dosing and parameters of the UV / H2O2 advanced oxidation device, significantly improving the processing efficiency and reducing the operating cost.

[0006] To achieve the above objectives, the present invention employs the following technical solution: Firstly, a method for parameter control of a BiLSTM-based UV / H2O2 advanced oxidation device is provided, comprising the following steps: An edge-cloud collaborative architecture is constructed, in which the influent and effluent data of the UV / H2O2 advanced oxidation unit are collected in real time at the edge and uploaded to the cloud; The cloud-based system uses Granger causality tests to filter influent and effluent data types. A lightweight BiLSTM model was built, using the filtered influent data, effluent data and feedback features as input data, and the corresponding hydrogen peroxide dosage and ultraviolet power as label data. The lightweight BiLSTM model was trained to obtain the control parameter prediction model. The cloud sends the control parameter prediction model to the edge device, and the edge device collects influent and effluent data in real time and inputs them into the control parameter prediction model to predict the hydrogen peroxide dosage and ultraviolet power. The edge end drives the UV / H2O2 advanced oxidation unit to adjust the hydrogen peroxide dosage and UV power based on the predicted hydrogen peroxide dosage and UV power.

[0007] Preferably, the influent data includes influent flow rate, influent turbidity, influent conductivity, and influent UV. 254 , Influent TOC (Total Organic Carbon), Influent New Pollutant Concentration, Influent Temperature, Influent pH (Acidity / Alkalinity), Influent Hydrogen Peroxide Concentration; The effluent data includes effluent UV. 254 Effluent turbidity, effluent conductivity, effluent TOC, concentration of new pollutants in effluent, effluent temperature, effluent pH, effluent hydrogen peroxide concentration, and ultraviolet light intensity; The new pollutants include antibiotics, tetracyclines, and odorants.

[0008] Preferably, the feedback characteristic components are constructed through PID business logic, including proportional feedback characteristics, integral feedback characteristics, and derivative feedback characteristics, and the expressions for the feedback characteristic components are as follows: , , , in, For the current predicted time, for Deviation in dosage at any time , for The hydrogen peroxide dosage value at all times. Set the hydrogen peroxide value. The current time-proportional feedback feature, The integral feedback feature at the current time. This represents the differential feedback characteristic at the current moment.

[0009] Preferably, Granger causality test is used to screen influent and effluent data types, retaining data types with p-values ​​less than 0.05; Specifically, it includes: ADF stationarity test and differential transformation were performed on the 24-hour time series data windows of influent and effluent data. Selecting the optimal lag order for the VAR model based on the AIC / BIC criterion; Using pollutant concentration as the dependent variable, we tested the Granger causality of each parameter, retaining data types with p-values ​​less than 0.05.

[0010] Preferably, the lightweight BiLSTM model consists of two independent LSTM layers, one for processing forward sequences and the other for processing reverse sequences; each LSTM layer includes an input layer, three LSTM cell layers, and an output layer, wherein each LSTM cell layer contains a forget gate, an input gate, and an output gate.

[0011] Preferably, the data acquisition frequency is 0.25 hours.

[0012] Secondly, a parameter control system for a BiLSTM-based UV / H2O2 advanced oxidation device is provided, including: Edge systems include: The information acquisition module is used to collect influent and effluent data of the UV / H2O2 advanced oxidation unit in real time and upload them to the cloud system. The parameter prediction module is used to input the real-time collected influent and effluent data into the control parameter prediction model to predict the hydrogen peroxide dosage and ultraviolet power. The execution drive module adjusts the hydrogen peroxide dosage and ultraviolet power based on the predicted hydrogen peroxide dosage and ultraviolet power. Cloud systems, including: The Granger causality test module dynamically filters influent and effluent data types using the Granger causality test. The model training module is used to build a lightweight BiLSTM model. It uses the filtered influent data, effluent data and feedback features as input data, and the corresponding hydrogen peroxide dosage and ultraviolet power as label data to train the model and generate a control parameter prediction model. The model distribution module distributes the trained control parameter prediction model to the edge system.

[0013] Preferably, the data filtering module is integrated into the NVIDIA Jetson Nano chip and dynamically switches the BiLSTM input data channel through FPGA programmable logic.

[0014] Thirdly, a parameter control device for a BiLSTM-based UV / H2O2 advanced oxidation apparatus is provided, comprising a processor and a memory storing program instructions, wherein the processor is configured to execute the BiLSTM-based UV / H2O2 advanced oxidation apparatus parameter control method when running the program instructions.

[0015] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the aforementioned parameter control method for a BiLSTM-based UV / H2O2 advanced oxidation device.

[0016] The advantages of this invention are: This invention utilizes Granger spectroscopy training data and the BiLSTM machine learning algorithm to accurately predict and control the dosage of chemicals (such as hydrogen peroxide) and ultraviolet light intensity in UV / H2O2 advanced oxidation devices. This precise control effectively avoids the increased costs and equipment damage caused by excessive chemical dosing and increased ultraviolet power in traditional methods, significantly improving processing efficiency.

[0017] By optimizing the control of chemical dosage and ultraviolet light intensity, this invention effectively reduces the cost of chemical dosing and the energy consumption of ultraviolet lamps, extends the service life of equipment, and thus significantly saves overall operating costs. This has significant economic benefits for large-scale water treatment plants.

[0018] The system incorporates Granger causality tests for dynamic feature selection, quantifying causal relationships between variables and identifying the true driving factors influencing pollutant degradation. This feature selection mechanism avoids interference from redundant data, enhances the model's reliability and generalization ability, and enables the system to maintain stable treatment performance under different water quality conditions.

[0019] This invention constructs a closed-loop optimization mechanism. The edge controller drives the dosing pump and ultraviolet lamp based on the final parameters, while simultaneously feeding the execution results back to the digital platform. This closed-loop control enables the system to continuously collect operational data, constantly optimize the control strategy, and form a virtuous cycle of continuous improvement. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0021] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0022] Figure 2 This is a schematic diagram illustrating odorant prediction according to an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 like Figure 1 As shown, a parameter control method for a BiLSTM-based UV / H2O2 advanced oxidation device includes the following steps: S1: Construct an edge-cloud collaborative architecture, whereby the edge collects influent and effluent data from the UV / H2O2 advanced oxidation unit in real time and uploads them to the cloud; S2: The cloud uses Granger causality tests to filter influent and effluent data types; S3: Build a lightweight BiLSTM model, using the filtered influent data, effluent data and feedback features as input data, and the corresponding hydrogen peroxide dosage and ultraviolet power as label data, train the lightweight BiLSTM model, and obtain the control parameter prediction model. S4: The cloud sends the control parameter prediction model to the edge, and the edge collects influent and effluent data in real time and inputs them into the control parameter prediction model to predict the hydrogen peroxide dosage and ultraviolet power. S5: The edge end drives the UV / H2O2 advanced oxidation device to adjust the hydrogen peroxide dosage and UV power based on the predicted hydrogen peroxide dosage and UV power.

[0025] As a refinement of the above embodiments, the influent data includes influent flow rate, influent turbidity, influent conductivity, and influent UV. 254 Influent TOC, influent new pollutant concentration, influent temperature, influent pH, and influent hydrogen peroxide concentration; The effluent data includes effluent UV. 254 The following parameters are measured: effluent turbidity, effluent conductivity, effluent TOC (total organic matter), concentration of new pollutants in effluent, effluent temperature, effluent pH, effluent hydrogen peroxide concentration, and ultraviolet light intensity. The new pollutants include antibiotics, tetracyclines, and odorants.

[0026] As a refinement of the above embodiments, the feedback characteristic components are constructed through PID business logic and include proportional feedback characteristics, integral feedback characteristics, and derivative feedback characteristics. The expressions for the feedback characteristic components are as follows: , , , in, For the current predicted time, for Deviation in dosage over time , for The hydrogen peroxide dosage value at all times. Set the value for hydrogen peroxide. The current time-proportional feedback feature, The integral feedback feature at the current time. This represents the differential feedback characteristic at the current moment.

[0027] Feedback characteristic variables can reflect the hydrogen peroxide (H2O2) addition process in UV / H2O2 advanced oxidation devices, ensuring effluent water quality, enhancing effluent safety, forming closed-loop control, increasing the reliability and generalization ability of the model, and the proportional, integral and differential characteristics variables can better reflect the linear and nonlinear relationship between water quality variables and dosage, enhancing the nonlinear expression of the model.

[0028] As a refinement of the above embodiments, the Granger causality test filters the data types of influent and effluent, retaining data types with p-values ​​less than 0.05; Specifically, it includes: ADF stationarity test and differential transformation were performed on the 24-hour time series data windows of influent and effluent data. Selecting the optimal lag order for the VAR model based on the AIC / BIC criterion; Using pollutant concentration as the dependent variable, we tested the Granger causality of each parameter, retaining data types with p-values ​​less than 0.05.

[0029] Currently, existing methods rely on fixed input features (such as TOC, UV). 254 By leveraging the advantages of Granger criterion testing, the causal relationships (non-correlational relationships) between variables are quantified to identify the true driving factors affecting pollutant degradation (e.g., in the patent, 2-MIB is causally affected by UV power and hydrogen peroxide concentration, while temperature is not causally related). Dynamic feature screening improves model prediction accuracy, avoids redundant data interference, and reduces edge computing load.

[0030] The Granger causality test and feature selection are performed locally using an embedded AI chip (such as NVIDIA Jetson Nano) with a pre-installed causal analysis algorithm library (Python-based). The input feature dimensions are dynamically adjusted based on the selection results, and the FPGA programmable logic switches the data channel to the BiLSTM model in real time. Specifically, the code is as follows: defgranger_feature_selection(data_stream): #Step 1: Data Stationarity Test (ADF) stationary_data=adf_test(data_stream) #Step 2: Construct a VAR model and calculate the Granger causal F-statistic var_model=VAR(stationary_data) best_lag = select_optimal_lag(var_model) # Select the lag order using the AIC / BIC criterion #Step 3: Perform a causal test (p=0.05 significance level) causal_features=[] forfeatureinininput_features: granger_result=granger_causation_test(var_model, target="Pollutant_Conc", candidate=feature, lag=best_lag) if granger_result.p_value < 0.05: # Pass causality test causal_features.append(feature) `return causal_features` # Outputs a list of causal correlation features. #Real-time application: Update the feature set every 2 hours (balancing computational overhead and adaptability) whileTrue: current_features=granger_feature_selection(last_24h_data) update_bilstm_inputs(current_features) # Dynamically reconfigure the BiLSTM input layer sleep(2*60*60)# Sleep for 2 hours Because the interactions between the multidimensional water quality data in the dataset are very complex, and each parameter has a hidden and close relationship with the removal of micropollutants, appropriate data dimensionality reduction methods are selected for data feature selection before model training. The Granger causality test is used to select features for data on the removal of typical micropollutants by advanced oxidation processes.

[0031] In this embodiment, the data types described in the following tables were collected and Granger causality tests and screenings were performed. Taking odorants as an example, the specific data are shown in Tables 1 and 2 below.

[0032] Table 1. Stationarity Test Table Table 2. Granger causality test of each factor and odorant. As a refinement of the above embodiment, the BiLSTM neural network structure of the prediction model comprises two independent LSTM layers, one processing the forward sequence and the other processing the reverse sequence. Its output is a combination of the outputs of these two LSTM layers. Specifically, for the input sequence X=(x1,x2,…,x…) T X=(x1,x2,…,x) T The forward LSTM layer generates a forward hidden state sequence (h1→, h2→, ..., h...). T →)(h1,h2,…h T The inverse LSTM layer generates an inverse hidden state sequence (h1←,h2←,…,h…). T ←)(h1,h2,…,h T The final output of BiLSTM is the concatenation of these two hidden state sequences (h). t →,h t ←)(h t ,h t The LSTM neural network described includes an input layer, three LSTM cell layers, and an output layer. The LSTM cell layers contain a forget gate, an input gate, and an output gate.

[0033] The main structural parameters and network structure of the BiLSTM model used in this embodiment can be set as follows: 1 convolutional layer, 64 convolutional kernels, 3 pooling windows, 200 training epochs, a learning rate of 0.001, and a forgetting rate of 0.4. These parameters are chosen to minimize the computational cost and training time of the BiLSTM model while ensuring its accuracy.

[0034] Example 2 A parameter control system for a BiLSTM-based UV / H2O2 advanced oxidation device includes: Edge systems include: The information acquisition module is used to collect influent and effluent data of the UV / H2O2 advanced oxidation unit in real time and upload them to the cloud system. The parameter prediction module is used to input the real-time collected influent and effluent data into the control parameter prediction model to predict the hydrogen peroxide dosage and ultraviolet power. The execution drive module adjusts the hydrogen peroxide dosage and ultraviolet power based on the predicted hydrogen peroxide dosage and ultraviolet power. Cloud systems, including: The Granger causality test module dynamically filters influent and effluent data types using the Granger causality test. The model training module is used to build a lightweight BiLSTM model. It uses the filtered influent data, effluent data and feedback features as input data, and the corresponding hydrogen peroxide dosage and ultraviolet power as label data to train the model and generate a control parameter prediction model. The model distribution module distributes the trained control parameter prediction model to the edge system.

[0035] As a refinement of the above embodiment, the information acquisition module acquires 20+ dimension raw data at a frequency of 0.25h; the edge caches a 24-hour time series data window (to meet the minimum sample size for Granger test).

[0036] The data filtering module is integrated into the NVIDIA Jetson Nano chip and dynamically switches the BiLSTM input data channel through FPGA programmable logic.

[0037] This disclosure also provides a parameter control device for a BiLSTM-based UV / H2O2 advanced oxidation apparatus, including a processor and a memory. Optionally, the device may further include a communication interface and a bus. The processor, communication interface, and memory can communicate with each other via the bus. The communication interface can be used for information transmission. The processor can call logical instructions in the memory to execute the BiLSTM-based UV / H2O2 advanced oxidation apparatus parameter control method of the above embodiments.

[0038] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0039] Memory, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as the program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor executes the program instructions / modules stored in the memory to perform functional applications and data processing, thereby implementing the parameter control method for the BiLSTM-based UV / H2O2 advanced oxidation device in the above embodiments.

[0040] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory may include high-speed random access memory and may also include non-volatile memory.

[0041] This disclosure provides a computer-readable storage medium storing computer-executable instructions configured to execute the above-described parameter control method for a BiLSTM-based UV / H2O2 advanced oxidation device.

[0042] The aforementioned computer-readable storage medium may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0043] The technical solutions of this disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this disclosure. The aforementioned storage medium can be a non-transitory storage medium, including: a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and other media capable of storing program code; it can also be a transient storage medium.

[0044] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A parameter control method for a BiLSTM-based UV / H2O2 advanced oxidation device, characterized in that, Includes the following steps: An edge-cloud collaborative architecture is constructed, in which the influent and effluent data of the UV / H2O2 advanced oxidation unit are collected in real time at the edge and uploaded to the cloud; The cloud-based system uses Granger causality tests to filter influent and effluent data types. A lightweight BiLSTM model was built, using the filtered influent data, effluent data and feedback features as input data, and the corresponding hydrogen peroxide dosage and ultraviolet power as label data. The lightweight BiLSTM model was trained to obtain the control parameter prediction model. The cloud sends the control parameter prediction model to the edge device, and the edge device collects influent and effluent data in real time and inputs them into the control parameter prediction model to predict the hydrogen peroxide dosage and ultraviolet power. The edge end drives the UV / H2O2 advanced oxidation unit to adjust the hydrogen peroxide dosage and UV power based on the predicted hydrogen peroxide dosage and UV power.

2. The parameter control method for a BiLSTM-based UV / H2O2 advanced oxidation device according to claim 1, characterized in that, The influent data includes influent flow rate, influent turbidity, influent conductivity, and influent UV. 254 Influent TOC, influent new pollutant concentration, influent temperature, influent pH, and influent hydrogen peroxide concentration; The effluent data includes effluent UV. 254 Effluent turbidity, effluent conductivity, effluent TOC, concentration of new pollutants in effluent, effluent temperature, effluent pH, effluent hydrogen peroxide concentration, and ultraviolet light intensity; The new pollutants include antibiotics, tetracyclines, and odorants.

3. The parameter control method for a BiLSTM-based UV / H2O2 advanced oxidation device according to claim 1, characterized in that, The feedback characteristic components are constructed through PID business logic and include proportional feedback characteristics, integral feedback characteristics, and derivative feedback characteristics. The expressions for the feedback characteristic components are as follows: ， ， ， in, For the current predicted time, for Deviation in dosage over time , for The hydrogen peroxide dosage value at all times Set the value for hydrogen peroxide. The current time-proportional feedback feature, The integral feedback feature at the current time. This represents the differential feedback characteristic at the current moment.

4. The parameter control method for a BiLSTM-based UV / H2O2 advanced oxidation device according to claim 1, characterized in that, Granger causality test was used to filter influent and effluent data types, retaining data types with p-values ​​less than 0.05; Specifically, it includes: ADF stationarity test and differential transformation were performed on the 24-hour time series data windows of influent and effluent data. Selecting the optimal lag order for the VAR model based on the AIC / BIC criterion; Using pollutant concentration as the dependent variable, we tested the Granger causality of each parameter, retaining data types with p-values ​​less than 0.

05.

5. The parameter control method for a BiLSTM-based UV / H2O2 advanced oxidation device according to claim 1, characterized in that, The lightweight BiLSTM model consists of two independent LSTM layers, one for processing forward sequences and the other for processing reverse sequences. Each LSTM layer includes an input layer, three LSTM cell layers, and an output layer. Each LSTM cell layer contains a forget gate, an input gate, and an output gate.

6. The parameter control method for a BiLSTM-based UV / H2O2 advanced oxidation device according to claim 1, characterized in that, The data collection frequency is 0.25 hours.

7. A parameter control system for a BiLSTM-based UV / H2O2 advanced oxidation device, characterized in that, The method for controlling parameters of a BiLSTM-based UV / H2O2 advanced oxidation device as described in any one of claims 1-6 includes: Edge systems include: The information acquisition module is used to collect influent and effluent data of the UV / H2O2 advanced oxidation unit in real time and upload them to the cloud system. The parameter prediction module is used to input the real-time collected influent and effluent data into the control parameter prediction model to predict the hydrogen peroxide dosage and ultraviolet power. The execution drive module adjusts the hydrogen peroxide dosage and ultraviolet power based on the predicted hydrogen peroxide dosage and ultraviolet power. Cloud systems, including: The Granger causality test module dynamically filters influent and effluent data types using the Granger causality test. The model training module is used to build a lightweight BiLSTM model. It uses the filtered influent data, effluent data and feedback features as input data, and the corresponding hydrogen peroxide dosage and ultraviolet power as label data to train the model and generate a control parameter prediction model. The model distribution module distributes the trained control parameter prediction model to the edge system.

8. The parameter control system for the UV / H2O2 advanced oxidation device based on BiLSTM according to claim 7, characterized in that, The data filtering module is integrated into the NVIDIA Jetson Nano chip and dynamically switches the BiLSTM input data channel through FPGA programmable logic.

9. A parameter control device for a BiLSTM-based UV / H2O2 advanced oxidation apparatus, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to execute, when running the program instructions, the parameter control method for a BiLSTM-based UV / H2O2 advanced oxidation device as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the parameter control method for the BiLSTM-based UV / H2O2 advanced oxidation device as described in any one of claims 1-6 above.

Images