Method and system for treating organic wastewater

Abstract

The application relates to the technical field of wastewater treatment, and particularly discloses a treatment method and system for organic wastewater, which utilizes layered B-phase vanadium dioxide as an efficient Fenton-like catalyst to degrade pollutants in the organic wastewater under normal temperature stirring conditions in cooperation with persulfate, and further introduces a deep learning algorithm in the oxidation-reduction reaction process of the organic wastewater, extracts time sequence characteristic representations of key reaction parameters in the reaction system by monitoring high-frequency data streams of multiple sources such as oxidation-reduction potential, PH, temperature, dissolved oxygen and conductivity in the reactor, and simultaneously combines the current reaction time to construct a reaction progress prediction model, so that intelligent judgment and intervention warning of the reaction endpoint in the organic wastewater treatment process are realized. In this way, the problems of insufficient reaction or over-reaction can be effectively avoided, and the treatment effect and resource utilization rate of the organic wastewater are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wastewater treatment, and more particularly, to a treatment method and system for organic wastewater. BACKGROUND

[0002] Fenton and Fenton-like methods have become one of the research hotspots for organic wastewater treatment due to their simple operation, fast degradation speed, and economic and environmental protection. Among them, heterogeneous catalytic systems have attracted much attention due to their easy recovery of catalysts, wide pH range, and other characteristics.

[0003] For example, the invention patent with publication number CN114656024A proposes a treatment method for organic wastewater, which uses vanadium dioxide as a Fenton-like catalyst and adds it together with persulfate into the organic wastewater to remove organic pollutants under specific conditions. However, such chemical oxidation treatment process often relies on empirical reaction time setting or offline sampling analysis to determine the reaction endpoint, the former may lead to insufficient reaction or over-reaction, resulting in poor treatment effect or resource waste; the latter has problems such as poor real-time performance, long analysis period, high labor cost, etc., making it difficult to achieve precise control and dynamic optimization of the complex reaction process. That is, for the treatment of organic wastewater with complex reaction mechanism, the lack of real-time and intelligent reaction endpoint judgment and process control means is the main defect.

[0004] Therefore, an optimized treatment method and system for organic wastewater are expected. SUMMARY

[0005] To solve the above technical problems, the present application is proposed. The embodiments of the present application provide a treatment method and system for organic wastewater, which uses layered B-phase vanadium dioxide as an efficient Fenton-like catalyst to degrade pollutants in organic wastewater in conjunction with persulfate under normal temperature stirring conditions, and further introduces a deep learning algorithm in the redox reaction process of organic wastewater, extracts the time sequence feature representation of key reaction parameters in the reaction system by real-time monitoring of the high-frequency data stream of multiple sensors such as oxidation-reduction potential, pH, temperature, dissolved oxygen, and conductivity in the reactor, and constructs a reaction progress prediction model by combining the current reaction time, to realize intelligent judgment and intervention warning of the reaction endpoint in the organic wastewater treatment process. In this way, the problems of insufficient reaction or over-reaction can be effectively avoided, and the treatment effect and resource utilization rate of organic wastewater can be improved.

[0006] Correspondingly, according to one aspect of the present application, a treatment method for organic wastewater is provided, which comprises:

[0007] Vanadium dioxide is used as a Fenton-like catalyst, which is added to organic wastewater together with persulfate to remove organic pollutants under the condition of normal temperature and stirring, wherein the vanadium dioxide is layered B-phase vanadium dioxide.

[0008] During the reaction, real-time high-frequency data streams from each sensor in the reactor are obtained in real time, including redox potential data stream, PH data stream, temperature data stream, dissolved oxygen data stream and conductivity data stream.

[0009] The current reaction time and the real-time high-frequency data stream are input into the trained machine learning model to obtain the probability of the reaction reaching the end point at the current time.

[0010] In response to the probability of the reaction reaching the end point at the current time exceeding a preset threshold, a production intervention prompt signal is generated.

[0011] According to another aspect of the present application, an organic wastewater treatment system is provided, which comprises:

[0012] A catalytic reaction module is used to use vanadium dioxide as a Fenton-like catalyst, which is added to organic wastewater together with persulfate to remove organic pollutants under the condition of normal temperature and stirring, wherein the vanadium dioxide is layered B-phase vanadium dioxide.

[0013] A reaction monitoring module is used to obtain real-time high-frequency data streams from each sensor in the reactor in real time during the reaction, including redox potential data stream, PH data stream, temperature data stream, dissolved oxygen data stream and conductivity data stream.

[0014] A reaction end point prediction module is used to input the current reaction time and the real-time high-frequency data stream into the trained machine learning model to obtain the probability of the reaction reaching the end point at the current time.

[0015] A production intervention prompt module is used to generate a production intervention prompt signal in response to the probability of the reaction reaching the end point at the current time exceeding a preset threshold.

[0016] Compared with the prior art, the organic wastewater treatment method and system provided by the application utilize layered B-phase vanadium dioxide as an efficient Fenton-like catalyst to degrade pollutants in organic wastewater under normal temperature stirring conditions in cooperation with persulfate, and further introduce a deep learning algorithm in the redox reaction process of the organic wastewater, extract the time sequence feature representation of key reaction parameters in the reaction system by monitoring the high-frequency data flow of multiple sensors such as the oxidation-reduction potential, pH, temperature, dissolved oxygen and conductivity in the reactor in real time, and construct a reaction progress prediction model by combining the current reaction time, so as to realize intelligent judgment and intervention warning of the reaction endpoint in the organic wastewater treatment process. In this way, the problems of insufficient reaction or over-reaction can be effectively avoided, and the treatment effect and resource utilization rate of the organic wastewater are improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application taken in conjunction with the accompanying drawings. The drawings provided in the present application are used to provide further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 Flowchart of the organic wastewater treatment method according to the embodiments of the present application.

[0019] Figure 2 Data flow diagram of the organic wastewater treatment method according to the embodiments of the present application.

[0020] Figure 3 Flowchart of step S3 in the organic wastewater treatment method according to the embodiments of the present application.

[0021] Figure 4 Flowchart of step S32 in the organic wastewater treatment method according to the embodiments of the present application.

[0022] Figure 5 Block diagram of the organic wastewater treatment system according to the embodiments of the present application. DETAILED DESCRIPTION

[0023] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein. It is worth noting that in the present application, all actions of obtaining data are carried out in compliance with the corresponding data protection regulations and policies of the place, and with the authorization given by the corresponding device owner.

[0024] Figure 1 A flow chart of the method for treating organic wastewater according to an embodiment of the present application. Figure 2 A data flow schematic diagram of the method for treating organic wastewater according to an embodiment of the present application. As shown in Figure 1 and Figure 2 The method for treating organic wastewater according to an embodiment of the present application includes the following steps: S1, vanadium dioxide is used as a Fenton-like catalyst, which is added to organic wastewater together with persulfate for reaction at normal temperature under stirring to remove organic pollutants, wherein the vanadium dioxide is layered B-phase vanadium dioxide; S2, during the reaction, a current reaction time and real-time high-frequency data streams from sensors in the reactor are obtained in real time, the real-time high-frequency data streams including redox potential data streams, PH data streams, temperature data streams, dissolved oxygen data streams and conductivity data streams; S3, the current reaction time and the real-time high-frequency data streams are input into a trained machine learning model to obtain a probability that the reaction reaches an end point at a current time; S4, in response to the probability that the reaction reaches the end point at the current time exceeding a preset threshold, a production intervention prompt signal is generated.

[0025] In the above method for treating organic wastewater, in step S1, vanadium dioxide is used as a Fenton-like catalyst, which is added to the organic wastewater together with persulfate, and the reaction is carried out at room temperature under stirring to remove organic pollutants. The vanadium dioxide used herein is layered B-phase vanadium dioxide. It should be understood that layered B-phase vanadium dioxide has a unique layered structure and chemical reactivity, and can efficiently activate persulfate to generate oxidizing substances such as hydroxyl radicals, effectively degrade organic pollutants, and is environmentally friendly and reusable. Therefore, based on the activation mechanism of persulfate in advanced oxidation, vanadium dioxide is used as a Fenton-like catalyst to synergize with persulfate, and its catalytic performance is used to realize the oxidative decomposition of organic pollutants. In the specific implementation process, the pre-prepared layered B-phase vanadium dioxide powder is added to the reactor containing the organic wastewater to be treated at the same time or sequentially with a certain amount of persulfate (such as sodium persulfate, potassium persulfate, etc.). Under the condition of room temperature (for example, room temperature 20-30°C), the catalyst, oxidant and pollutant molecules are fully contacted by mechanical stirring (such as magnetic stirring, paddle stirring) for sufficient mixing and mass transfer. In this process, the unique layered structure of layered B-phase vanadium dioxide and the electronic properties of multivalent vanadium can effectively activate persulfate ions or peroxymonosulfate ions, promote their decomposition to generate sulfate radicals, hydroxyl radicals and other active radicals with strong oxidizing properties. These radicals can non-selectively attack organic pollutant molecules, and through electron transfer, addition, substitution and other reaction pathways, they are gradually oxidized and decomposed into small molecular inorganic substances (such as carbon dioxide, water) or intermediate products with lower toxicity and easy biodegradation. In this way, the target pollutants in the organic wastewater can be effectively removed, and the chemical oxygen demand (COD), biochemical oxygen demand (BOD) and toxicity of the wastewater can be reduced. At the same time, layered B-phase vanadium dioxide as a heterogeneous catalyst shows good catalytic activity and potential recycling value.

[0026] In the method for treating organic wastewater, the step S2, during the reaction, real-time high-frequency data streams from the current reaction time and various sensors in the reactor are obtained, including redox potential data stream, pH data stream, temperature data stream, dissolved oxygen data stream and conductivity data stream. It should be understood that the chemical oxidation reaction process of organic wastewater is complex and dynamic, and depends on the experience to set the reaction time or to judge the reaction endpoint by offline sampling analysis. Not only lack of real-time and accuracy, difficult to adapt to the impact of wastewater component fluctuation, but also may lead to insufficient reaction and cause the treatment effect not up to standard, or excessive reaction and cause waste of energy and chemical agents, and increase the operation cost. Therefore, in order to realize real-time and accurate monitoring of the reaction progress, the present application is based on multi-parameter sensing monitoring technology, and various sensors are arranged in the reactor to realize comprehensive collection of reaction process data by cooperating with a timing system. Specifically, in the organic wastewater treatment reactor or the circulating pipeline, targeted online sensors are installed, including oxidation reduction potential (ORP) probe, pH electrode, temperature sensor, dissolved oxygen (DO) sensor and conductivity sensor. These sensors are connected with a data acquisition system (such as PLC, DAQ card or integrated water quality analyzer), and from the beginning of the reaction (t=0), the instantaneous readings of various parameters are continuously collected at a set high frequency (for example, every second, every several seconds or higher frequency), and the time of the current reaction is recorded. Among them, the oxidation reduction potential reflects the overall change of the oxidation reduction intensity of the system, and tends to be stable at the reaction endpoint; the change of pH value indicates the movement of acid-base balance and the generation and consumption of intermediate products; the change of temperature can reflect the heat release and absorption characteristics and reaction rate of the reaction; the change of dissolved oxygen concentration is related to some oxidation pathways or side reactions; and the conductivity can reflect the change of total ion concentration in the system. The dynamic evolution of the above-mentioned various parameters together reflects the chemical reaction process and pollutant degradation state in the organic wastewater treatment process, and is an important basis for understanding the reaction progress. Therefore, through real-time monitoring of the above high-frequency data stream, comprehensive data basis can be provided for subsequent reaction progress prediction.

[0027] In the implementation process, to obtain real-time data of the current reaction time and high-frequency data stream from various sensors in the reactor, the appropriate sensor type needs to be selected for different monitoring requirements in terms of hardware selection and deployment. The oxidation-reduction potential (ORP) sensor should use a composite electrode with a pollution-resistant membrane structure, such as a platinum-gold electrode combined with a reference electrode, to reduce the adsorption of organic pollutants on the electrode surface. The measurement range should cover -2000mV to +2000mV, with an accuracy of within ±5mV, to ensure that the subtle changes in the oxidation-reduction reaction intensity in the reaction system can be captured. The pH sensor uses a composite sensor integrated with a glass electrode and a reference electrode, with a gel electrolyte type that can withstand strong acids and strong bases to avoid the risk of leakage of traditional liquid electrolytes in wastewater treatment scenarios. The measurement range is set to 0-14, with an accuracy of ±0.05pH, to reflect the dynamic changes in hydrogen ion concentration in real time during the reaction process. The temperature sensor preferably uses a platinum resistance (Pt100) sensor with a three-wire connection to eliminate the effects of wire resistance, with a measurement range of 0-100°C and an accuracy of ±0.5°C. Considering the possible temperature gradient in the reactor, 2-3 temperature sensors can be deployed at different heights in the reactor to improve the reliability of temperature data through mean value calculation. Temperature is a key parameter for reaction kinetics, and its accurate monitoring is crucial for evaluating reaction rate. The dissolved oxygen (DO) sensor can use a fluorescence method dissolved oxygen electrode, which is less affected by reducing substances such as sulfides in water and has a shorter response time than traditional polarographic methods. The measurement range is 0-20mg / L, with an accuracy of ±0.1mg / L. The flow cell design allows water samples to continuously flow over the electrode surface, avoiding the influence of bubble attachment on measurement accuracy. Changes in dissolved oxygen content can reflect the degree of oxidation reaction consumption and the oxygen transfer efficiency in the system. The conductivity sensor selects a four-loop electrode structure to reduce electrode polarization effects, with a measurement range of 0-200mS / cm and an accuracy of ±1%. Considering the high electrolyte concentration in wastewater, the sensor needs to have an automatic range switching function. Changes in conductivity are closely related to the decomposition of organic matter, ion generation, and other processes during the reaction process, making it one of the important indicators of reaction progress.

[0028] The installation layout of the sensors needs to be optimized according to the structure of the reactor. For a cylindrical reactor, the redox potential sensor, pH sensor, and conductivity sensor can be uniformly arranged along the circumferential direction at a height of 1 / 3 of the reactor from the bottom, with the electrodes immersed at least 10 cm below the liquid level to avoid exposure of the electrodes to air due to liquid level fluctuations. The temperature sensor can be inserted into the center of the reactor to a depth of 1 / 2 of the reactor radius to measure the temperature of the bulk solution. The dissolved oxygen sensor is installed in a turbulent region of the flow field near the stirring paddle to ensure sufficient mixing of the water sample and improve measurement representativeness. The installation interface of all sensors needs to use a flange sealing structure to prevent wastewater leakage, and the sensor probe needs to be equipped with a quickly detachable protective sleeve for regular calibration and maintenance.

[0029] At the same time, the construction of the data acquisition system needs to meet the requirements of high-frequency sampling and synchronization. A multi-channel data acquisition card (DAQ) is used as the core acquisition device, such as the USB-6366 model from NI Company, which has 8 analog input channels with a maximum sampling rate of 1.25MS / s, which can meet the high-frequency acquisition requirements of 5 sensors (each channel sampling rate 10Hz). To avoid cross-talk between sensor signals, each channel needs to be equipped with an independent signal conditioning circuit, including an anti-aliasing filter (cut-off frequency set to 5Hz, satisfying the Nyquist sampling theorem), and a signal amplification circuit (gain adjusted according to the sensor output signal range, such as amplifying 0-5V signal to the input range of the acquisition card). The data acquisition software can be developed based on the LabVIEW platform, setting a timing sampling task to ensure synchronous acquisition of data from each channel through hardware triggering, with a uniform sampling frequency of 10Hz, i.e. 10 sets of data collected per second. Such high-frequency sampling can capture transient changes during the reaction process.

[0030] In addition, time synchronization mechanism is the key to ensure the consistency of data timing. A GPS clock synchronization device is deployed on site in the reactor, connected to the data acquisition card through an RS-485 interface, and calibrated every hour to control the error within ±10ms. At the same time, a timestamp is added to each set of collected data in the data acquisition software, accurate to the millisecond level, with a format of "YYYY-MM-DD HH:MM:SS:ms", facilitating subsequent alignment and analysis of time series data for different parameters. For multiple reactors deployed in a distributed manner, the IEEE 1588 Precision Time Protocol (PTP) can be used to synchronize the time of each acquisition node through a switch, ensuring the time consistency of cross-reactor data.

[0031] On the other hand, the data transmission and storage architecture needs to balance real-time and reliability. The data collected on site is transmitted to the data server in the control room through an industrial Ethernet switch (such as Hirschmann RS20-0802M2T), and data interaction is carried out using the OPC UA (Open Platform Communications Unified Architecture) protocol, which has cross-platform and secure encryption features to ensure data integrity and security during transmission. The transmission link uses a redundant design, with a main link for fiber transmission and a backup link for twisted pair. When the main link fails, it automatically switches to the backup link, with a switching time of less than 50ms. The data server uses a dual-hardware hot standby architecture, with a Windows Server 2019 operating system, and a time series database (such as InfluxDB) is deployed. This database is optimized for time series data and supports fast writing and querying of high-frequency data, and can store at least 6 months of historical data. When storing data, each sensor's data is stored in separate tables, with table structures including timestamp, parameter value, and quality flag (to identify whether the data is valid), and a time index is established to improve query efficiency.

[0032] To ensure data quality, a perfect abnormality handling mechanism is also needed. At the hardware level, each sensor needs to be equipped with a self-diagnosis function to monitor the electrode state (such as the zero drift of the oxidation-reduction potential sensor and the slope change of the pH sensor) in real time. When a sensor failure is detected, an error code is sent to the data acquisition system and an alarm is triggered. At the software level, a data validity checking module is developed, using a combination of threshold judgment and trend analysis: for oxidation-reduction potential, the normal range is set to -1000mV to +1500mV, and data outside this range is marked as invalid; for pH, the normal range is set to 2-12, and data outside this range is marked as invalid; for temperature data, if the change exceeds 5°C within 1 minute, it is judged as abnormal data. For invalid data, the system automatically performs interpolation to ensure the continuity of the data sequence. At the same time, the sensors are calibrated regularly: the oxidation-reduction potential sensor is calibrated once a week with a standard solution (such as quinhydrone solution), the pH sensor is calibrated with pH=4.00, pH=7.00, and pH=10.00 buffer solutions, the temperature sensor is calibrated monthly with a standard constant temperature water bath, the dissolved oxygen sensor is calibrated every two weeks with a zero point and full scale, and the conductivity sensor is calibrated with a standard potassium chloride solution. The calibration records need to be saved completely for traceability.

[0033] In addition, to realize real-time visualization and interaction of data, a Web-based monitoring interface can be developed using HTML5+JavaScript technology, which communicates with the data server in real time through the WebSocket protocol to display the real-time change trend of each parameter in the form of dynamic curves, supports zooming, panning and other interactive operations, and facilitates operators to master the reaction state in real time. The interface sets data query function, which can retrieve historical data according to time interval, generate report and export, providing data support for process optimization. At the same time, set the early warning threshold, when the change rate of a parameter exceeds the preset value (such as the change of oxidation-reduction potential per minute exceeds 50 mV, the change of pH value per minute exceeds 0.5), the system automatically sends sound and light warning, prompting the operator to pay attention to the abnormal reaction.

[0034] In the system integration and debugging stage, full-link performance testing is required. Simulate the reactor operating state, generate standard signals to simulate the output of each sensor through the signal generator, test the integrity and real-time performance of data acquisition, transmission and storage, and ensure that the delay time from data acquisition to interface display is less than 500 ms. In actual application, first carry out small-scale debugging, deploy sensors in laboratory-scale reactors, and compare with offline analysis data (such as organic matter concentration determined by high-performance liquid chromatography), optimize data acquisition frequency and calibration period, and ensure the consistency of online monitoring data and offline analysis results, with an error of less than ±5%.

[0035] In the above method for treating organic wastewater, the step S3 inputs the current reaction time and the real-time high-frequency data stream into the trained machine learning model to obtain the probability of the reaction reaching the end point at the current time. That is, in order to realize real-time online intelligent judgment of the reaction end point in the oxidation-reduction reaction process of organic wastewater, the present application further introduces a deep learning algorithm, uses a machine learning model trained by a large amount of historical reaction data to learn and identify the time sequence evolution law of each key parameter in the reaction process and the potential correlation with the target reaction end point, thereby realizing accurate prediction of the current reaction progress.

[0036] Figure 3 The flow chart of step S3 in the method for treating organic wastewater according to the embodiments of the present application. As Figure 3As shown, the step S3 comprises: S31, extracting time sequence features of each data in the real-time high-frequency data stream respectively to obtain a reaction progress ORP end time sequence feature representation, a reaction progress PH end time sequence feature representation, a reaction progress temperature end time sequence feature representation, a reaction progress dissolved oxygen end time sequence feature representation and a reaction progress conductivity end time sequence feature representation; S32, performing time sequence fine-grained enhancement coding on the reaction progress ORP end time sequence feature representation, the reaction progress PH end time sequence feature representation, the reaction progress temperature end time sequence feature representation, the reaction progress dissolved oxygen end time sequence feature representation and the reaction progress conductivity end time sequence feature representation respectively to obtain an enhanced reaction progress ORP end time sequence feature representation, an enhanced reaction progress PH end time sequence feature representation, an enhanced reaction progress temperature end time sequence feature representation, an enhanced reaction progress dissolved oxygen end time sequence feature representation and an enhanced reaction progress conductivity end time sequence feature representation; S33, fusing the enhanced reaction progress ORP end time sequence feature representation, the enhanced reaction progress PH end time sequence feature representation, the enhanced reaction progress temperature end time sequence feature representation, the enhanced reaction progress dissolved oxygen end time sequence feature representation and the enhanced reaction progress conductivity end time sequence feature representation to obtain a reaction progress multi-dimensional data time sequence feature joint representation; S34, inputting the current reaction time and the reaction progress multi-dimensional data time sequence feature joint representation into the trained machine learning model to obtain a prediction result, the prediction result being a probability of the reaction reaching an end point at a current time.

[0037] Specifically, the step S31 extracts the time sequence features of each data in the real-time high-frequency data stream respectively to obtain a reaction progress ORP end time sequence feature representation, a reaction progress PH end time sequence feature representation, a reaction progress temperature end time sequence feature representation, a reaction progress dissolved oxygen end time sequence feature representation, and a reaction progress conductivity end time sequence feature representation. In a specific example of the present application, 1D-CNN-based time sequence analysis is performed on the real-time high-frequency data stream respectively to obtain the reaction progress ORP end time sequence feature representation, the reaction progress PH end time sequence feature representation, the reaction progress temperature end time sequence feature representation, the reaction progress dissolved oxygen end time sequence feature representation, and the reaction progress conductivity end time sequence feature representation. It can be understood that, since the original real-time high-frequency sensor data stream contains a large amount of original point information, there is noise interference and information redundancy, and it is difficult to directly reveal the deep time sequence trend changes of each parameter. Therefore, the present application further utilizes the powerful feature extraction capability of the convolutional neural network (CNN), and by performing one-dimensional convolution coding along the time dimension on each sensor data stream, the key time sequence patterns of each key parameter are extracted, the noise interference and information redundancy are reduced, and the deep laws of each parameter changing with time are captured. Specifically, the 1D-CNN can automatically learn and identify the local correlation patterns in the data sequence, such as rising / falling trend, peak, valley, plateau, and change rate, by sliding the filter (convolution kernel) in the convolution layer along the time dimension of each high-frequency data stream. The subsequent pooling layer (such as maximum pooling or average pooling) helps to reduce the feature dimension, retain the most significant feature information, and enhance the translational invariance of the model. After multiple convolution and pooling operations, a series of fixed-dimension abstract time sequence feature encoding vectors are finally outputted as the time sequence feature representations of each reaction parameter, i.e., the reaction progress ORP end time sequence feature representation, the reaction progress PH end time sequence feature representation, the reaction progress temperature end time sequence feature representation, the reaction progress dissolved oxygen end time sequence feature representation, and the reaction progress conductivity end time sequence feature representation. In this way, the key dynamic information of each parameter evolving with reaction time is effectively captured, and the data amount and computational complexity are greatly reduced, providing an efficient and refined feature input for subsequent current reaction progress prediction.

[0038] Specifically, the step S32, the reaction progress ORP end time sequence feature representation, the reaction progress PH end time sequence feature representation, the reaction progress temperature end time sequence feature representation, the reaction progress dissolved oxygen end time sequence feature representation and the reaction progress conductivity end time sequence feature representation are respectively time sequence fine-grained enhancement coded to obtain an enhanced reaction progress ORP end time sequence feature representation, an enhanced reaction progress PH end time sequence feature representation, an enhanced reaction progress temperature end time sequence feature representation, an enhanced reaction progress dissolved oxygen end time sequence feature representation and an enhanced reaction progress conductivity end time sequence feature representation. It can be understood that the time sequence feature representation of each sensor parameter obtained by the 1D-CNN network fuses the overall time sequence information of the high-frequency data stream, but the macroscopic features may mask the subtle changes, long-term dependence relationships or key events at specific time points in the reaction process that are crucial for judging the reaction endpoint, such as short-term rapid jumps, plateau formation or slight fluctuations in PH value. Therefore, in order to deeply mine the details and local dynamics of each sensor parameter and improve the sensitivity to the key points of the reaction process, the application further introduces a time sequence fine-grained enhancement coding technology to perform feature enhancement processing on the time sequence feature representation of each sensor parameter. Next, the reaction progress PH end time sequence feature representation is taken as an example for detailed description.

[0039] Figure 4 The flowchart of step S32 in the organic wastewater treatment method according to the embodiment of the application. As shown in the figure, Figure 4 S321, the reaction progress PH end time sequence feature representation is content-aware to obtain a reaction progress PH end local time sequence segment content-aware coding feature map; S322, the reaction progress PH end local time sequence segment content-aware coding feature map is input into a spatial attention layer to obtain a reaction progress PH end local time sequence segment content spatial saliency-aware coding feature map; S323, the reaction progress PH end local time sequence segment content spatial saliency-aware coding feature map is feature shape reshaped to obtain an enhanced reaction progress PH end time sequence feature vector as the enhanced reaction progress PH end time sequence feature representation.

[0040] In one specific example of the present application, the step S321 comprises: first, based on a preset window, the reaction progress PH end timing feature representation is feature segment divided to obtain a set of reaction progress PH end local timing feature vectors. It should be understood that, due to the stage characteristics of the change of PH value in the wastewater treatment process (such as the sudden drop of PH value at the beginning of the reaction due to the generation of free radicals, the platform period formed due to the action of the buffer system in the middle period, and the slow rise in the later period due to the accumulation of by-products), the global reaction progress PH end timing feature representation cannot deeply describe the key local events. Therefore, based on the sliding window mechanism, the reaction progress PH end timing feature representation is decomposed into local fragments with context association to form a set of reaction progress PH end local timing feature vectors, thereby providing a basic structural unit for subsequent fine analysis, which is conducive to focusing on the details of PH change in different reaction stages in subsequent analysis.

[0041] Then, after the set of reaction progress PH end local timing feature vectors is reshaped into a reaction progress PH end local timing segment encoding feature map, content perception based on convolution coding is performed to obtain a reaction progress PH end local timing segment content perception encoding feature map, which is expressed by the formula:

[0042]

[0043]

[0044] wherein, represents the set of reaction progress PH end local timing feature vectors, , , and represents the first, second, third and fourth reaction progress PH end local timing feature vectors in the set , is the number of vectors in the set of reaction progress PH end local timing feature vectors, represents feature shape reshaping, represents convolution coding operation, represents the reaction progress PH end local timing segment content perception encoding feature map.

[0045] ​​That is, in order to effectively model the complex dependency relationship inside the reaction progress PH end local timing feature vector (such as the nonlinear association between the PH crash starting point and the valley), the application further reshapes the set feature shape of the reaction progress PH end local timing feature vector into a highly structured reaction progress PH end local timing segment encoding feature map to capture the local dependency relationship and context information between each reaction progress PH end local timing feature vector by using the convolutional neural network good at capturing local association patterns, thereby obtaining a reaction progress PH end local timing segment content-aware encoding feature map.

[0046] In a specific example of the present application, the step S322 comprises: first, calculating a dynamic convolution kernel parameter based on the reaction progress PH end local timing segment content-aware encoding feature map. More specifically, the reaction progress PH end local timing segment content-aware encoding feature map is globally averaged along the channel dimension to obtain a reaction progress PH end local timing segment content-aware encoding vector, and the reaction progress PH end local timing segment content-aware encoding vector is input into a multi-layer perceptron model to obtain the dynamic convolution kernel parameter, which is expressed by the formula:

[0047]

[0048]

[0049] wherein, represents a global average pooling operation, represents a reaction progress PH end local timing segment content-aware encoding vector, is a multi-layer perceptron model, and respectively represent the weight matrix and bias vector of the first layer of the multi-layer perceptron model, represents a ReLU activation function, represents a hidden state feature vector of the first layer of the multi-layer perceptron model, and respectively represent the weight matrix and bias vector of the second layer of the multi-layer perceptron model, represents a hidden state feature vector of the second layer of the multi-layer perceptron model, represents a dynamic convolution kernel parameter.

[0050] Here, considering that the pH change patterns at different reaction stages are of varying importance in determining the reaction endpoint—for example, the starting point, trough, and duration of the initial pH drop directly affect the efficiency of free radical generation, while the stability of the plateau phase in the middle stage reflects the effectiveness of the buffer system—these key pH change patterns are crucial for accurately predicting the reaction endpoint. Therefore, this application further introduces a spatial attention mechanism to highlight key local pH change patterns and suppress non-critical information. Specifically, during spatial attention calculation, considering that fixed convolutional kernel parameters lack adaptability to the temporal change patterns of different sensor data, it may lead to insufficient extraction of key information. Therefore, to improve the sensitivity and adaptability to the temporal change patterns of different sensor data, this application further dynamically generates dedicated convolutional kernel parameters based on the content characteristics of the local temporal segment at the current reaction progress pH end, to achieve accurate capture of salient features in the content-aware encoding feature map of the local temporal segment at the current reaction progress pH end, thereby improving the accuracy of subsequent spatial attention calculations.

[0051] In particular, in a preferred embodiment of this application, the input of the reaction progress PH local temporal segment content-aware encoding vector into a multilayer perceptron model to obtain the dynamic convolution kernel parameters includes: first, performing iterative optimization on the bias vector of the multilayer perceptron model based on parameter space phase stability preservation to obtain an optimized bias vector; then, based on the optimized bias vector, using the multilayer perceptron model to process the reaction progress PH local temporal segment content-aware encoding vector to obtain the dynamic convolution kernel parameters.

[0052] It should be understandable that in generating dynamic convolution kernel parameters... During the process, global mean pooling is performed on the channel dimension of the local temporal segment content-aware encoding feature map of the reaction progress PH terminal to obtain the local temporal segment content-aware encoding vector of the reaction progress PH terminal. Then, using the weight matrix... and and bias vector and For vectors During modulation, when the feature distribution of the local temporal pattern at the PH end of the reaction progress depends on the parameter distribution modulation of the weight matrix and bias vector, it is also desirable to construct a closed-loop manifold in the parameter space, thereby making the dynamic convolution kernel parameters... It can improve the fidelity of subsequent spatial saliency feature extraction.

[0053] Therefore, if and If we consider the bias vector as the interface feature state control in the weight parameter space, then... and An additional phase factor will be introduced undoubtedly, and in order to maintain the phase coherence stability of the eigenstate, the eigenmanifold basis representation is first introduced as:

[0054]

[0055]

[0056] wherein, represents a first-order reaction progress PH end local timing eigenmanifold basis representation vector, represents a second-order reaction progress PH end local timing eigenmanifold basis representation vector.

[0057] Then, in the case of considering that the weight matrix constructs a feature migration coupling matrix element, the feature migration degree is:

[0058]

[0059] wherein, is a reaction progress PH end local timing feature migration encoding vector;

[0060] In this way, the symmetry breaking of the anisotropic constraint matrix element of the bias phase can be realized according to the bias phase:

[0061]

[0062]

[0063] wherein, represents the calculation of the L2 norm, and respectively represent the optimized bias vectors of the first layer and the second layer of the multi-layer perception model.

[0064] That is, by iteratively optimizing the bias vectors and , the defect of the additional phase that destroys the surface state of the weight modulation can be avoided while introducing the rotation phase perception of the local connection through the bias vector, so that the weight matrix and and the parameter distribution modulation of the bias vector and form a closed path for the generation of the dynamic convolution kernel parameter , and improve the feature extraction fidelity of the dynamic convolution kernel parameter .

[0065] Then, based on the dynamic convolution kernel parameter, the reaction progress PH end local timing segment content perceptual coding feature map is subjected to spatial saliency feature extraction to obtain a reaction progress PH end local timing segment content spatial saliency perceptual coding feature map, which is expressed by a formula as follows:

[0066]

[0067]

[0068] wherein, is a convolution operation, represents a Sigmoid activation function, represents a reaction progress PH end local timing segment content spatial attention weight distribution map, represents multiplication by a position point, represents a reaction progress PH end local timing segment content spatial saliency perceptual coding feature map.

[0069] That is, based on the dynamic convolution kernel parameter, the reaction progress PH end local timing segment content perceptual coding feature map is subjected to convolution coding to learn the context association and relative importance between local features, and the spatial attention weight distribution map of the reaction progress PH end local timing segment content perceptual coding feature map is output through a sigmoid normalization function. Further, the original reaction progress PH end local timing segment content perceptual coding feature map is subjected to weighted processing through the spatial attention weight distribution map, so that the key PH local change mode is enhanced and the non-key information is weakened, thereby obtaining the reaction progress PH end local timing segment content spatial saliency perceptual coding feature map. In this way, the model can focus more on the PH change details that are crucial to the reaction progress judgment, such as the starting point of the sudden drop, the valley value, the plateau stability, etc., thereby improving the prediction accuracy of the reaction endpoint.

[0070] In one specific example of the present application, the reaction progress PH end local timing segment content spatial saliency perceptual coding feature map is subjected to feature shape remodeling to obtain an enhanced reaction progress PH end timing feature vector as the enhanced reaction progress PH end timing feature representation, which is expressed by a formula as follows:

[0071]

[0072] wherein, represents an enhanced reaction progress PH end timing feature vector.

[0073] That is, by remodeling the feature shape of the reaction progress PH end local timing segment content space significant perception coding feature map into a one-dimensional vector form, a normalized one-dimensional feature vector format is provided to facilitate subsequent multi-modal parameter fusion and regression decision.

[0074] Specifically, the step S33 fuses the enhanced reaction progress oxidation-reduction potential end timing feature representation, the enhanced reaction progress PH end timing feature representation, the enhanced reaction progress temperature end timing feature representation, the enhanced reaction progress dissolved oxygen end timing feature representation and the enhanced reaction progress conductivity end timing feature representation to obtain a reaction progress multi-dimensional data timing feature joint representation. It should be understood that the reaction endpoint of organic wastewater treatment is the result of the comprehensive action of the coordinated changes of multiple physical and chemical parameters, and each sensor data reflects the state of the reaction system from different aspects, and there is complex mutual correlation and complementary information between each other. Therefore, based on the multi-modal information fusion technology, the enhanced timing feature representations of various sensor parameters are fused at the feature level to fully utilize the complementarity and synergy between different sensor data, overcome the limitations of single parameter monitoring, and form a feature representation that can fully reflect the overall comprehensive state of the reaction system, i.e. the reaction progress multi-dimensional data timing feature joint representation.

[0075] Specifically, the step S34, the current reaction time and the reaction progress multi-dimensional data time series feature joint representation input into the trained machine learning model to obtain a prediction result, which is the probability of the reaction reaching the endpoint at the current time. It should be understood that the current reaction time is a basic dimension for measuring the progress of the reaction, and the reaction progress multi-dimensional data time series feature joint representation depicts the overall state evolution of the reaction system at the current time and in the past period of time. Therefore, in order to make a probabilistic prediction of whether the reaction reaches the endpoint based on all available information (reaction time and reaction system state) at the current time, the present application adopts a regression algorithm based on supervised learning to realize real-time estimation of the reaction endpoint probability by inputting the current reaction time and the reaction progress multi-dimensional data time series feature joint representation into a pre-trained machine learning model. Specifically, first, after normalizing the current reaction time, it is embedded at the tail of the reaction progress multi-dimensional data time series feature joint representation as supplementary information of the time dimension. Then, the reaction progress multi-dimensional data time series feature joint representation fused with the time information is input into the machine learning model trained by a large amount of historical reaction data. Through the nonlinear transformation of multiple dense layers (Dense layers) inside the model, the deep-level features of the current reaction system state are learned, and through the Sigmoid activation function of the output layer, the output result is mapped into the probability interval of 0 to 1, wherein the output value closer to 1 indicates the higher probability of the reaction reaching the endpoint, and vice versa, the output value closer to 0 indicates the lower probability of the reaction reaching the endpoint. In this way, the powerful fitting and generalization ability of machine learning can be used to reliably predict whether the reaction reaches the endpoint at the current time, providing timely decision support for the operator.

[0076] In the method for treating organic wastewater described above, the step S4 generates a production intervention prompt signal in response to the probability of the reaction reaching the end point at the current time exceeding a preset threshold. That is, when the probability of the reaction reaching the end point at the current time output by the model exceeds the preset threshold (for example, 0.90), the system will automatically trigger the production intervention prompt signal to prompt the operator that the current reaction system has approached or reached the predetermined reaction end point, and suggest taking the next step or taking appropriate processing measures. The preset threshold is set by comprehensively considering historical reaction data, process requirements and economic benefits, aiming to ensure that the operator can be notified in time and take appropriate operation measures when the reaction approaches or reaches the end point. In actual application, the production intervention prompt signal can take various forms, for example, displaying a prominent visual alarm (such as a color-changing indicator light, a pop-up prompt) on the control interface of the operator, issuing a sound alarm, or directly triggering subsequent production operations (such as stopping stirring, closing the feed valve, starting the water pump, or switching to the next processing unit, etc.) through control logic (such as in a PLC or DCS system). In this way, a closed-loop control from data perception to decision response is realized, which not only ensures that the organic wastewater treatment achieves the expected purification effect, avoids under-treatment or over-treatment caused by human judgment errors or delays, but also effectively saves production costs such as energy and chemicals, and improves the automation level and operation efficiency of the entire processing flow.

[0077] In summary, the method for treating organic wastewater according to the embodiments of the present application is illustrated, which utilizes layered B-phase vanadium dioxide as an efficient Fenton-like catalyst to degrade pollutants in organic wastewater under normal temperature stirring conditions in cooperation with persulfate, and further introduces a deep learning algorithm in the redox reaction process of organic wastewater, extracts the time sequence feature representation of key reaction parameters in the reaction system by real-time monitoring of the high-frequency data stream of multiple sensors such as oxidation-reduction potential, pH, temperature, dissolved oxygen and conductivity in the reactor, and constructs a reaction progress prediction model by combining the current reaction time, to realize intelligent judgment and intervention warning of the reaction end point in the organic wastewater treatment process. In this way, the problems of insufficient reaction or over-reaction can be effectively avoided, and the treatment effect and resource utilization rate of organic wastewater can be improved.

[0078] Further, the present application also provides a treatment system for organic wastewater.

[0079] Figure 5 A block diagram of the treatment system for organic wastewater according to the embodiments of the present application is shown in FIG. 1. As shown in FIG. 1, the treatment system for organic wastewater according to the embodiments of the present application comprises a reaction tank 1, a stirring device 2, a sensor device 3, a data acquisition and processing device 4, a control device 5 and a production intervention prompt device 6. Figure 5As shown, the organic wastewater treatment system 100 according to the embodiment of the present application comprises: a catalytic reaction module 110 for removing organic pollutants by adding vanadium dioxide as a Fenton-like catalyst and persulfate into the organic wastewater to react under the condition of normal temperature and stirring, wherein the vanadium dioxide is layered B-phase vanadium dioxide; a reaction monitoring module 120 for acquiring the current reaction time and real-time high-frequency data stream from sensors in the reactor during the reaction process, wherein the real-time high-frequency data stream comprises redox potential data stream, PH data stream, temperature data stream, dissolved oxygen data stream and conductivity data stream; a reaction endpoint prediction module 130 for inputting the current reaction time and the real-time high-frequency data stream into the trained machine learning model to obtain the probability of the reaction reaching the endpoint at the current time; and a production intervention prompting module 140 for generating a production intervention prompting signal in response to the probability of the reaction reaching the endpoint at the current time exceeding a preset threshold.

[0080] Here, those skilled in the art can understand that the specific operations of each module in the above organic wastewater treatment system have been described in detail in the above description of the organic wastewater treatment method according to the embodiment of the present application, and therefore, the repeated description thereof will be omitted. Figures 1 to 4

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.​

Claims

1. A method for treating organic wastewater, characterized in that, include: S1, using vanadium dioxide as a Fenton-like catalyst, it and persulfate are added to organic wastewater and reacted under normal temperature and stirring conditions to remove organic pollutants, wherein the vanadium dioxide is layered B-phase vanadium dioxide; S2, during the reaction process, the current reaction time and real-time high-frequency data streams from various sensors in the reactor are acquired in real time. The real-time high-frequency data streams include redox potential data streams, pH data streams, temperature data streams, dissolved oxygen data streams, and conductivity data streams. S3, input the current reaction time and the real-time high-frequency data stream into the trained machine learning model to obtain the probability that the reaction reaches the endpoint at the current moment; S4, in response to the probability that the reaction reaches the endpoint at the current moment exceeds a preset threshold, a production intervention prompt signal is generated; Step S3 includes: S31, extracting the temporal features of each data item in the real-time high-frequency data stream to obtain temporal feature representations of the reaction progress at the redox potential, pH, temperature, dissolved oxygen, and conductivity; S32, performing temporal fine-grained enhancement encoding on the temporal feature representations of the reaction progress at the redox potential, pH, temperature, dissolved oxygen, and conductivity to obtain enhanced temporal feature representations of the reaction progress at the redox potential and pH. The process involves: S33, fusing the enhanced reaction progress time-series features at the redox potential, pH, temperature, dissolved oxygen, and conductivity ends to obtain a joint representation of the reaction progress multidimensional data time-series features; S34, inputting the current reaction duration and the joint representation of the reaction progress multidimensional data time-series features into the trained machine learning model to obtain a prediction result, which is the probability that the reaction has reached its endpoint at the current moment. Step S32 includes: S321, performing content awareness on the temporal feature representation of the reaction progress at the pH end to obtain a content-aware encoding feature map of the local temporal segment of the reaction progress at the pH end; S322, inputting the content-aware encoding feature map of the local temporal segment of the reaction progress at the pH end into a spatial attention layer to obtain a spatially salient encoding feature map of the content of the local temporal segment of the reaction progress at the pH end; S323, reshaping the feature shape of the spatially salient encoding feature map of the content of the local temporal segment of the reaction progress at the pH end to obtain an enhanced temporal feature vector of the reaction progress at the pH end as the enhanced temporal feature representation of the reaction progress at the pH end. Step S322 includes: calculating dynamic convolution kernel parameters based on the content-aware coding feature map of the local temporal segment at the pH end of the reaction progress; and extracting spatial saliency features from the content-aware coding feature map of the local temporal segment at the pH end of the reaction progress based on the dynamic convolution kernel parameters to obtain the spatial saliency coding feature map of the content of the local temporal segment at the pH end of the reaction progress.

2. The method for treating organic wastewater according to claim 1, characterized in that, Step S31 includes: The real-time high-frequency data stream is subjected to time-series analysis based on 1D-CNN to obtain the time-series characteristics of the reaction progress at the redox potential, pH, temperature, dissolved oxygen, and conductivity.

3. The method for treating organic wastewater according to claim 2, characterized in that, Step S321 includes: Based on a preset window, the temporal feature representation of the reaction progress at the pH end is segmented into feature segments to obtain a set of local temporal feature vectors at the pH end of the reaction progress; After reshaping the set of local temporal feature vectors at the pH end of the reaction progress into a local temporal segment encoding feature map at the pH end of the reaction progress, content-aware features based on convolutional coding are applied to obtain the content-aware encoding feature map of the local temporal segment at the pH end of the reaction progress.

4. The method for treating organic wastewater according to claim 3, characterized in that, Based on the content-aware encoding feature map of the local temporal segment at the pH end of the reaction progress, the dynamic convolution kernel parameters are calculated, including: The content-aware encoding feature map of the local temporal segment at the pH end of the reaction progress is subjected to global mean pooling along the channel dimension to obtain the content-aware encoding vector of the local temporal segment at the pH end of the reaction progress. The content-aware encoding vector of the local temporal segment at the pH end of the reaction progress is then input into the multilayer perceptron model to obtain the dynamic convolution kernel parameters.

5. The method for treating organic wastewater according to claim 4, characterized in that, The content-aware encoding vector of the local temporal segment at the pH end of the reaction progress is input into a multilayer perceptron model to obtain the dynamic convolution kernel parameters, including: The bias vector of the multilayer perceptron model is iteratively optimized based on the preservation of phase stability in the parameter space to obtain the optimized bias vector; Based on the optimized bias vector, the multilayer perceptron model is used to process the content-aware encoding vector of the local temporal segment at the pH end of the reaction progress to obtain the dynamic convolution kernel parameters.

6. An organic wastewater treatment system for implementing the organic wastewater treatment method according to any one of claims 1-5, characterized in that, include: A catalytic reaction module is used to react vanadium dioxide as a Fenton-like catalyst with persulfate in organic wastewater at room temperature and under stirring conditions to remove organic pollutants. The vanadium dioxide is layered B-phase vanadium dioxide. The reaction monitoring module is used to acquire the current reaction time and real-time high-frequency data streams from various sensors in the reactor during the reaction process. The real-time high-frequency data streams include redox potential data streams, pH data streams, temperature data streams, dissolved oxygen data streams, and conductivity data streams. The reaction endpoint prediction module is used to input the current reaction duration and the real-time high-frequency data stream into a trained machine learning model to obtain the probability that the reaction has reached its endpoint at the current moment. The production intervention prompt module is used to generate a production intervention prompt signal in response to the probability that the reaction reaches the endpoint at the current moment exceeds a preset threshold.

Images