Intelligent control method for self-cleaning filter based on perception of Internet of Things

Abstract

The invention relates to the technical field of intelligent control of filters, and discloses an intelligent control method for a self-cleaning filter based on sensing of the Internet of Things. The method comprises the following steps: collecting multi-source sensing data such as inlet and outlet water pressure difference, flow fluctuation frequency, turbidity change rate and vibration spectrum characteristics when the filter runs; performing hierarchical clustering on the data according to a preset blockage level threshold value, and generating a blockage feature matrix with a timestamp; the matrix dynamic state is monitored in real time, and when the turbidity change rate and the covariance of the vibration spectrum characteristics exceed a dynamic threshold value, self-adaptive cleaning strategy switching is triggered; in combination with the time-space matching degree of a historical cleaning period and a current matrix, decomposing sub-feature sets of high-frequency blockage and low-frequency deposition modes, and correcting the weight of the high-frequency sub-feature sets through cross validation of fluid dynamics and mechanical wear parameters; and finally, a preferential cleaning area is reversely positioned along the structural topological graph of the filter, a three-dimensional cleaning path instruction is generated, and the opening degree of a blow-down valve is synchronously adjusted, so that the control and operation performance of the filter is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of filter intelligent control, in particular to a self-cleaning filter intelligent control method based on Internet of Things sensing. BACKGROUND

[0002] In the fields of water treatment, industrial circulating water system, water supply and drainage engineering, etc., as the key equipment to ensure the cleanliness of fluid medium, the running stability and cleaning efficiency of the self-cleaning filter directly affect the energy consumption level and running life of the whole system. At present, the self-cleaning filters on the market mostly adopt fixed period cleaning or single pressure difference triggered cleaning control mode, which has obvious limitations.

[0003] The filter using fixed period cleaning will perform cleaning operation according to the preset time regardless of the actual clogging situation. When the actual clogging degree of the filter is lighter, the early cleaning will cause waste of water resources and electric energy, and the frequent cleaning action will also increase the mechanical wear of the internal components of the filter, shortening the service life of the equipment; when the actual clogging degree of the filter is heavier, if the preset cleaning time is not reached, the filtration efficiency will be greatly reduced, and even the fluid flow will be blocked, affecting the normal operation of the whole system.

[0004] The control mode relying on single inlet and outlet water pressure difference triggered cleaning, although considering the actual clogging situation of the filter to some extent, can only reflect the overall clogging state of the filter, and cannot accurately capture the local characteristics and clogging types. In the actual operation process, the clogging degrees of different regions in the filter often differ, and some regions may have serious clogging, while the overall pressure difference has not reached the trigger threshold, at which time the system cannot timely perform targeted cleaning on the local clogging region, causing continuous deterioration of the filtration effect; in addition, when the filter is blocked by different nature pollutants, such as deposition of particulate impurities and adhesion of viscous substances, the single pressure difference signal cannot distinguish these different clogging modes, which is easy to cause the cleaning strategy to be unmatched with the actual clogging situation, resulting in incomplete cleaning or over-cleaning.

[0005] The control logic of the existing self-cleaning filter lacks comprehensive analysis and utilization of multi-source operation data. The data generated during the operation of the filter, such as flow fluctuation, turbidity change, vibration signal, etc., all contain rich information about the clogging state, but the existing technology fails to effectively integrate these data, cannot build a comprehensive clogging feature model, and thus the accuracy of the clogging state judgment is low. In terms of cleaning path planning, the existing filter usually adopts a fixed cleaning path, which cannot adjust the cleaning trajectory according to the real-time clogging feature matrix, and it is difficult to efficiently clean the priority clogging area, further reducing the cleaning efficiency and the recovery speed of the filtering performance. The existence of these problems makes the existing self-cleaning filter unable to meet the current demand for efficient and accurate filtering equipment in the field of industrial production and water treatment in terms of intelligent degree, operation economy and filtering stability. SUMMARY

[0006] The purpose of the present application is to provide an intelligent control method for a self-cleaning filter based on Internet of Things perception, to solve the problems raised in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides an intelligent control method for a self-cleaning filter based on Internet of Things perception, which comprises:

[0008] Collecting multi-source sensing data during the operation of the self-cleaning filter, the multi-source sensing data including the pressure difference between the inlet and outlet water, the flow fluctuation frequency, the turbidity change rate and the vibration spectrum feature;

[0009] Hierarchical clustering the multi-source sensing data according to the preset clogging level threshold to generate a clogging feature matrix containing a time stamp;

[0010] Real-time monitoring the dynamic change trend of the clogging feature matrix, and triggering an adaptive cleaning strategy switching when the covariance of the turbidity change rate and the vibration spectrum feature exceeds the dynamic covariance threshold;

[0011] Based on the spatio-temporal matching degree of the historical cleaning period and the current clogging feature matrix, decomposing the sub-feature sets corresponding to the high-frequency clogging mode and the low-frequency deposition mode;

[0012] Combining the cross-validation results of the fluid dynamics parameters and the mechanical wear parameters to correct the weights of the high-frequency clogging mode sub-feature set;

[0013] According to the corrected sub-feature set, reversely positioning the priority cleaning area along the filter structure topology graph, generating a three-dimensional cleaning path instruction and synchronously adjusting the opening degree of the blowdown valve.

[0014] Preferably, the multi-source sensing data collected during the operation of the self-cleaning filter specifically includes:

[0015] Synchronously acquiring the instantaneous pressure fluctuation curve of the filter inlet and outlet through a distributed pressure sensor;

[0016] capturing vortex frequency components of the fluid in the pipeline using an ultrasonic flow meter;

[0017] record the light transmittance gradient of the water body continuously using an optical turbidimeter;

[0018] Install a vibration acceleration sensor to collect the broadband vibration signal of the filter screen support.

[0019] Preferably, the hierarchical clustering of multi-source sensing data according to the preset blockage level threshold value comprises:

[0020] Time-domain alignment of peak interval of instantaneous pressure fluctuation curve and energy proportion of vortex frequency component;

[0021] Calculate the mutual information entropy value of the light transmittance gradient and the main frequency band of the broadband vibration signal;

[0022] Hierarchical clustering and division of pressure difference, turbidity and vibration characteristics based on mutual information entropy value.

[0023] Preferably, the trigger of adaptive cleaning strategy switching comprises:

[0024] Establish a sliding time window correlation model of the first derivative sequence of turbidity change rate and the second moment of vibration spectrum characteristics;

[0025] When the correlation model's determination coefficient is lower than the dynamic covariance threshold, start the composite mode of rotary cleaning and axial pulse cleaning.

[0026] Preferably, the decomposition of the sub-feature set corresponding to the high-frequency blockage mode and the low-frequency deposition mode comprises:

[0027] Extract the period when the pressure difference rise rate exceeds the critical rate in the historical cleaning period as the high-frequency blockage template;

[0028] Dynamically time warping match the current blockage feature matrix with the high-frequency blockage template to separate out the transient pulse type sub-feature.

[0029] Preferably, the weight correction of the high-frequency blockage mode sub-feature set comprises:

[0030] According to the Reynolds number in the fluid dynamics parameter, correct the effective action radius of the vortex frequency component;

[0031] Based on the material fatigue coefficient in the mechanical wear parameter, adjust the energy distribution weight of the vibration signal.

[0032] Preferably, the reverse positioning priority cleaning area comprises:

[0033] Construct the mapping relationship between filter screen pore distribution and fluid shear force in the filter structure topology graph;

[0034] The overlapping area of the eddy current frequency component and the pore distribution in the modified sub-feature set is identified as a primary cleaning target.

[0035] Preferably, the generation of the three-dimensional cleaning path instruction comprises:

[0036] Converting the coordinate information of the primary cleaning target into a polar angle-axial displacement control sequence of the cleaning suction nozzle;

[0037] Adjusting the execution time interval of the control sequence according to the real-time updated turbidity gradient.

[0038] Preferably, the synchronous adjustment of the blowdown valve opening degree comprises:

[0039] Dynamically calculating the blowdown flow demand according to the duration of the transient pulse type sub-feature;

[0040] Mapping the flow demand to a pulse width modulation signal to drive the blowdown valve by using a fuzzy logic controller.

[0041] Preferably, the method further comprises:

[0042] After completing the execution of the three-dimensional cleaning path instruction, re-collecting multi-source sensing data and verifying the decay rate of the clogging feature matrix;

[0043] When the decay rate does not reach a preset threshold, iteratively triggering secondary decomposition and weight modification of the high-frequency clogging mode sub-feature set.

[0044] Compared with the prior art, the present application has the following beneficial effects:

[0045] By collecting multi-source sensing data and conducting comprehensive analysis, the limitations of the existing self-cleaning filter control method are effectively broken through. It can comprehensively capture key information such as the inlet and outlet water pressure difference, flow fluctuation frequency, turbidity change rate and vibration frequency spectrum characteristics during the operation of the filter. These multi-dimensional data are no longer limited to the monitoring of a single parameter, but from multiple levels of overall operation state and local features, a complete data system reflecting the filter clogging situation is constructed, making the perception of the clogging state more comprehensive and accurate, avoiding the clogging judgment deviation caused by single data monitoring, and providing a rich and reliable information foundation for subsequent accurate control.

[0046] In the congestion feature analysis link, the method performs hierarchical clustering on the multi-source sensing data according to the preset congestion level threshold to generate a congestion feature matrix with a timestamp mark. This hierarchical clustering method can effectively distinguish and classify data of different congestion degrees and different congestion types. The addition of the timestamp enables the change of the congestion feature to have a time dimension traceability, facilitating clear understanding of the development process and change law of the congestion, and instead of the fuzzy judgment of the congestion state in the prior art, the method can accurately define the stage and trend of the congestion, providing a clear basis for subsequent adjustment of the cleaning strategy.

[0047] When the covariance of the turbidity change rate and the vibration spectrum feature exceeds the dynamic covariance threshold, the method can trigger the adaptive cleaning strategy switching. This design fully considers the dynamic change characteristics of the filter congestion state. The turbidity change rate reflects the real-time change of the fluid cleanliness, and the vibration spectrum feature implies the operating state change of the internal components of the filter due to congestion. The monitoring of the covariance of the two can sensitively capture the change of the congestion mode. Through the setting of the dynamic covariance threshold, the problem that the fixed threshold cannot adapt to different operating conditions is avoided, so that the cleaning strategy can be adjusted in time according to the actual change of the congestion mode, and the cleaning method is no longer fixed, effectively dealing with different properties and different degrees of congestion, and improving the flexibility and adaptability of the cleaning strategy.

[0048] Based on the spatio-temporal matching degree of the historical cleaning period and the current congestion feature matrix, the sub-feature sets corresponding to the high-frequency congestion mode and the low-frequency deposition mode are decomposed. This process realizes accurate classification of the congestion mode. The high-frequency congestion mode usually corresponds to short-term and sudden congestion, while the low-frequency deposition mode represents long-term and gradual congestion. Through the decomposition of the sub-feature set, different types of congestion can be clearly distinguished, and then the direction for subsequent targeted processing is provided. The weight of the high-frequency congestion mode sub-feature set is corrected by cross-verification of the fluid dynamics parameters and the mechanical wear parameters, further improving the accuracy and reliability of the sub-feature set. The fluid dynamics parameters can reflect the influence of the flow state of the fluid in the filter on the congestion, and the mechanical wear parameters consider the wear of the equipment components during the cleaning process. The cross-verification of the two makes the weight correction more scientific and reasonable, avoids the deviation that may be caused by single parameter correction, and ensures that the subsequent control action based on the sub-feature set is more accurate.

[0049] In terms of cleaning path planning and execution, the method locates the priority cleaning area along the filter structure topology graph in reverse direction according to the corrected sub-feature set, generates three-dimensional cleaning path instructions and synchronously adjusts the blowdown valve opening. The reverse direction locating of the priority cleaning area can accurately lock the area with more serious blockage, avoids the invalid cleaning of the area without or with slight blockage by the traditional fixed cleaning path, and greatly improves the cleaning efficiency. Compared with the traditional two-dimensional or fixed trajectory, the three-dimensional cleaning path instructions can more comprehensively and deeply clean the priority cleaning area, ensuring more thorough cleaning. At the same time, the synchronous adjustment of the blowdown valve opening can reasonably control the blowdown amount according to the actual needs in the cleaning process, reduce the waste of water resources, and reduce the operating cost while ensuring the cleaning effect. In addition, the entire control process is based on the Internet of Things sensing technology to realize real-time monitoring and dynamic adjustment, so that the operation of the filter is always in the optimal state, effectively prolongs the service life of the equipment, and improves the operation stability and economy of the entire filtration system. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 The working principle diagram of the intelligent control method of the self-cleaning filter based on Internet of Things sensing is described.

[0051] Figure 2 The flowchart of multi-source sensing data collection is described.

[0052] Figure 3 The flowchart of high-frequency and low-frequency blockage mode sub-feature set decomposition is described. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0054] Please refer to Figure 1The application provides an intelligent control method for a self-cleaning filter based on Internet of Things sensing, which comprises the following steps: after the collected data is preprocessed, hierarchical clustering analysis is performed according to a preset blockage level threshold, a blockage feature matrix with a time stamp is generated, and the matrix can dynamically reflect the running state of the filter. The system continuously monitors the change trend of the blockage feature matrix, and when the covariance of the turbidity change rate and the vibration frequency spectrum feature exceeds the dynamically set covariance threshold, the switching mechanism of the adaptive cleaning strategy is automatically triggered. Based on the spatiotemporal matching degree of the historical cleaning cycle data and the current blockage feature matrix, the system decomposes the sub-feature sets corresponding to the high-frequency blockage mode and the low-frequency deposition mode, thereby identifying different types of blockage behaviors. Combined with the cross-validation results of the fluid dynamics parameters and the mechanical wear parameters, the weight of the high-frequency blockage mode sub-feature set is corrected to optimize the cleaning decision. According to the corrected sub-feature set, the priority cleaning area is located along the filter structure topology graph in reverse, three-dimensional cleaning path instructions are generated, and the opening of the blowdown valve is adjusted synchronously, thereby realizing precise and efficient cleaning operation.

[0055] Embodiment 1: see Figure 2 The data acquisition link constitutes the sensing basis of the entire intelligent control system, and various types of sensors deployed at key positions of the filter are used to synchronously acquire running parameters. Distributed pressure sensors are precisely installed near the water inlet and outlet flanges of the filter. These sensors use high-precision strain gauge technology and can capture the instantaneous fluctuations of fluid pressure in the pipeline at a response speed of milliseconds. The pressure data is output in the form of continuous analog signals and is recorded as a digital pressure fluctuation curve with high-precision time stamps after analog-digital conversion, thereby providing raw data support for calculating the real-time water inlet and outlet pressure difference. The ultrasonic flowmeter is installed perpendicular to the pipeline axis, and its transducer works in an alternating emission and reception mode. By measuring the time difference of ultrasonic beams propagating in the forward and reverse directions, the average flow rate and instantaneous flow of the fluid can be calculated, and the vortex frequency components can be analyzed from flow noise through advanced signal processing technology. These frequency components reflect the vortex shedding phenomenon when the fluid passes through the filter screen, and their energy distribution is related to the blockage condition of the filter screen surface.

[0056] The probe of the optical turbidity meter is directly immersed in the outlet main pipeline of the filter. The probe emits a beam of visible light of a specific wavelength through the water to be measured, and uses a photoelectric detector on the opposite side to measure the attenuation of the transmitted light intensity. The concentration change of suspended particulate matter in the water body will cause continuous changes in the light transmittance. By calculating the change gradient of the light transmittance per unit time, the key parameter of the turbidity change rate can be obtained. The temperature compensation circuit built-in the optical turbidity meter ensures the stability of the measurement data when the water temperature fluctuates. The vibration acceleration sensor is firmly installed on the bearing seat or vibration conduction key point of the filter screen support through a magnetic base. It uses the piezoelectric effect principle to sense the wideband mechanical vibration of the filter screen generated under the action of fluid impact and cleaning mechanism. The vibration signal covers from low-frequency structural resonance to high-frequency fluid impact noise. The complete vibration spectrum characteristics are converted from time domain signal to frequency domain for analysis through fast Fourier transform. All these sensor nodes are connected to the data acquisition gateway through the industrial Internet of Things protocol, realize millisecond-level time synchronization of data aggregation, and form a continuous multi-source sensing data stream.

[0057] After obtaining the original sensing data, the system performs a hierarchical clustering operation according to the preset clogging level threshold to generate a clogging feature matrix. This process first requires time-domain alignment of heterogeneous data, accurately matching the peak interval sequence of the instantaneous pressure fluctuation curve with the energy proportion sequence of the eddy current frequency component. The time alignment algorithm is based on the dynamic time warping principle, compensating for the possible small time delay caused by different sensor data acquisition and transmission paths, and ensuring that all feature parameters reflect the filter state on the same time reference. Subsequently, the system calculates the mutual information entropy value between the time sequence of the light transmittance change gradient and the energy of the main frequency band of the wideband vibration signal (such as the vibration energy integral of a specific high frequency band or low frequency band). The calculation of mutual information entropy value does not depend on linear assumption, but quantifies the statistical dependence between the two random variables by estimating the KL divergence of their joint probability distribution and respective marginal probability distribution. A higher mutual information entropy value indicates that the abnormal change of turbidity and the specific vibration mode of the filter screen structure are strongly associated, often indicating that a certain type of clogging is occurring.

[0058] Based on the calculated mutual information entropy values, the system performs hierarchical clustering division on multiple dimensional feature parameters such as the water inlet and outlet pressure difference, turbidity change rate, and vibration frequency spectrum characteristics. The clustering algorithm uses a bottom-up aggregation hierarchical clustering method, with mutual information entropy values as the basis for measuring the similarity between feature vectors, and a pre-set clogging level threshold as the distance criterion for controlling the clustering and merging process. Low-level clogging may only show a slight increase in pressure difference, while high-level clogging is usually accompanied by simultaneous abnormalities in turbidity change rate and specific vibration frequency band energy. The clustering process automatically merges data points with similar mutual information entropy patterns into the same cluster, and each cluster represents a clogging state or development stage. The final generated clogging feature matrix is a structured data object, with row vectors corresponding to each sampling time point and column vectors containing standardized feature values after clustering division. Each feature vector is labeled with an accurate timestamp, and the matrix is dynamically updated to provide standardized and quantitative state input for subsequent trend analysis, pattern recognition, and cleaning decisions.

[0059] Example 2: refer to Figure 3 The real-time monitoring function relies on a continuously running state analysis engine that processes the continuously updated clogging feature matrix in a sliding window manner. Each feature vector in the matrix is labeled with an accurate timestamp, allowing the system to track the evolution of pressure difference, turbidity change rate, and vibration frequency spectrum characteristics over time. To quantify the dynamic relationship between these parameters, the system establishes a sliding time window correlation model between the first derivative sequence of turbidity change rate and the second moment of vibration frequency spectrum characteristics. The first derivative sequence is obtained by calculating the difference between adjacent time points of turbidity change rate, reflecting the acceleration direction and intensity of turbidity change. The second moment of vibration frequency spectrum characteristics describes the distribution of vibration energy in the frequency domain and is an indicator of vibration state stability. The sliding time window correlation model uses a fixed-length data window, with the window width set according to the typical operating period of the filter. At each analysis time, the model calculates the Pearson correlation coefficient or similar linear correlation index between the first derivative sequence and the second moment sequence within the current window. This correlation coefficient is called the decision coefficient, and its value range reflects the linear correlation strength between the two sequences within a specific time window. The dynamic covariance threshold is not a fixed value, but a boundary value dynamically adjusted by machine learning algorithms based on historical operating data, current inlet water quality load, and cumulative equipment operating time. When the decision coefficient of the correlation model continuously falls below this dynamic threshold, it indicates that the abnormal change pattern of turbidity and the mechanical vibration response of the filter screen have decoupled significantly. This decoupling often indicates uneven distribution of clogging material on the filter screen surface or the formation of special adhesive structures. At this time, the system determines that the current state requires intervention, triggering the switching mechanism of the adaptive cleaning strategy.

[0060] After triggering the adaptive cleaning strategy switching, the control system switches from the basic one-way flushing or timed flushing mode to the composite mode combining rotary cleaning and axial pulse cleaning. The execution mechanism of rotary cleaning is a rotary spray arm installed inside the filter screen, which is accurately driven by a servo motor. The rotary speed can be infinitely adjusted according to the severity of the blockage. During the rotation, the nozzles on the spray arm are aimed at the inner surface of the filter screen to spray high-pressure water flow to peel off the dirt. Axial pulse cleaning is achieved through a pulse generator connected in parallel with the main water inlet pipeline. The pulse generator is usually a secondary pipeline controlled by a high-speed electromagnetic valve, which can intermittently inject short-time, high-pressure fluid pulses into the filter screen. Such pulses can generate instantaneous pressure difference fluctuations and fluid shear forces inside the filter screen, which help to shake off and flush away stubborn blockages. Rotary cleaning and axial pulse cleaning need to be cooperatively controlled in time sequence. For example, the axial pulse can be triggered synchronously when the rotary spray arm passes through the identified key blockage area, thereby forming a composite cleaning effect with precise spatial positioning and time coordination. The switching decision of this strategy is issued by the central controller in real time, and the feedback signals of the execution mechanism are monitored to ensure that the action is in place. In the parallel path of triggering the cleaning strategy switching, the system starts the decomposition process of the sub-feature set corresponding to the high-frequency blockage mode and the low-frequency deposition mode. This decomposition process strongly depends on the mining and comparison of historical data. The system extracts data recorded in the past multiple cleaning cycles from the historical database. This critical rate is an engineering threshold calculated based on the maximum working pressure difference allowed by the filter and the safety margin. After extracting these data segments, they are processed and standardized to form one or more representative "high-frequency blockage templates". These templates are essentially feature patterns that describe the typical change patterns of parameters such as pressure, turbidity, and vibration when rapid blockage occurs. Next, the system dynamically time warps the current blockage feature matrix calculated in real time with these historical high-frequency blockage templates. Dynamic time warping is a nonlinear time series alignment algorithm that can overcome the stretching, distortion, and other deformation problems of different time series on the time axis, find the optimal correspondence between two sequences, and evaluate their similarity by calculating the minimum cumulative distance between the current sequence and the template sequence.

[0061] Based on the results of dynamic time warping matching, the system can separate the "transient pulse type sub-features" from the current complex mixed plugging characteristics, which are highly similar to the historical high-frequency plugging templates. These sub-features usually correspond to the rapid plugging events caused by a large number of particles impacting the filter screen in a short time, characterized by rapid pressure difference rise and significant increase in specific high-frequency vibration energy. The sub-features with lower matching degree and relatively slow changes are classified as the "low-frequency deposition mode" corresponding sub-feature set. This mode is usually caused by the long-term adsorption, deposition of fine particles or slow growth of biofilm. The two types of separated sub-feature sets will be given different processing priorities and cleaning strategy parameters. The high-frequency plugging mode usually requires immediate response and strong cleaning action due to its suddenness and rapid impact on system pressure loss, while the low-frequency deposition mode can be gradually eliminated by a more gentle and longer period cleaning method. This accurate decomposition makes the cleaning action more targeted and avoids waste of energy and resources. The implementation of the whole process relies on efficient time series analysis algorithms, sufficient capacity of historical database and computing power capable of processing real-time data, ensuring the accuracy and timeliness of pattern decomposition.

[0062] Taking a self-cleaning filter for an industrial cooling water circulation system as an example, the system has been handling cooling water containing trace amounts of oil and inorganic salt particles for a long time. During a certain period of time, the sensor data shows that the filter operating parameters start to fluctuate abnormally. The real-time monitoring module detects that the dynamic change trend of the plugging characteristic matrix appears a certain pattern, the inlet and outlet pressure difference shows a stepwise slow rise, and the light transmittance change gradient recorded by the optical turbidity meter appears multiple rapid drops followed by slow recoveries in a certain period of time. The vibration acceleration sensor captures the vibration energy of the filter support in the 800Hz to 1200Hz frequency band, which continuously increases but is accompanied by intermittent peaks. The system establishes a correlation model between the first derivative sequence of turbidity change rate and the second moment of vibration frequency spectrum characteristics using a built-in algorithm, which uses a time window of 5 minutes, slides every 10 seconds and recalculates the determination coefficient of the two sequences in the window. During the monitoring period of half an hour, the system finds that the determination coefficient gradually decreases from the initial 0.85 to 0.45, and the current dynamic covariance threshold automatically calculated according to the inlet water quality load is 0.60. When the determination coefficient first falls below 0.60, the state analysis engine determines that the current operating state is abnormal, triggering the switching instruction of the adaptive cleaning strategy, and the control system immediately issues an instruction to the cleaning execution mechanism to switch from the conventional low-pressure backflushing mode to the composite mode of rotary cleaning and axial pulse cleaning.

[0063] After the composite cleaning mode is started, the rotating cleaning unit is first activated, driving the spray arms inside the filter to rotate at a uniform speed of 15 revolutions per minute, covering the entire filtering circumferential surface. Almost simultaneously, the axial pulse cleaning module begins to work, which controls a high-speed electromagnetic valve to inject high-pressure water pulses with a duration of 0.3 seconds into the main pipeline at a frequency of 2 times per second. While the composite cleaning mode is running, the system performs a parallel operation of decomposing the corresponding feature set of the high-frequency clogging mode and the low-frequency deposition mode. This operation first accesses the historical database to retrieve all recorded cleaning events in the past three months, and extracts twelve typical "high-frequency clogging templates" from them. The common feature of these templates is that the pressure difference rise rate exceeds the critical rate of 0.1 bar / min, and is accompanied by a sharp fluctuation in turbidity. The system performs dynamic time warping matching between the current real-time constructed clogging feature matrix and these historical templates. The dynamic time warping algorithm finds the most matching path between the current data sequence and each template sequence by stretching or compressing the time axis. The matching result finds that the data in the current matrix from about 10:15 to 8 minutes later is highly similar to template No. 7 in the historical template library, which records a rapid clogging event caused by a short-term large influx of oil. The cumulative matching error is below the set tolerance. Based on this matching result, the system successfully separates a group of "transient pulse type sub-features" from the current complex mixed features. This group of sub-features clearly shows that during the period from 10:15 to 10:23, the pressure difference rise rate suddenly accelerates, the turbidity change rate derivative appears two significant negative pulses, and the vibration energy near 800 Hz appears a synchronous sharp peak.

[0064] The separated transient pulse type sub-features are marked as high frequency jamming mode, while the slow but continuous rising trend of pressure difference after 10:23 and the steady accumulation of low frequency energy below 200Hz in the vibration signal are classified as the sub-feature set of low frequency deposition mode. This decomposition immediately affects the fine-tuning of the cleaning strategy, the control system judges that the high frequency jamming event may be related to a short-term process fluctuation, and the corresponding contaminants may have been effectively removed by the current composite cleaning action, while the low frequency deposition mode reflects the inherent accumulation of long-term operation. Therefore, the system dynamically adjusts the cleaning parameters on the basis of maintaining the composite cleaning mode: for the filter screen area corresponding to the identified high frequency jamming period, an additional set of higher intensity axial pulses is triggered when the spray arm rotates through this area, in order to more thoroughly remove the residues of this impulsive jamming. For the low frequency deposition mode, the system plans to automatically start a longer period and lower intensity preventive cleaning program after the end of the current high intensity composite cleaning, aiming to gradually remove chronic deposits. Throughout the process, sensor data is continuously monitored and the jamming feature matrix is constantly updated, providing real-time feedback for dynamic adjustment of the cleaning strategy. This example shows how to achieve complete closed-loop control from triggering cleaning to precise optimization of the cleaning strategy through real-time data analysis, historical pattern matching and feature decomposition.

[0065] Embodiment 3: The weight correction process is a fine decision-making link that integrates multi-physical field information, and its core is to adjust the importance proportion of different sub-features in subsequent decision-making by combining the cross-validation results of fluid dynamics parameters and mechanical wear parameters. Fluid dynamics parameters are mainly derived from real-time calculation and analysis of the fluid state in the pipeline, among which Reynolds number is a key indicator, which is calculated by the basic parameters of fluid density, average flow rate, characteristic length and dynamic viscosity of the fluid. The value of Reynolds number directly determines whether the fluid flow is laminar or turbulent, and further affects the generation, development and decay characteristics of vortex in the filter. The system uses the calculated Reynolds number to correct the parameter of "effective action radius of vortex frequency component" in the high frequency jamming mode sub-feature set. In the laminar flow state of low Reynolds number, the scale of vortex is small and decays quickly, and its effective action radius is relatively limited. In the turbulent flow state of high Reynolds number, the vortex energy is stronger and the influence range is wider, and the effective action radius will increase significantly. The correction process is realized through a mapping function based on historical data and fluid mechanics model, which maps the Reynolds number to a correction factor to scale the weight of the feature item representing the influence of vortex in the sub-feature set, so that the system's judgment of the jamming position is more consistent with the actual fluid mechanics environment.

[0066] The introduction of mechanical wear parameters adds a time dimension and equipment health state consideration to the weight correction. The mechanical wear parameters are obtained by deep analysis of long-term collected vibration spectrum characteristics. The "material fatigue coefficient" is a comprehensive index that reflects the cumulative damage degree of key load-bearing components such as filter screens and supports under long-term alternating stress. The estimation of the material fatigue coefficient relies on long-term trend analysis of vibration signals, especially the frequency band energy related to the structure's natural frequency, combined with the cumulative length of equipment operation. Through a special wear model, the system evaluates the current structure's fatigue state. Based on the calculated material fatigue coefficient, the system adjusts the weight of the features related to vibration signals in the high-frequency blockage mode sub-feature set. For example, when the material fatigue coefficient is high, indicating a decline in the structure's health, the system appropriately reduces the absolute weight of the high-frequency vibration energy feature. Because at this time, stronger vibration may be partially caused by mechanical problems such as structural loosening, rather than completely caused by blockage. At the same time, the weight proportion of pure fluid feature items based on pressure difference and turbidity change may be correspondingly increased. This dynamic adjustment avoids misjudgment due to changes in the equipment's own state, making the cleaning decision more stable and reliable.

[0067] After the weight correction of the high-frequency clogging pattern sub-feature set is completed, the system initiates the process of reverse positioning of the priority cleaning area, the core of which is to establish a spatial mapping relationship between the internal physical structure of the filter and the fluid-clogging interaction. The system first needs to construct an accurate filter structure topology map, which is a three-dimensional digital model that not only includes the overall geometric shape and installation orientation of the filter screen, but more importantly, details the microscopic structure information of the pore distribution density, pore size, and support rib layout of different areas on the filter screen. This model is usually generated during the filter design stage and stored as static data in the control system. Subsequently, the system estimates the fluid shear force distribution when flowing through different positions of the filter screen based on real-time collected flow and pressure data, combined with computational fluid dynamics principles. The size of the fluid shear force is directly related to the flow velocity gradient and is the main driving force for flushing the clogging material. Establishing a mapping relationship between the filter screen pore distribution and the fluid shear force means that the physical structure features and the fluid dynamic environment are spatially related, for example, the low-porosity area overlaps with the low-shear force area, which is usually considered as a high-risk area that is prone to clogging and difficult to clean. The system uses the corrected high-frequency clogging pattern sub-feature set, especially the "vortex frequency component" feature corrected by Reynolds number, to identify overlapping areas. The specific distribution pattern of the vortex frequency component in the frequency spectrum carries information about the location of the clogging material, because the clogging material will disturb the flow field and change the characteristic frequency of the local vortex. The system analyzes the corrected vortex frequency feature in the frequency-space domain and maps it to the filter structure topology map to form a "vortex influence intensity" distribution map. The final positioning operation is to identify the overlapping areas between the "vortex influence intensity distribution map" and the previously established "pore-shear force mapping relationship map". Specifically, the system looks for areas that meet the following conditions: first, the area shows abnormally high values in the vortex influence intensity distribution map, indicating strong flow disturbance, which is likely caused by clogging; second, the area corresponds to a high clogging risk in the structure topology map; and third, the area is relatively weak in fluid shear force distribution and lacks self-cleaning ability. These overlapping areas that meet multiple unfavorable conditions are marked as "primary cleaning targets" by the system, which are the key positions that need to be cleaned first and thoroughly. This positioning method integrates real-time sensor data, fluid mechanics principles, and device structure information, achieving precise reverse mapping from abstract "features" to specific "spatial locations".

[0068] The entire weight correction and reverse positioning process is achieved through a comprehensive evaluation function containing multiple parameters, which quantifies and integrates the above factors. The function form for evaluating the cleaning urgency of each potential grid cell (small area after discretizing the filter screen) is as follows:

[0069]

[0070] wherein: represents the cleaning priority index of the grid cell, the higher the index value, the more the cell needs to be prioritized for cleaning. is the total number of high-frequency clogging sub-features under consideration. is the initial importance coefficient assigned to the sub-feature, which is set according to the base impact of the feature at system initialization. is the dynamic weight factor specially designed for the sub-feature, which is self-adaptively adjusted according to real-time fluid dynamics conditions (e.g. Reynolds number) and equipment mechanical wear state, embodying the dynamic revision of the weight. represents the feature intensity representation value of the sub-feature on the grid cell, which is derived from the interpolation or inversion calculation of sensor data in space. is the fluid dynamics correction coefficient, which is usually a function of Reynolds number under current working conditions, for quantifying the influence of fluid state on clogging and cleaning. is the mechanical wear correction coefficient, which reflects the influence degree of equipment health state indicators such as material fatigue coefficient on feature reliability and cleaning decision. The control system can automatically screen out the first-level cleaning targets that need to be prioritized by traversing the values of all discrete grid cells on the filter, and setting a threshold value, providing direct spatial coordinate input for generating accurate three-dimensional cleaning path.

[0071] Embodiment 4: The process of generating three-dimensional cleaning path instructions starts with the processing of first-level cleaning target coordinate information, which is defined in the structural coordinate system of the filter itself, usually a cylindrical coordinate system with the center axis of the filter as the reference, and the coordinates contain three dimensions of radial distance, polar angle and axial height. The control system needs to convert these static three-dimensional coordinate information into a motion control sequence of the cleaning execution mechanism, usually a rotatable and axially movable spray and suction arm, and the core of this conversion is to perform coordinate transformation to decompose the spatial position of the target point (radial R, polar angle θ, axial Z) into two basic actions that the spray and suction arm needs to perform: rotational motion around the axis (corresponding to polar angle θ) and linear motion along the axis (corresponding to axial Z). Since the radial position of the spray and suction arm is usually fixed, the motion control sequence is mainly composed of a series of polar angle-axial displacement pairs, and each pair (θ, Z) instruction directs the spray and suction arm to move to a specific position on the inner surface of the filter.

[0072] Considering the cleaning efficiency and smoothness, the generated original path point sequence also needs to be optimized, the system will adopt path planning methods such as nearest neighbor algorithm or genetic algorithm to sort all target points, the goal is to make the total path of the spray and suction arm moving from one point to the next point shortest, the motion time is the least, and avoid unnecessary reciprocating motion, the optimized path sequence ensures the efficiency of the cleaning action. However, this motion sequence is not immutable, it will be dynamically adjusted by the real-time updated turbidity gradient, the turbidity gradient reflects the concentration change rate of the discharged sewage during the cleaning process, the control system will set a reference range of the turbidity gradient, when the real-time monitored gradient value is at a high level, it means that the current cleaning position is discharging a large amount of pollutants, the system will appropriately prolong the cleaning residence time at this position and its adjacent area, that is, increase the execution time interval of the control sequence at this point; on the contrary, if the turbidity gradient rapidly decreases to a lower level, it means that the pollutants in this area have been basically removed, the system will shorten the residence time, or even end the cleaning of this point in advance, and move to the next target point quickly. This dynamic speed regulation mechanism based on turbidity feedback makes the cleaning process have a certain adaptability, which can allocate time and resources according to the actual cleaning effect. While generating and executing the three-dimensional cleaning path instructions, the system needs to adjust the opening of the blowdown valve to match the cleaning action, the control of the blowdown valve is based on the blowdown flow demand dynamically calculated by the duration of the transient pulse type sub-feature. The transient pulse type sub-feature usually corresponds to high-intensity, short-time cleaning events, such as the period when axial pulse cleaning occurs, the duration of such events can be directly obtained from the cleaning strategy instructions, the calculation model of the blowdown flow demand will comprehensively consider the estimated amount of pollutants that may be stirred up by each pulse cleaning, the design flow of the filter and the baseline blowdown amount required to maintain the normal pressure of the system, a simplified calculation example is as follows: assuming that in a pulse cleaning with a duration of Δt, according to the historical data model, the estimated volume of additional pollutant suspension that needs to be discharged is V, then the average additional blowdown flow demand during this period can be represented as The total instantaneous blowdown flow demand is equal to the basic blowdown flow The sum of which. To convert the calculated continuous flow demand into a driving signal for the valve actuator, the system employs a fuzzy logic controller, whose input variables usually include "flow demand error" (the difference between the calculated demand and the actual flow sensor feedback) and "error rate of change", and the output variable is the duty cycle of the pulse width modulation signal driving the blowdown valve. The controller has pre-set fuzzy rules inside, such as "if flow demand error is positive and large, and error rate of change is positive, then increase PWM duty cycle substantially", through the process of fuzzy reasoning and defuzzification, the continuous flow demand is mapped to a smoothly varying PWM signal, which precisely drives the opening degree of the electric or pneumatic blowdown valve, ensuring that during the cleaning blowdown phase, the sewage can be discharged in time, avoiding secondary pollution, and during the non-intensive cleaning phase, the blowdown volume is maintained at a low level to reduce water waste.

[0073] To illustrate the process more specifically, consider a simplified example scenario, assume that the filter is divided into several axial regions, after the positioning of embodiment 3, the system identifies three main primary cleaning target points, their initial coordinates, the optimized motion sequence, and the planned residence time adjusted according to the real-time turbidity gradient, refer to Table 1. It should be noted that the point sequence and coordinates in the actual system will be much more complex, this table is only for illustration.

[0074] Table 1: Cleaning path planning and dynamic adjustment table

[0075]

[0076] According to the table, the cleaning execution process is as follows: the spray-suction arm first moves to point A (θ = 120°, Z = 1.2 m), and starts cleaning according to the instruction, while the turbidity sensor monitors the turbidity of the discharged water. Since the feedback shows that the turbidity gradient remains high, indicating that the point is severely clogged and is being effectively cleaned, the control system automatically extends the originally planned 10-second dwell time to 15 seconds to ensure the cleaning effect. When the cleaning at point A is approaching the end, the system has pre-instructed the blowdown valve to adjust the opening to a larger state according to the cleaning intensity and time at point A. After completion, the spray-suction arm moves to point B (θ = 45°, Z = 0.8 m) according to the optimized order, and the initial turbidity gradient is also high. The system plans to execute 12 seconds of cleaning (with slight extension), but after 6 seconds of cleaning, the turbidity gradient begins to drop significantly, indicating that the contaminants are rapidly decreasing. The control system judges that there is no need to use the full 12 seconds, so it ends the cleaning at point B at the 8th second in advance, and instructs the blowdown valve to reduce the opening as the cleaning intensity decreases. Subsequently, the spray-suction arm moves to point C (θ = 270°, Z = 1.5 m), where the turbidity gradient has been at a low level, indicating that the clogging is relatively light. The system therefore shortens the cleaning time to 8 seconds, and the blowdown valve is maintained at a relatively small opening. Throughout the process, the execution of the three-dimensional cleaning path is closely coupled with the adjustment of the blowdown valve opening, forming a closed-loop control system that works cooperatively.

[0077] Example 5: After the three-dimensional cleaning path instruction is completely executed, the blowdown valve is closed, and the system returns to the normal filtering state for a predetermined stable time, the verification program of the intelligent control system is started. The first action of this program is to re-collect a complete set of representative multi-source sensor data. The standards for this round of data collection are exactly the same as before cleaning, including obtaining the latest water pressure difference curve through the distributed pressure sensor, measuring the flow fluctuation frequency after the system returns to steady state using the ultrasonic flow meter, recording the light transmittance and its change gradient of the water body after cleaning using the optical turbidity meter, and collecting the broadband vibration signal of the filter screen support again through the vibration acceleration sensor. The data collected this time constitutes the original data set of the "post-cleaning state", and its purpose is to form a comparable benchmark with the "clogging feature matrix" recorded before cleaning.

[0078] The new data collected is immediately sent to the data processing module for the same feature extraction and construction process as before, generating a new "cleaned congestion feature matrix" with the current timestamp. The dimensions, structure, and feature meaning of this new matrix are identical to the matrix generated before cleaning, ensuring comparability. The core of the verification process is to calculate the "decay rate" of the new matrix relative to the pre-cleaning matrix. The decay rate is a comprehensive quantitative indicator that is not simply the percentage decline of a single parameter, but a measure of the overall decline of the system's abnormal state represented by the entire feature matrix. When calculating the decay rate, the system will find the state vector of the same length in the post-cleaning matrix corresponding to each feature vector in the pre-cleaning matrix, then use vector norm calculation or feature weighted Euclidean distance algorithm to quantify the difference between each pair of vectors, and finally statistically integrate the difference values at all time points to obtain a total decay rate value. This decay rate value reflects the overall effect of the cleaning operation, and the higher the value, the more thorough the cleaning and the better the filter state recovery. The system compares the calculated actual decay rate with a preset threshold, which is determined during filter design or long-term operation optimization and represents the minimum effect standard expected to be achieved by this cleaning. It is a key parameter related to system efficiency and safety. If the actual decay rate reaches or exceeds the preset threshold, the system determines that the cleaning task is successfully completed, and the filter can continue to operate normally. The system archives all data from this cleaning, including cleaning parameters, paths, and the final effect decay rate, into the historical database for future learning samples.

[0079] If the actual attenuation rate does not reach the preset threshold, it means that the cleaning effect is not ideal, and the state of the filter fails to fully recover to the expected level, and there may be residual stubborn blockages or improper selection of cleaning strategies. At this time, the system will not immediately start another identical cleaning cycle, but will trigger an iterative process that is more targeted. The core of this iterative process is "secondary decomposition and weight correction". The system will return to the analysis step of the high-frequency blockage mode sub-feature set, but the input conditions for this analysis have changed. First, the system will consider the just-completed cleaning action, which was not effective, as a new important event and include it in the analysis. The cleaning path, cleaning intensity, and real-time turbidity data all provide new clues to understanding the characteristics of the blockage. Based on these new information, the system performs secondary decomposition on the high-frequency blockage mode sub-feature set in the current (i.e., after cleaning but still not ideal) state. This decomposition may use different time window scales or more refined matching algorithms, aiming to separate stubborn blockage features that are insensitive to the previous cleaning strategy from the residual signal. In an actual scenario of chemical circulating water filtration, after the initial cleaning, the pressure difference decreased significantly but did not fully recover to normal, the energy in a certain specific medium-high frequency band in the vibration spectrum was reduced but still higher than the baseline, and the turbidity change rate also showed small fluctuations. The overall attenuation rate calculated did not reach the threshold. The system determines that the cleaning is not complete and triggers the iterative process. In the secondary decomposition, the system finds that the residual vibration features are highly similar to those of a known "low-frequency deposition mode" formed by the slow deposition of viscous polymers, but were not given priority in the initial decomposition due to their weak energy and slow changes. Based on this finding, the system performs weight correction on the high-frequency blockage mode sub-feature set, significantly reducing the weight of the specific vibration frequency band feature, as analysis shows that it is not the main contradiction of the residual blockage; at the same time, the system increases the analysis weight of the small "plateau" in the pressure difference recovery curve, which suggests the presence of a local penetrating blockage. Subsequently, the iterative decision is generated: the system decides not to repeat the comprehensive three-dimensional path cleaning, but generates a targeted and local supplementary cleaning path that only targets the specific axial height area suspected of being a penetrating blockage, uses a higher pressure point pulse cleaning mode, and appropriately extends the cleaning action time at that point. After this iterative cleaning is completed, the system will again collect data for effect verification and calculate a new attenuation rate. If it still does not meet the standard, it can theoretically continue to iterate, but usually will alarm when it reaches the safe operation threshold or the number of iterations exceeds the limit, prompting the need for human intervention. This iterative mechanism based on effect verification enables the cleaning control to have a certain learning and adaptive ability, gradually approaching the optimal cleaning strategy.

[0080] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0081] While the embodiments of the application have been shown and described herein, it is to be understood that the scope of the application, jointly pointed out in the appended claims, is not to be limited to the above-described embodiments but can be otherwise variously changed, modified, replaced, and altered within the principles and spirit of the present application.

Claims

1. An Internet of Things (IoT) perception-based self-cleaning filter intelligent control method, characterized in that, The method comprises the following steps: Collecting multi-source sensing data from the operation of the cleaning filter, the multi-source sensing data including an inlet-outlet water pressure difference, a flow fluctuation frequency, a turbidity change rate, and a vibration spectrum feature; Hierarchical clustering of the multi-source sensing data according to preset blockage level thresholds to generate a blockage feature matrix containing time stamp markers; Real-time monitoring of the dynamic change trend of the blockage feature matrix, and triggering adaptive cleaning strategy switching when the covariance of the turbidity change rate and the vibration spectrum feature exceeds a dynamic covariance threshold; Decomposing a sub-feature set corresponding to a high-frequency blockage mode and a low-frequency deposition mode based on the spatiotemporal matching degree of the historical cleaning period and the current blockage feature matrix; Weight correction of the high-frequency blockage mode sub-feature set based on the cross-validation results of the fluid dynamics parameters and the mechanical wear parameters; Reverse positioning of a preferential cleaning area according to the corrected sub-feature set along a filter structure topology graph, generation of a three-dimensional cleaning path instruction, and synchronous adjustment of the blowdown valve opening degree. 2.The IoT-aware self-cleaning filter intelligent control method of claim 1, wherein, The multi-source sensing data collected from the operation of the cleaning filter specifically includes: Synchronous acquisition of the instantaneous pressure fluctuation curve of the filter inlet and outlet by a distributed pressure sensor; Capture of the vortex frequency component of the fluid in the pipeline by an ultrasonic flowmeter; Continuous recording of the water body light transmittance change gradient by an optical turbidimeter; Installation of a vibration acceleration sensor to collect the wideband vibration signal of the filter screen support. 3.The IoT-aware self-cleaning filter intelligent control method of claim 2, wherein, The hierarchical clustering of the multi-source sensing data according to the preset blockage level thresholds includes: Time domain alignment of the peak interval of the instantaneous pressure fluctuation curve and the energy proportion of the vortex frequency component; Calculation of the mutual information entropy value of the light transmittance change gradient and the main frequency band of the wideband vibration signal; Hierarchical clustering and division of the pressure difference, turbidity, and vibration features based on the mutual information entropy value. 4.The IoT-aware self-cleaning filter intelligent control method of claim 3, wherein, The adaptive cleaning strategy switching includes: Establishment of a sliding time window correlation model of the first derivative sequence of the turbidity change rate and the second moment of the vibration spectrum feature; When the correlation model's determination coefficient is lower than the dynamic covariance threshold, start the composite mode of rotary cleaning and axial pulse cleaning. 5.The IoT-aware self-cleaning filter intelligent control method of claim 4, wherein, The decomposition of the sub-feature set corresponding to the high-frequency blockage mode and the low-frequency deposition mode includes: Extracting the period when the pressure difference rising rate exceeds the critical rate in the historical cleaning period as a high-frequency blockage template; Dynamic time warping matching of the current blockage feature matrix and the high-frequency blockage template to separate out the transient pulse type sub-feature. 6.The IoT-aware self-cleaning filter intelligent control method of claim 5, wherein, The weight correction of the high-frequency blockage mode sub-feature set includes: According to the Reynolds number in the fluid dynamics parameters, correct the effective action radius of the vortex frequency component; Based on the material fatigue coefficient in the mechanical wear parameters, adjust the energy distribution weight of the vibration signal. 7.The IoT-aware self-cleaning filter intelligent control method of claim 6, wherein, The reverse positioning of the preferential cleaning area includes: Constructing the mapping relationship between the filter screen pore distribution and the fluid shear force in the filter structure topology graph; Identifying the overlapping area of the vortex frequency component and the pore distribution in the corrected sub-feature set as the primary cleaning target. 8.The IoT-aware self-cleaning filter intelligent control method of claim 7, wherein, The generation of the three-dimensional cleaning path instruction includes: Converting the coordinate information of the primary cleaning target into a polar angle-axial displacement control sequence of the cleaning suction nozzle; Adjusting the execution time interval of the control sequence according to the real-time updated turbidity gradient. 9.The IoT-aware self-cleaning filter intelligent control method of claim 8, wherein, The synchronous adjustment of the blowdown valve opening degree includes: The emission flow demand is dynamically calculated according to the duration of the transient pulse type sub-feature; A fuzzy logic controller is used to map the flow demand to a pulse width modulation signal to drive the emission valve. 10.The IoT-aware self-cleaning filter intelligent control method of claim 1, wherein, Further comprising: After completing the three-dimensional cleaning path instruction execution, re-acquire multi-source sensing data and verify the decay rate of the blockage feature matrix; When the decay rate does not reach the preset threshold, iteratively trigger the secondary decomposition and weight correction of the high-frequency blockage mode sub-feature set.

Images