Intelligent self-cleaning system for water treatment sensor probe based on multi-mode sensing

Abstract

The invention discloses a water treatment sensor probe intelligent self-cleaning system based on multi-modal sensing, and relates to the technical field of water treatment, and the system comprises a data sensing module which is used for collecting multi-sensor data of a sludge concentration meter, a sludge level meter, a water inlet and outlet flow meter and a turbidimeter in a water treatment system, and agent usage amount data in a water treatment process; establishing a water treatment process database; and the embedded central processor module is electrically connected with the data sensing module, is internally provided with a pollutant monitoring model, and analyzes and compares the acquired multi-source data by using a preset AI algorithm. Data of multiple sensors such as a sludge concentration meter, a sludge level meter, a water inlet and outlet flow meter and a turbidimeter are widely collected through the data sensing module, the rich data are transmitted to the embedded central processing unit module, the embedded central processing unit module comprehensively analyzes the multi-source data through a preset AI algorithm, the accuracy of the current data is intelligently recognized, and the accuracy of the current data is improved. And the pollution degree of the sensor probe can be accurately judged.

Description

Technical Field

[0001] The present invention relates to the technical field of water treatment, and in particular to an intelligent self-cleaning system of a water treatment sensor probe based on multimodal perception. Background Art

[0002] In the current field of water treatment technology, photoelectric sensors are widely used in sewage treatment. Among them, "water treatment sensor probes" are core monitoring components, such as sludge concentration probes, mud level meter probes, and turbidity meter probes. They convert relevant parameters in the water into detectable signals through specific physical principles to monitor key indicators in the water treatment process, and play an important role in the precise control of the entire water treatment system.

[0003] However, due to the extremely complex water treatment environment, these "water treatment sensor probes" will inevitably be affected by sludge adhesion and scale, which will lead to inaccurate measurement data and seriously hinder the precise control of the water treatment process.

[0004] At present, the cleaning technology for sensor probes mainly relies on the following three methods:

[0005] Manual cleaning: Clean the probe at a fixed cycle or after a problem occurs. This method lacks real-time performance and cannot detect probe contamination and clean it in time, which may cause long-term inaccurate measurement data and affect the control accuracy of the water treatment process.

[0006] Mechanical brushing device: With the help of a stepper motor, the nylon brush head is driven to rotate at a fixed cycle (such as once every 4 hours) to remove the attachments on the probe surface. However, the cleaning cycle of this method is fixed and cannot be flexibly adjusted according to the actual degree of contamination of the probe, which easily leads to waste of resources and poor adaptability to complex working conditions. For example, when encountering fiber entanglements, the mechanical brushing device is difficult to effectively remove, but still works according to a fixed procedure, and cannot prompt human intervention in time.

[0007] High-pressure water flushing device: The high-pressure water pump (pressure ≥ 3MPa) is triggered periodically through a preset program to remove pollutants using a jet of water. Similarly, it also has similar problems as the mechanical scrubbing device, relying on a fixed cleaning cycle and unable to dynamically adjust according to the actual pollution situation. In addition, when faced with complex working conditions, it lacks data monitoring and judgment capabilities and cannot prompt the staff to handle it in time, resulting in work delays. Summary of the invention

[0008] In order to solve the above technical problems, an intelligent self-cleaning system for water treatment sensor probes based on multimodal perception is provided. This technical solution solves the above problems.

[0009] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0010] Intelligent self-cleaning system for water treatment sensor probes based on multi-modal perception, including:

[0011] Data perception module: used to collect multi-sensor data of sludge concentration meters, sludge level meters, influent and effluent flow meters, and turbidity meters in the water treatment system, as well as data on the dosage of chemicals used in the water treatment process, and establish a water treatment process database;

[0012] Embedded central processor module: electrically connected to the data perception module, with a built-in pollutant monitoring model, using a preset AI algorithm to analyze and compare the multi-source data collected, calculating pollution correlation indicators, and judging the pollution status of the sensor probes in combination with the measurement data of each probe;

[0013] Execution module: electrically connected to the embedded central processor module, receiving instructions issued by the embedded central processor module, and performing cleaning operations on the sensor probes. The execution module includes a mechanical brushing unit and a high-pressure water flushing unit, and can control the cleaning time and cleaning intensity respectively;

[0014] Feedback module: electrically connected to the embedded central processor module, after the cleaning operation is completed, collecting the data of the data perception module and transmitting it to the embedded central processor module. The feedback module includes a data acquisition sub-module and a data analysis sub-module. The data acquisition sub-module is responsible for collecting sensor data before and after cleaning, and the data analysis sub-module preprocesses the collected data, extracts key features, and then transmits it to the embedded central processor module;

[0015] Reinforcement learning optimization module: integrated in the embedded central processor module, using the DQN algorithm, iteratively optimizing the cleaning strategy library according to the data provided by the feedback module. The cleaning strategy library includes cleaning timing, cleaning duration, and cleaning intensity.

[0016] Preferably, the data perception module includes a multi-sensor acquisition unit, an auxiliary data acquisition unit, and a data fusion unit;

[0017] The multi-sensor acquisition unit is used to collect various types of sensor data in real time;

[0018] The auxiliary data acquisition unit is used to collect auxiliary data such as the dosage of chemicals used;

[0019] The data fusion unit fuses the data obtained by the multi-sensor acquisition unit and the auxiliary data acquisition unit to generate a complete water treatment data information flow.

[0020] Preferably, the pollutant monitoring model in the embedded central processor module includes a data comparison unit, a pollution degree judgment unit, and an abnormal situation warning unit:

[0021] The data comparison unit compares the collected real-time data with the built-in standard data models under different working conditions;

[0022] The pollution degree judgment unit judges the pollution degree of the sensor probe according to the data comparison result and in combination with a preset AI algorithm;

[0023] When the pollution degree reaches the warning threshold, the abnormal situation warning unit issues a warning signal.

[0024] Preferably, the mechanical brushing unit of the execution module includes a motor drive unit and a rotation control unit:

[0025] The motor drive unit is used to drive the brush head to rotate;

[0026] The rotation control unit receives instructions from the embedded central processor module, controls the rotation speed and rotation time of the motor, and thus controls the brushing intensity and brushing duration.

[0027] Preferably, the high-pressure water flushing unit of the execution module includes a high-pressure water pump device, a pressure regulation unit and a water flow jet control unit:

[0028] The high-pressure water pump device is used to generate high-pressure water flow;

[0029] The pressure regulation unit receives instructions from the embedded central processor module, adjusts the pressure of the high-pressure water pump, and controls the flushing intensity;

[0030] The water flow jet control unit controls the jet angle and jet time of the water flow.

[0031] Preferably, the embedded central processor module also integrates a time series prediction model based on Transformer and a historical data analysis model, including a trend prediction unit, a working condition analysis unit and an intervention prompt unit:

[0032] The trend prediction unit analyzes the data trend based on the time series prediction model of Transformer and predicts the pollution trend;

[0033] The working condition analysis unit analyzes the complexity of the working condition in combination with the historical data analysis model and judges whether human intervention is required;

[0034] When it is judged that human intervention is required, the intervention prompt unit issues a prompt signal.

[0035] Preferably, the time series prediction model based on Transformer uses the formula P t+n = f(P t , P t-1 , …, P t-m ) for pollution trend prediction, where Pt+n is the predicted pollution index at the future nth moment, P t , P t-1 , …, P t-m are the actual pollution indices at the past m + 1 moments, and f is the prediction function trained based on the Transformer model.

[0036] Preferably, the calculation formula of the pollution correlation index is:

[0037] I = α × F + β × T + γ × D

[0038] where F represents the influent flow rate, T represents the turbidity, D represents the dosage of the chemical agent, and α, β, and γ are the weight coefficients determined according to the actual working conditions.

[0039] Preferably, the data analysis sub-module of the feedback module includes a feature extraction unit, a data comparison unit, and a cleaning effect evaluation unit:

[0040] The feature extraction unit extracts key features from the collected data, such as the change rate of the data before and after cleaning, etc.;

[0041] The data comparison unit compares the data features before and after cleaning;

[0042] The cleaning effect evaluation unit evaluates the cleaning effect according to the data comparison result and transmits the evaluation result to the reinforcement learning optimization module.

[0043] Preferably, the DQN algorithm of the reinforcement learning optimization module includes a state definition unit, an action selection unit, a reward calculation unit, and a network update unit:

[0044] The state definition unit defines information such as sensor data, pollution degree, and cleaning strategy as the system state;

[0045] The action selection unit selects a cleaning action through the policy network according to the current state;

[0046] The reward calculation unit calculates the reward value according to the cleaning effect evaluation result;

[0047] The network update unit updates the policy network and the target network using the reward value and the new state to realize the iterative optimization of the cleaning strategy library.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows: The data perception module widely collects multi-sensor data such as sludge concentration meters, sludge level meters, influent and effluent flow meters, and turbidity meters. These rich data are transmitted to the embedded central processor module. The embedded central processor module uses a preset AI algorithm to comprehensively analyze the multi-source data, intelligently identify the accuracy of the current data, and then accurately judge the pollution degree of the sensor probe. Different from the traditional method of judging pollution relying on single data or simple experience, the present invention can more comprehensively and accurately grasp the probe pollution situation, providing a reliable basis for subsequent cleaning decisions. And the reinforcement learning algorithm is adopted. After the cleaning operation is completed, the feedback module will quickly collect the data of the data perception module and transmit it to the embedded central processor module. The processor evaluates the cleaning effect based on these data and iteratively optimizes the cleaning strategy library including cleaning timing, cleaning duration, and cleaning intensity. For example, when it is judged that the probe pollution is light, the system automatically shortens the cleaning time and reduces the cleaning intensity, avoiding unnecessary energy consumption. The reasonable cleaning strategy reduces the over-cleaning and damage to the probe, thus effectively extending the service life of the probe. At the same time, the embedded central processor module integrates a time series prediction model and a historical data analysis model based on Transformer. Through these models, the system can predict the pollution trend in advance. Before the complex working conditions (such as the sharp increase in the influent turbidity of the water plant during the rainy season) may cause probe pollution, it can predict the pollution risk and initiate preventive cleaning in advance to ensure the accuracy of the measurement data. When the system encounters the complex situation that the probe is entangled by fiber-like debris and cannot complete self-cleaning, the feedback module will automatically identify it and prompt the worker to perform manual intervention through the embedded central processor module to solve the problem in time, ensuring the stable operation of the system under complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a system structure framework diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations.

[0051] Referring to Figure 1 as shown, the intelligent self-cleaning system for water treatment sensor probes based on multi-modal perception includes:

[0052] A data perception module: used to collect multi-sensor data of sludge concentration meters, sludge level meters, influent and effluent flow meters, and turbidity meters in the water treatment system, as well as the data of the dosage of chemicals in the water treatment process, and establish a water treatment process database;

[0053] Embedded Central Processing Unit Module: Electrically connected to the data perception module, with a built-in pollutant monitoring model, which analyzes and compares multi-source data collected by using a preset AI algorithm, calculates pollution correlation indicators, and judges the pollution status of the sensor probe in combination with the measurement data of each probe;

[0054] Execution Module: Electrically connected to the embedded central processing unit module, receives instructions issued by the embedded central processing unit module, and performs a cleaning operation on the sensor probe. The execution module includes a mechanical brushing unit and a high-pressure water flushing unit, and can control the cleaning time and cleaning intensity respectively;

[0055] Feedback Module: Electrically connected to the embedded central processing unit module, after the cleaning operation is completed, collects the data of the data perception module and transmits it to the embedded central processing unit module. The feedback module includes a data acquisition sub-module and a data analysis sub-module. The data acquisition sub-module is responsible for collecting sensor data before and after cleaning, and the data analysis sub-module preprocesses the collected data, extracts key features and then transmits them to the embedded central processing unit module;

[0056] Reinforcement Learning Optimization Module: Integrated in the embedded central processing unit module, adopts the DQN algorithm, and iteratively optimizes the cleaning strategy library according to the data provided by the feedback module. The cleaning strategy library includes the cleaning timing, cleaning duration, and cleaning intensity.

[0057] Specifically, first, the data perception module collects various sensor data and chemical agent usage data, and establishes a database to provide comprehensive information for subsequent analysis. The embedded central processing unit module has a built-in pollutant monitoring model, analyzes and compares multi-source data through a preset AI algorithm, calculates pollution correlation indicators to judge the pollution status of the sensor probe. The execution module, according to the processor instructions, uses the mechanical brushing unit and the high-pressure water flushing unit to clean the probe, and can control the cleaning time and intensity. The feedback module collects and processes data after cleaning, providing data support for the reinforcement learning optimization module. The reinforcement learning optimization module adopts the DQN algorithm and optimizes the cleaning strategy library according to the feedback data. By comprehensively collecting data through multi-modal perception, the accuracy of judging the pollution of the sensor probe is improved. The cleaning operation is accurately controlled according to the pollution status, avoiding unnecessary cleaning and reducing energy consumption. At the same time, the reinforcement learning continuously optimizes the cleaning strategy, enabling the system to adapt to different pollution situations, extending the service life of the probe, ensuring the accuracy of sensor measurement in the water treatment process, and improving the operation efficiency of the water treatment system.

[0058] The data perception module consists of a multi-sensor acquisition unit, an auxiliary data acquisition unit, and a data fusion unit;

[0059] The multi-sensor acquisition unit is used to collect various types of sensor data in real time;

[0060] The auxiliary data acquisition unit is used to collect auxiliary data such as the dosage of the medicament;

[0061] The data fusion unit fuses the data obtained by the multi-sensor acquisition unit and the auxiliary data acquisition unit to generate a complete water treatment data information flow.

[0062] The refined structure of the specific data perception module

[0063] Principle: Through multi-source data acquisition and fusion processing, the system can understand the operating state of the water treatment system from multiple dimensions. Compared with the single data acquisition method, this multi-modal perception data acquisition method can more accurately reflect the pollution situation of the sensor probe, provide strong support for subsequent accurate judgment of the pollution degree and formulation of reasonable cleaning strategies, and improve the adaptability of the system to complex working conditions.

[0064] The pollutant monitoring model in the embedded central processor module includes a data comparison unit, a pollution degree judgment unit, and an abnormal situation warning unit:

[0065] The data comparison unit compares the collected real-time data with the built-in standard data models under different working conditions;

[0066] The pollution degree judgment unit judges the pollution degree of the sensor probe according to the data comparison result and in combination with a preset AI algorithm;

[0067] When the pollution degree reaches the warning threshold, the abnormal situation warning unit issues a warning signal.

[0068] Specific principle: Through comparison with the standard data model and judgment by the AI algorithm, the accuracy and scientificity of the pollution degree judgment are improved. The abnormal situation warning function enables the system to detect potential problems in a timely manner, take measures in advance, avoid inaccurate sensor measurements caused by severe pollution, ensure the stable operation of the water treatment system, and reduce losses caused by equipment failures.

[0069] The mechanical brushing unit of the execution module includes a motor drive unit and a rotation control unit:

[0070] The motor drive unit is used to drive the brush head to rotate;

[0071] The rotation control unit receives instructions from the embedded central processor module, controls the rotation speed and rotation time of the motor, and thus controls the brushing intensity and brushing duration.

[0072] Specifically, the mechanical scrubbing unit consists of a motor drive unit and a rotation control unit, which can flexibly adjust the scrubbing parameters according to the processor instructions. For light pollution, the scrubbing intensity and time can be reduced to avoid damage to the probe caused by excessive cleaning; for heavy pollution, the scrubbing intensity and time can be increased to ensure the cleaning effect. This precise control improves the cleaning efficiency, extends the service life of the probe, and also reduces energy consumption during the cleaning process.

[0073] The high-pressure water flushing unit of the execution module includes a high-pressure water pump device, a pressure regulating unit and a water jet control unit:

[0074] The high-pressure water pump device is used to generate high-pressure water flow;

[0075] The pressure regulating unit receives instructions from the embedded central processing unit module, adjusts the pressure of the high-pressure water pump, and controls the flushing intensity;

[0076] The water jet control unit controls the jet angle and jet time of the water jet.

[0077] Specifically, the high-pressure water flushing unit includes a high-pressure water pump device, a pressure regulating unit and a water jet control unit. In conjunction with the mechanical scrubbing unit, the high-pressure water flushing unit provides another cleaning method and can accurately control the flushing parameters. By adjusting the pressure, angle and time, it can effectively remove pollutants of different types and degrees and improve the cleaning effect. At the same time, the flushing parameters are reasonably adjusted according to the pollution situation to avoid the waste of water resources and excessive flushing of the probe, ensuring the normal use of the probe and the stable operation of the system.

[0078] The embedded CPU module also integrates a Transformer-based time series prediction model and a historical data analysis model, including a trend prediction unit, a working condition analysis unit, and an intervention prompt unit:

[0079] The trend prediction unit analyzes data trends based on the Transformer time series prediction model to predict pollution trends;

[0080] The working condition analysis unit analyzes the complexity of the working condition in combination with the historical data analysis model to determine whether human intervention is required;

[0081] The intervention prompt unit sends a prompt signal when determining that human intervention is required.

[0082] Specifically, it can predict the pollution trend in advance, enabling the system to take preventive cleaning measures before pollution occurs and reducing the impact of pollution on sensor measurements. The analysis of the complexity of working conditions and timely human intervention prompts improve the system's ability to handle complex working conditions, ensure the stable operation of the system, avoid problems in the water treatment process due to system failures, and improve the reliability and safety of the water treatment system.

[0083] The Transformer-based time series prediction model uses the formula P t+n = f(P t , P t-1 , …, P t-m ) to predict the pollution trend, where P t+n is the predicted pollution index at the future nth moment, and P t , P t-1 , …, P t-m are the actual pollution indices at the past m + 1 moments, and f is the prediction function trained based on the Transformer model.

[0084] Specifically, using this formula to predict the pollution trend enables the system to more accurately grasp the pollution change situation. Based on accurate predictions, the system can more reasonably arrange the cleaning plan, take preventive measures in advance, further improve the system's intelligence level and operation efficiency, and reduce measurement errors and system failures caused by pollution.

[0085] The calculation formula for the pollution correlation index is:

[0086] I = α × F + β × T + γ × D

[0087] where F represents the influent flow rate, T represents the turbidity, D represents the chemical dosage, and α, β, and γ are weight coefficients determined according to the actual working conditions.

[0088] Specifically, compared with judging pollution by a single index, this formula comprehensively considers multiple factors, making the pollution judgment more accurate and scientific. Based on a more accurate pollution judgment, the system can formulate a more reasonable cleaning strategy, improve the pertinence and effectiveness of cleaning, and then enhance the overall performance of the system, ensuring the precise control of the water treatment process.

[0089] The data analysis sub-module of the feedback module includes a feature extraction unit, a data comparison unit, and a cleaning effect evaluation unit:

[0090] The feature extraction unit extracts key features from the collected data, such as the change rate of data before and after cleaning, etc.;

[0091] The data comparison unit compares the data features before and after cleaning;

[0092] The cleaning effect evaluation unit evaluates the cleaning effect according to the data comparison result and transmits the evaluation result to the reinforcement learning optimization module.

[0093] Specifically, by analyzing the data before and after cleaning to evaluate the cleaning effect, accurate feedback information is provided for the reinforcement learning optimization module. Based on this information, the reinforcement learning optimization module optimizes the cleaning strategy, enabling the system to continuously improve the cleaning effect, enhance the cleaning efficiency, further ensure the normal operation of the sensor probe, and ensure the stable operation of the water treatment system.

[0094] The DQN algorithm of the reinforcement learning optimization module includes a state definition unit, an action selection unit, a reward calculation unit, and a network update unit:

[0095] The state definition unit defines information such as sensor data, pollution degree, and cleaning strategy as the system state;

[0096] The action selection unit selects a cleaning action through the policy network according to the current state;

[0097] The reward calculation unit calculates the reward value according to the cleaning effect evaluation result;

[0098] The network update unit updates the policy network and the target network using the reward value and the new state to achieve iterative optimization of the cleaning strategy library.

[0099] Specifically, by continuously iteratively optimizing the cleaning strategy library, the system can automatically adjust the cleaning strategy according to different pollution situations and cleaning effects, find the optimal cleaning solution, not only improve the cleaning efficiency, reduce energy consumption, but also further extend the service life of the probe, enhance the intelligent level and adaptive ability of the system, and ensure the long-term stable operation of the water treatment system.

[0100] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. The intelligent self-cleaning system for a water treatment sensor probe based on multimodal perception, characterized in that Including: Data perception module: It is used to collect multi-sensor data of sludge concentration meters, sludge level meters, influent and effluent flow meters, and turbidity meters in the water treatment system, as well as the data of chemical agent usage during the water treatment process, and establish a water treatment process database; Embedded central processor module: Electrically connected to the data perception module, with a built-in pollutant monitoring model, using a preset AI algorithm to analyze and compare the collected multi-source data, calculating pollution correlation indicators, and judging the pollution status of the sensor probes by combining the measurement data of each probe; Execution module: Electrically connected to the embedded central processor module, receiving instructions issued by the embedded central processor module, and performing cleaning operations on the sensor probes. The execution module includes a mechanical brushing unit and a high-pressure water flushing unit, and can control the cleaning time and cleaning intensity respectively; Feedback module: Electrically connected to the embedded central processor module, after the cleaning operation is completed, collecting the data of the data perception module and transmitting it to the embedded central processor module. The feedback module includes a data acquisition sub-module and a data analysis sub-module. The data acquisition sub-module is responsible for collecting sensor data before and after cleaning, and the data analysis sub-module preprocesses the collected data, extracts key features, and then transmits it to the embedded central processor module; Reinforcement learning optimization module: Integrated in the embedded central processor module, using the DQN algorithm, iteratively optimizing the cleaning strategy library according to the data provided by the feedback module. The cleaning strategy library includes cleaning timing, cleaning duration, and cleaning intensity.

2. The intelligent self-cleaning system for a water treatment sensor probe based on multi-modal perception according to claim 1, wherein The data perception module includes a multi-sensor acquisition unit, an auxiliary data acquisition unit, and a data fusion unit; The multi-sensor acquisition unit is used to collect various sensor data in real time; The auxiliary data acquisition unit is used to collect auxiliary data such as chemical agent usage; The data fusion unit fuses and processes the data obtained by the multi-sensor acquisition unit and the auxiliary data acquisition unit to generate a complete water treatment data information flow.

3. The intelligent self-cleaning system for the water treatment sensor probe based on multi-modal perception according to claim 1, wherein, The pollutant monitoring model in the embedded central processor module includes a data comparison unit, a pollution degree judgment unit, and an abnormal situation warning unit: The data comparison unit compares the collected real-time data with the standard data models under different working conditions built in; The pollution degree judgment unit judges the pollution degree of the sensor probes according to the data comparison results, combined with a preset AI algorithm; The abnormal situation warning unit issues a warning signal when the pollution degree reaches the warning threshold.

4. The intelligent self-cleaning system for the water treatment sensor probe based on multi-modal perception according to claim 1, wherein, The mechanical brushing unit of the execution module includes a motor drive unit and a rotation control unit: The motor drive unit is used to drive the brush head to rotate; The rotation control unit receives instructions from the embedded central processor module, controls the rotation speed and rotation time of the motor, and thus controls the brushing intensity and brushing duration.

5. The intelligent self-cleaning system for the water treatment sensor probe based on multi-modal perception according to claim 1, wherein The high-pressure water flushing unit of the execution module includes a high-pressure water pump device, a pressure adjustment unit, and a water flow injection control unit: The high-pressure water pump device is used to generate high-pressure water flow; The pressure adjustment unit receives instructions from the embedded central processor module, adjusts the pressure of the high-pressure water pump, and controls the flushing intensity; The water flow injection control unit controls the injection angle and injection time of the water flow.

6. The intelligent self-cleaning system for the water treatment sensor probe based on multi-modal perception according to claim 1, wherein The embedded central processing unit module also integrates a time series prediction model and a historical data analysis model based on Transformer, including a trend prediction unit, a working condition analysis unit, and an intervention prompt unit: The trend prediction unit analyzes data trends and predicts pollution trends based on the time series prediction model of Transformer; The working condition analysis unit analyzes the complexity of the working condition in combination with the historical data analysis model to determine whether human intervention is required; When it is determined that human intervention is required, the intervention prompt unit issues a prompt signal.

7. The intelligent self-cleaning system for the water treatment sensor probe based on multimodal perception according to claim 1, wherein The Transformer-based time series prediction model uses formula P t+n = f(P t , P t-1 , …, P t-m ) to predict the pollution trend, where P t+n is the predicted pollution index at the future nth moment, and P t , P t-1 , …, P t-m are the actual pollution indices at the past m + 1 moments, and f is the prediction function trained based on the Transformer model.

8. The intelligent self-cleaning system for a water treatment sensor probe based on multi-modal perception according to claim 1, wherein, The calculation formula for the pollution correlation index is: I = α×F + β×T + γ×D Wherein, F represents the influent flow rate, T represents the turbidity, D represents the chemical dosage, and α, β, and γ are weight coefficients determined according to the actual working conditions.

9. The intelligent self-cleaning system for the water treatment sensor probe based on multi-modal perception according to claim 1, characterized in that, The data analysis sub-module of the feedback module includes a feature extraction unit, a data comparison unit, and a cleaning effect evaluation unit: The feature extraction unit extracts key features from the collected data, such as the change rate of the data before and after cleaning, etc.; The data comparison unit compares the data features before and after cleaning; The cleaning effect evaluation unit evaluates the cleaning effect according to the data comparison result and transmits the evaluation result to the reinforcement learning optimization module.

10. The intelligent self-cleaning system for the water treatment sensor probe based on multi-modal perception according to claim 1, characterized in that, The DQN algorithm of the reinforcement learning optimization module includes a state definition unit, an action selection unit, a reward calculation unit, and a network update unit: The state definition unit defines information such as sensor data, pollution degree, and cleaning strategy as the system state; The action selection unit selects a cleaning action through the policy network according to the current state; The reward calculation unit calculates the reward value according to the cleaning effect evaluation result; The network update unit updates the policy network and the target network using the reward value and the new state to achieve iterative optimization of the cleaning strategy library.

Images