Intelligent control method and system for micro-interface oil removal device based on backwashing circulation

Abstract

The invention provides an intelligent control method and system for a micro-interface oil removal device based on backwashing circulation, and relates to the technical field of intelligent control, and the method comprises the steps: obtaining real-time operation state data and historical backwashing data of the micro-interface oil removal device, constructing a pollution state evolution model considering a space-time coupling effect, and obtaining a pollution state evolution model; predicting a pollutant accumulation trend and a diffusion rule of each region to obtain a regional pollution evaluation result with time sequence characteristics; constructing a multi-objective optimization function in combination with a system backwashing resource constraint condition, and generating an initial pulse sequence scheme; performing dynamic reconstruction on the initial pulse sequence scheme based on pollutant spatial distribution characteristics and a time-varying rule to form an optimized pulse sequence with spatial suitability; and establishing a grading evaluation criterion in a backwashing operation execution process, performing zoning quantitative evaluation on different areas, and selectively adjusting pollution state evolution model parameters. According to the invention, intelligent optimization of a backwashing strategy is realized, and the oil removal efficiency and the backwashing resource utilization rate are improved.

Description

Technical Field

[0001] This invention relates to intelligent control technology, and more particularly to an intelligent control method and system for a micro-interface oil removal device based on backwashing cycle. Background Technology

[0002] Micro-interface oil removal technology is an important technique in industrial water treatment. It achieves oil-water separation at a microscale interface and is widely used in the treatment of oily wastewater in industries such as petrochemicals, machinery manufacturing, and metal processing. This technology utilizes special interface materials and structural designs to capture and separate oil from water at the microscopic level, offering advantages such as high separation efficiency, small footprint, and low operating costs. However, during long-term operation, contaminants gradually accumulate on the surface of the micro-interface material, leading to a decline in separation performance. Therefore, periodic backwashing is necessary to restore its oil removal capacity. The backwashing circulation system removes contaminants adhering to the micro-interface through reverse water or airflow impact, and is a key element in maintaining stable operation of the device.

[0003] Traditional backwash control methods primarily rely on fixed time intervals or differential pressure thresholds for triggering. This approach struggles to adapt to the complexity and non-uniformity of contaminant accumulation in real-world operations. In practical applications, the degree of contamination varies significantly across different regions of the micro-interface. The accumulation rate and distribution characteristics of contaminants are influenced by various factors, including fluctuations in influent water quality, flow field distribution, and local velocity differences, exhibiting pronounced spatiotemporal coupling characteristics. Furthermore, backwashing itself is a dynamic process requiring coordinated consideration of multiple parameters such as backwash intensity, backwash duration, and pulse frequency to ensure effective cleaning while minimizing water and energy consumption.

[0004] In existing technologies, backwashing control methods for micro-interface oil removal devices generally employ a uniform backwashing strategy, applying the same intensity and rhythm of cleaning operations to the entire micro-interface region, failing to fully consider the differences in contamination states in different areas. This "one-size-fits-all" control approach leads to over-cleaning of lightly contaminated areas, wasting backwashing resources, while heavily contaminated areas are under-cleaned, failing to effectively restore their separation performance. Furthermore, existing control methods lack the ability to dynamically predict the contaminant accumulation process, failing to adjust the backwashing strategy in advance based on the evolution trend of the contamination state. They can only passively respond after contamination accumulates to a certain level, resulting in periodic fluctuations in device performance and affecting overall operational stability. Summary of the Invention

[0005] The present invention provides an intelligent control method and system for a micro-interface oil removal device based on backwashing cycle, which can solve the problems in the prior art.

[0006] A first aspect of the present invention provides an intelligent control method for a micro-interface oil removal device based on backwashing circulation, comprising:

[0007] Real-time operating status data and historical backwash data of the micro-interface oil removal device are obtained. Based on the operating status data, a pollution state evolution model considering the spatiotemporal coupling effect is constructed. The pollution state evolution model is used to predict the pollutant accumulation trend and diffusion law of each region of the micro-interface, and obtain the zonal pollution assessment results with time-series characteristics.

[0008] Based on the pollutant accumulation rate and diffusion characteristics in the zonal pollution assessment results, and combined with the system backwashing resource constraints, a multi-objective optimization function is constructed to generate initial pulse sequence schemes for zonals with different pollution levels.

[0009] Based on the spatial distribution characteristics and time-varying patterns of pollutants in the zonal pollution assessment results, the initial pulse sequence scheme is dynamically reconstructed. By adjusting the temporal relationship and range of action of each pulse unit, an optimized pulse sequence with spatial adaptability is formed.

[0010] During the backwashing operation of the optimized pulse sequence, a graded evaluation criterion for the backwashing effect is established. Based on the differences in pollution characteristics in different regions of the micro-interface, the actual effect of the backwashing process is evaluated quantitatively by region. Based on the results of the quantitative evaluation by region, the parameters of the pollution state evolution model are selectively adjusted for regions with different pollution levels to optimize and improve the backwashing strategy.

[0011] Based on the operational status data, a pollution state evolution model considering spatiotemporal coupling effects is constructed. This model predicts the pollutant accumulation trends and diffusion patterns in each region of the micro-interface, yielding temporal-series-based zonal pollution assessment results, including:

[0012] Multidimensional spectral decomposition is performed on the pressure difference characteristics, flow characteristics and pollutant accumulation characteristics in the operation status data. The decomposed feature information is then finely reconstructed according to the spatial structure spectrum of the micro-interface to construct a pollution state evolution model that reflects the pollution state of each region. By tracing the inter-regional pollutant migration relationship in the pollution state evolution model, a spatial correlation spectrum is established.

[0013] Based on the spatiotemporal structural characteristics of the pollution state evolution model, the pollution state data at different times are expanded in the time domain. By hierarchically stripping and progressively recombining the dominant features of the expanded sequence, the evolution pattern of pollutant accumulation rate is restored. The migration path and diffusion intensity information in the spatial correlation spectrum are deeply coupled with the evolution pattern to construct a state evolution function.

[0014] Based on the spatiotemporal coupling relationship in the state evolution function, the trajectory of pollutant concentration change in each region of the micro-interface is dynamically tracked and predicted. By partitioning and analyzing the spatiotemporal characteristics of the pollutant concentration change trajectory, an assessment result reflecting the pollution evolution law of each region is generated.

[0015] Based on the spatiotemporal structural characteristics of the pollution state evolution model, pollution state data at different times are expanded in the time domain. By hierarchically stripping and progressively recombining the dominant features of the expanded sequence, the evolution pattern of pollutant accumulation rate is restored, including:

[0016] Based on the spatiotemporal structure spectrum in the pollution state evolution model, the pollution state information at different times is separated into multiple scales in the time domain to construct a feature spectrum matrix that reflects the progressive evolution of pollutants in the time dimension. The dominant fluctuation mode and secondary fluctuation mode of the pollutant accumulation process are extracted from the feature spectrum matrix.

[0017] Feature enhancement is performed based on the main changing trends in the dominant fluctuation mode. By hierarchically deconstructing and sequentially reorganizing its temporal energy distribution, the baseline evolution path of pollutant accumulation is extracted, and the local perturbation features in the secondary fluctuation mode are transformed into modulation functions of the baseline evolution path.

[0018] In the time-frequency joint domain, the reference evolution path and the correction effect of the modulation function are deeply integrated. By introducing time-varying weight coefficients, the fusion process is adaptively adjusted to reconstruct the evolution mode of pollutant accumulation rate.

[0019] Based on the pollutant accumulation rate and diffusion characteristics in the zoning pollution assessment results, and combined with the system backwashing resource constraints, a multi-objective optimization function is constructed to generate initial pulse sequence schemes for zoning with different pollution levels, including:

[0020] The pollutant accumulation rate and pollutant diffusion characteristics of each zone are analyzed from the pollution assessment results of the zones. Based on the pollutant accumulation rate, the pollutant removal demand intensity of each zone is decomposed. Based on the pollutant diffusion characteristics, the pollutant migration influence coefficient between each zone and adjacent zones is restored.

[0021] Based on the intensity of pollutant removal demand and the pollutant migration influence coefficient, a multi-objective optimization function is established with the joint optimization objectives of maximizing oil removal efficiency and minimizing backwashing energy consumption.

[0022] Import backwashing resource constraints, transform the backwashing resource constraints into the constraint boundary of the multi-objective optimization function, solve the multi-objective optimization function, and obtain the pulse parameter combination that maximizes the oil removal efficiency and minimizes the backwashing energy consumption to achieve Pareto optimality under the backwashing resource constraints.

[0023] The pulse energy allocation scheme and pulse duration allocation scheme for different pollution levels in the pulse parameter combination are combined in a gradient to construct an initial pulse sequence scheme.

[0024] Based on the spatial distribution characteristics and time-varying patterns of pollutants in the zonal pollution assessment results, the initial pulse sequence scheme is dynamically reconstructed. By adjusting the temporal relationship and range of action of each pulse unit, an optimized pulse sequence with spatial adaptability is formed, including:

[0025] The pollution assessment results of the zoning are subjected to multidimensional profile separation, and the pollutant concentration distribution data and concentration change rate data of each zoning are purified. A spatial coupling matrix that characterizes the correlation of pollutant concentration between zoning ...

[0026] Based on the concentration change rate data, we deepen the temporal extension analysis, track the pollutant concentration evolution trajectory of each zone in the future time window, and combine the gradient difference between the pollutant concentration evolution trajectory and the current pollutant concentration to identify the rapid deterioration area of ​​pollutants where the concentration increment exceeds the preset increment threshold.

[0027] The rapidly deteriorating region of the pollutant is projected onto the spatial propagation spectrum, and the rapidly deteriorating region located at the propagation source is selected as the intervention region. Based on the hierarchical relationship of the intervention region in the spatial propagation spectrum, the priority execution order of each pulse unit in the initial pulse sequence scheme is reconstructed.

[0028] Around the pollutant diffusion gradient between adjacent partitions in the spatial propagation spectrum, a domain expansion function is constructed for each pulse unit. Based on the domain expansion function, the radiation range of the pulse unit acting on the intervention area is progressively extended to multiple consecutive partitions in the spatial propagation spectrum to form an optimized pulse sequence.

[0029] During the backwashing operation of the optimized pulse sequence, a graded evaluation criterion for the backwashing effect is established. Based on the differences in contamination characteristics in different regions of the micro-interface, the actual effect of the backwashing process is quantitatively evaluated by region, including:

[0030] During the backwashing operation of the optimized pulse sequence, the backwash flow fluctuation pattern of different partitions of the micro-interface is dynamically identified. Combined with the distribution of abrupt change points of the differential pressure response characteristic curve, a spatiotemporal evolution spectrum characterizing the intensity of the backwashing action is constructed. The spatiotemporal evolution spectrum is deeply integrated with the real-time collected pollutant concentration change trajectory to form a multidimensional evaluation matrix reflecting the transient effect of the backwashing process.

[0031] Based on the flow-pressure difference response characteristics and pollutant migration characteristics in the multidimensional evaluation matrix, feature decomposition is performed. By tracking the principal components of the multidimensional feature space and tracing the information source, the dominant change pattern of the backwashing effect is restored. The dominant change pattern is correlated and mapped with the pollutant accumulation rate spectrum of each partition of the micro-interface to construct a hierarchical evaluation system of backwashing effect that reflects different pollution levels of partitions.

[0032] Based on the dynamic response characteristics in the backwash effect hierarchical evaluation system, the backwash response characteristics of each zone are decoupled in multiple levels according to pollutant removal rate, backwash intensity utilization rate and energy conversion efficiency. By quantitatively evaluating the decoupled indicators at each level, a backwash effect evaluation result with zone differences is formed.

[0033] Based on the flow-pressure difference response characteristics and pollutant migration characteristics in the multidimensional evaluation matrix, feature decomposition is performed. By tracing the principal components and information sources in the multidimensional feature space, the dominant change patterns of the backwashing effect are reconstructed, including:

[0034] Dynamic fluctuation information of flow-pressure difference response characteristics and pollutant migration characteristics is extracted from the multidimensional evaluation matrix. The dynamic fluctuation information is hierarchically separated according to different time scales. The separated hierarchical features are recombined and mapped in the time-frequency coupling domain to construct a multidimensional feature space that reflects the transient response of the system.

[0035] In the multidimensional feature space, the source analysis of the coordinated change pattern of pollutant concentration and flow-pressure difference response characteristics is performed. By deeply analyzing the propagation path and diffusion intensity of the coordinated change pattern, the driving factors and hindering factors in the evolution of backwashing effect are identified. Based on the spatiotemporal distribution law of the driving factors and the hindering factors, the dominant change pattern of backwashing effect is reconstructed.

[0036] A second aspect of the present invention provides an intelligent control system for a micro-interface oil removal device based on backwashing circulation, comprising:

[0037] The acquisition module is used to acquire real-time operating status data and historical backwash data of the micro-interface oil removal device. Based on the operating status data, a pollution state evolution model considering the spatiotemporal coupling effect is constructed. The pollution state evolution model is used to predict the accumulation trend and diffusion law of pollutants in each region of the micro-interface, and obtain the zonal pollution assessment results with time-series characteristics.

[0038] The module is used to construct a multi-objective optimization function based on the pollutant accumulation rate and diffusion characteristics in the zoning pollution assessment results, combined with the system backwashing resource constraints, and generate initial pulse sequence schemes for zoning with different pollution levels.

[0039] The determination module is used to dynamically reconstruct the initial pulse sequence scheme based on the spatial distribution characteristics and time-varying patterns of pollutants in the zonal pollution assessment results, and to form an optimized pulse sequence with spatial adaptability by adjusting the temporal relationship and range of action of each pulse unit.

[0040] The optimization module is used to establish a graded evaluation criterion for the backwashing effect during the execution of the optimized pulse sequence backwashing operation. Based on the differences in pollution characteristics in different regions of the micro-interface, it performs a zonal quantitative evaluation of the actual effect of the backwashing process. Based on the results of the zonal quantitative evaluation, it selectively adjusts the parameters of the pollution state evolution model for different pollution level regions to achieve optimization and improvement of the backwashing strategy.

[0041] A third aspect of the present invention provides an electronic device, comprising:

[0042] processor;

[0043] Memory used to store processor-executable instructions;

[0044] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0045] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0046] The beneficial effects of this application are as follows:

[0047] This invention constructs a pollution state evolution model that considers spatiotemporal coupling effects, enabling accurate prediction of pollutant accumulation trends and diffusion patterns in various regions of a micro-interface. This achieves dynamic tracking and forward-looking assessment of pollution states. Compared to traditional pollution monitoring methods that rely on a single time or spatial dimension, this invention fully considers the temporal and spatial interactions of pollutants, resulting in a more comprehensive and accurate pollution assessment. This provides a reliable data foundation for subsequent backwashing control and effectively enhances the system's ability to perceive and predict pollution states.

[0048] This invention employs a multi-objective optimization function to generate an initial pulse sequence scheme and dynamically reconstructs it based on the spatial distribution characteristics and time-varying patterns of pollutants, forming an optimized pulse sequence with spatial adaptability. This zone-differentiated backwashing strategy can implement targeted treatment according to the actual pollution level of different areas, avoiding the resource waste and insufficient cleaning problems caused by traditional uniform backwashing methods. While ensuring oil removal efficiency, it significantly reduces energy consumption and backwashing media consumption, improving the economic and environmental benefits of system operation. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the intelligent control method for a micro-interface oil removal device based on backwashing circulation, according to an embodiment of the present invention.

[0050] Figure 2 This is a flowchart of the multidimensional evaluation matrix eigenvalue decomposition and change pattern reconstruction in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0053] Figure 1 This is a flowchart illustrating the intelligent control method for a micro-interface oil removal device based on backwashing circulation, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0054] Real-time operating status data and historical backwash data of the micro-interface oil removal device are obtained. Based on the operating status data, a pollution state evolution model considering the spatiotemporal coupling effect is constructed. The pollution state evolution model is used to predict the pollutant accumulation trend and diffusion law of each region of the micro-interface, and obtain the zonal pollution assessment results with time-series characteristics.

[0055] Based on the pollutant accumulation rate and diffusion characteristics in the zonal pollution assessment results, and combined with the system backwashing resource constraints, a multi-objective optimization function is constructed to generate initial pulse sequence schemes for zonals with different pollution levels.

[0056] Based on the spatial distribution characteristics and time-varying patterns of pollutants in the zonal pollution assessment results, the initial pulse sequence scheme is dynamically reconstructed. By adjusting the temporal relationship and range of action of each pulse unit, an optimized pulse sequence with spatial adaptability is formed.

[0057] During the backwashing operation of the optimized pulse sequence, a graded evaluation criterion for the backwashing effect is established. Based on the differences in pollution characteristics in different regions of the micro-interface, the actual effect of the backwashing process is evaluated quantitatively by region. Based on the results of the quantitative evaluation by region, the parameters of the pollution state evolution model are selectively adjusted for regions with different pollution levels to optimize and improve the backwashing strategy.

[0058] In one optional implementation, a pollution state evolution model considering spatiotemporal coupling effects is constructed based on the operational status data. This model is used to predict the pollutant accumulation trends and diffusion patterns in each region of the micro-interface, yielding a temporal-series-based zonal pollution assessment result, including:

[0059] Multidimensional spectral decomposition is performed on the pressure difference characteristics, flow characteristics and pollutant accumulation characteristics in the operation status data. The decomposed feature information is then finely reconstructed according to the spatial structure spectrum of the micro-interface to construct a pollution state evolution model that reflects the pollution state of each region. By tracing the inter-regional pollutant migration relationship in the pollution state evolution model, a spatial correlation spectrum is established.

[0060] Based on the spatiotemporal structural characteristics of the pollution state evolution model, the pollution state data at different times are expanded in the time domain. By hierarchically stripping and progressively recombining the dominant features of the expanded sequence, the evolution pattern of pollutant accumulation rate is restored. The migration path and diffusion intensity information in the spatial correlation spectrum are deeply coupled with the evolution pattern to construct a state evolution function.

[0061] Based on the spatiotemporal coupling relationship in the state evolution function, the trajectory of pollutant concentration change in each region of the micro-interface is dynamically tracked and predicted. By partitioning and analyzing the spatiotemporal characteristics of the pollutant concentration change trajectory, an assessment result reflecting the pollution evolution law of each region is generated.

[0062] The pollution state evolution model is constructed based on multidimensional spectral decomposition technology using operational status data. The data acquisition module continuously monitors the differential pressure, flow characteristics, and pollutant accumulation characteristics of the micro-interface system, with a sampling frequency of 10 times per second and a data accuracy requirement of three decimal places. Differential pressure feature extraction employs a sliding window mechanism with a window length of 60 sampling points and a step size of 10 sampling points. The time-domain signal is converted to a frequency-domain spectrum using a Fast Fourier Transform. Differential pressure spectral decomposition retains frequency components with a power spectral density greater than 1% of the total energy, and the amplitude and phase information of each component are stored separately. Flow characteristic decomposition uses wavelet transform, selecting the 4th-order Daubechies wavelet as the basis function, with a decomposition level of 5 levels. The detail coefficients and approximation coefficients extracted from each level are stored as feature vectors. The coefficients of each level after flow spectral decomposition undergo threshold denoising, with the threshold set to 0.3 times the standard deviation of the coefficients. The pollutant accumulation characteristics are obtained by differential operation to obtain the cumulative rate sequence. After normalization of the sequence, principal component analysis is performed to retain the top principal components with a cumulative contribution rate of more than 95%. The principal component loading coefficients constitute the pollutant characteristic spectrum.

[0063] The decomposed feature information is refined and reconstructed according to the micro-interface spatial structure hierarchy. The micro-interface is divided into 9 equal-area rectangular regions, each corresponding to an independent state node. The node numbers are arranged in order from left to right and from top to bottom. Feature information reconstruction uses an interpolation mapping method, allocating the decomposed frequency domain features to each spatial node through a cubic spline interpolation function. The calculation of the reconstruction weight matrix is ​​based on the physical distance and fluid dynamic similarity between regions. The distance weight is calculated using a Gaussian decay function with a decay parameter set to 0.2, while the dynamic weight is determined using a Reynolds number similarity metric with a similarity threshold set to 0.85. The pollution state evolution model is constructed using a state-space representation, with a 27-dimensional state vector containing the pressure difference state, flow rate state, and pollutant concentration state of each of the 9 regions. The state transition matrix is ​​determined based on the reconstruction weights and physical constraints, with matrix element values ​​limited to between -1 and +1.

[0064] Tracing the source of pollutant migration relationships between regions is achieved through cross-correlation analysis. The cross-correlation function of pollutant concentration time series for each pair of adjacent regions is calculated, with a lag window set to -20 to +20 sampling points and a correlation coefficient threshold of 0.6. Correlation relationships exceeding the threshold are identified as valid migration paths, and migration intensity is quantified by the absolute value of the correlation coefficient. The migration direction is determined by the lag sign corresponding to the maximum correlation coefficient; a positive lag indicates migration from the previous region to the next, and a negative lag indicates reverse migration. Spatial correlation spectra are constructed using graph theory, with nine regions as nodes and valid migration paths as directed edges, the edge weight equal to the migration intensity value. The correlation spectrum is stored in the form of an adjacency matrix, where matrix elements represent the migration intensity of corresponding region pairs; pairs of regions with no migration relationship have elements of zero. The correlation spectrum is updated once per minute, and historical correlation spectra are stored at a depth of 24 hours to support long-term trend analysis and pattern recognition.

[0065] The spatiotemporal structural feature analysis of the pollution state evolution model is based on temporal expansion technology. Temporal expansion organizes the pollution state data into a matrix form according to the time series, with rows representing time points and columns representing state variables. The matrix dimension is the number of time steps multiplied by the number of state variables. The expansion window length is set to 300 time points, and the overlap length is 100 time points to ensure data continuity and smooth transition. After expanding the pollution state data at different time points, a high-dimensional spatiotemporal data matrix is ​​formed, with each row representing a snapshot of the full-space state at a given time point. Singular value decomposition (SVD) is used to extract the dominant features of the expanded sequence, performing decomposition operations on the spatiotemporal data matrix to obtain the left singular matrix, singular value matrix, and right singular matrix. Modes with singular values ​​greater than 1% of the total energy are retained as dominant features; the number of modes is typically between 5 and 15.

[0066] The hierarchical stripping of the dominant features of the unfolded sequence sorts the modes according to the magnitude of their singular values, separating high-energy modes from low-energy modes in turn. A separation threshold is set at a ratio of adjacent singular values ​​less than 0.5; modes meeting this condition are grouped into the same level. Each level contains several modes, and the modes within a level are merged into a representative mode through a weighted average. Progressive recombination is achieved through a weighted linear combination, with weight coefficients determined based on the mode's contribution to the total variance, calculated to four decimal places. The recombined feature sequence maintains the original temporal resolution, but the dimensionality is reduced to the number of dominant modes.

[0067] The evolution pattern of pollutant accumulation rate was obtained through time derivative calculation. First-order differences were calculated on the recombined dominant feature time series, with the difference step size set to one sampling interval, resulting in an instantaneous rate of change sequence. This rate of change sequence was further smoothed using a moving average filter with a filter window length of 5 sampling points. Cluster analysis was employed for evolution pattern identification, using the k-means algorithm. The number of clusters was set to 5, the maximum number of iterations was 100, and the convergence criterion was that the displacement of the cluster centers was less than 0.01. Initial cluster centers were selected using the k-means++ method to ensure a dispersed distribution. Each cluster represents a typical accumulation rate evolution pattern, and the pattern characteristics are described by the coordinates of the cluster centers, with the center coordinate dimension equal to the number of dominant features.

[0068] In the spatial correlation spectrum, migration paths and diffusion intensity information are deeply coupled with evolutionary patterns. For each migration path, evolutionary pattern labels are extracted from the starting and ending regions, and the similarity between the evolutionary patterns of the two regions is calculated. Euclidean distance is used for similarity calculation, with a distance threshold set at 0.3. Paths with similarity exceeding the threshold are marked as validly coupled paths. Diffusion intensity information is calculated by multiplying the path weight by the similarity to obtain the coupling strength value. Paths with a coupling strength greater than 0.1 are included in the state evolution function construction. The coupling relationship is represented in the form of a coupling matrix, where the rows and columns correspond to the source and target regions, respectively, and the matrix elements are the coupling strength values. The coupling matrix is ​​updated hourly, and the historical coupling matrix is ​​used to analyze the temporal evolution of the coupling relationship.

[0069] The state evolution function is constructed based on spatiotemporal coupling. It is expressed as a system of differential equations, with the number of equations equal to the number of regions, each region corresponding to one state differential equation. The right-hand side of the equations includes internal evolution terms for the current region and coupling terms from other regions. The coefficients of the internal evolution terms are determined by the evolutionary pattern eigenvectors, with values ​​ranging from -0.5 to +0.5. The coefficients of the coupling terms are equal to the coupling strength of the corresponding path, and the number of coupling terms equals the number of effective migration paths received by the region. The equations are solved using the fourth-order Runge-Kutta method, with an integration step size of 0.01 time units and a solution precision controlled to 6 significant digits. The evolution function parameters are updated hourly, and parameter smoothing employs an exponentially weighted moving average method with a smoothing factor of 0.9 to avoid numerical instability caused by drastic parameter changes.

[0070] The dynamic tracking of pollutant concentration changes in various regions of the micro-interface is based on forward integration of an evolutionary function. The prediction time window is set to the next 60 minutes, with a prediction step size of 1 minute. At each prediction time, predicted pollutant concentration values ​​for 9 regions are output. The tracking algorithm employs Kalman filtering, with the filter state vector containing concentration values ​​and rates of change, and an 18-dimensional state dimension. The process noise covariance matrix has diagonal elements set to 0.05 and off-diagonal elements set to 0.01, while the observation noise covariance is set to 0.02. The state transition matrix is ​​obtained by linearizing the evolutionary function, and the observation matrix consists of the first 9 rows of the identity matrix, corresponding to the concentration observations in the 9 regions. The filter prediction step uses the evolutionary function for state extrapolation, and the update step uses real-time observation data for state correction.

[0071] The uncertainty of the pollutant concentration change trajectory prediction results was assessed using confidence intervals. The prediction confidence level was set at 95%, and the confidence interval was estimated using the Monte Carlo method with 1000 sampling iterations. For each sampling, a normally distributed random perturbation was added to the evolution function parameters, with the perturbation standard deviation set to 10% of the parameter standard deviation. Quantile statistics of the predicted trajectory set yielded the upper and lower bounds of the confidence interval, and the width of the confidence interval served as a measure of prediction uncertainty.

[0072] The spatiotemporal characteristics of pollutant concentration change trajectories were analyzed through statistical analysis, calculating the mean, variance, maximum, and minimum pollutant concentrations for each region. The statistical window length was the full length of the prediction time window. Regional assessment indicators included pollution accumulation rate, concentration change amplitude, peak occurrence time, and duration. The accumulation rate was calculated as the ratio of concentration increment to time increment, expressed in milligrams per liter per minute (mg / L / min). The change amplitude was the difference between the maximum and minimum values, reflecting the degree of concentration fluctuation. The peak occurrence time was the moment the concentration reached its maximum value, expressed in minutes relative to the current time. The duration was the total time the concentration exceeded the average plus one standard deviation, expressed in minutes.

[0073] The assessment results reflecting the pollution evolution patterns in each region are generated in the form of numerical tables. Each table contains 36 data items, including four indicators for each of the nine regions. Numerical precision is maintained to two decimal places, and scientific notation is used to represent values ​​below 0.001 or above 1000. The assessment results are updated synchronously with the prediction frequency, with a new assessment table generated every minute. Historical assessment results are saved for seven days for long-term trend analysis and model performance evaluation. The assessment tables also include time-series charts for each indicator, with a resolution of 100 pixels per region and color coding using a gradient spectrum to represent numerical values.

[0074] In one optional implementation, based on the spatiotemporal structural characteristics of the pollution state evolution model, pollution state data at different times are expanded in the time domain. By hierarchically stripping and progressively recombining the dominant features of the expanded sequence, the evolution pattern of pollutant accumulation rate is restored, including:

[0075] Based on the spatiotemporal structure spectrum in the pollution state evolution model, the pollution state information at different times is separated into multiple scales in the time domain to construct a feature spectrum matrix that reflects the progressive evolution of pollutants in the time dimension. The dominant fluctuation mode and secondary fluctuation mode of the pollutant accumulation process are extracted from the feature spectrum matrix.

[0076] Feature enhancement is performed based on the main changing trends in the dominant fluctuation mode. By hierarchically deconstructing and sequentially reorganizing its temporal energy distribution, the baseline evolution path of pollutant accumulation is extracted, and the local perturbation features in the secondary fluctuation mode are transformed into modulation functions of the baseline evolution path.

[0077] In the time-frequency joint domain, the reference evolution path and the correction effect of the modulation function are deeply integrated. By introducing time-varying weight coefficients, the fusion process is adaptively adjusted to reconstruct the evolution mode of pollutant accumulation rate.

[0078] The reconstruction of the pollutant accumulation rate evolution pattern is achieved through multi-scale temporal domain separation technology based on the spatiotemporal structural features of the pollution state evolution model. The spatiotemporal structure spectrum is extracted by eigenvalue decomposition of the state matrix of the evolution model. The dimension of the state matrix is ​​set to the product of the number of regions and the number of state variables, and the eigenvalues ​​are sorted from largest to smallest to form the structural spectrum. The pollution state information collection frequency at different times is set to 5 times per second, with a data cache depth of 3600 time points, corresponding to a 12-minute historical data window. Multi-scale temporal domain separation employs wavelet packet decomposition, selecting the 5th-order Daubechies wavelet as the mother wavelet function. The decomposition level is set to 4 levels, with each level generating 2 sub-bands. Each sub-band corresponds to different time-scale characteristics, ranging from high-frequency local fluctuations to low-frequency long-term trends.

[0079] The feature spectrum matrix is ​​constructed by reorganizing the separation results at each time scale according to the progressive evolution of the time dimension. The number of rows in the matrix equals the number of time points, and the number of columns equals the number of sub-bands multiplied by the number of regions. The matrix elements are the coefficient values ​​of the corresponding time point, sub-band, and region. The feature spectrum matrix is ​​stored in a sparse matrix format, with a compression ratio typically exceeding 70%. Matrix updates employ a sliding window mechanism: when new data arrives, the oldest row is deleted and a new row is added to the end, maintaining a constant matrix dimension. Normalization of the feature spectrum matrix is ​​performed column-wise, with each column's coefficient divided by the maximum absolute value of that column. The normalized coefficients are limited to the range of -1 to +1.

[0080] The dominant wave patterns in the pollutant accumulation process are extracted from the eigenspectral matrix through singular value decomposition (SVD). SVD is performed on the normalized eigenspectral matrix to obtain a left singular matrix, singular value vectors, and a right singular matrix. The magnitude of the singular values ​​reflects the importance of the corresponding patterns. The top few patterns, whose singular values ​​account for more than 95% of the total energy, are retained as dominant wave patterns; the number of patterns is typically between 3 and 8. The temporal components of the dominant wave patterns are given by the corresponding columns of the left singular matrix, and the spatial components are given by the corresponding rows of the right singular matrix. Secondary wave patterns consist of the remaining smaller singular value patterns; these patterns mainly reflect local disturbances and noise components.

[0081] Feature enhancement of the main trend components in the dominant wave pattern is achieved through energy redistribution. The energy distribution of each dominant pattern across different frequency bands is calculated using a short-time Fourier transform (SFT) method, with a window length of 64 sampling points and an overlap rate of 50%. An energy distribution threshold of three times the average energy is set, and frequency bands exceeding this threshold are identified as main trend components. Feature enhancement is achieved by amplifying the amplitude of the main trend components, with an amplification factor set to an adaptive value between 2 and 5, dynamically adjusted based on the signal-to-noise ratio (SNR). The amplified components are then recombined to obtain the enhanced dominant wave pattern, and the enhancement effect is evaluated by comparing the peak SNR before and after enhancement.

[0082] The hierarchical deconstruction of the temporal energy distribution layers the enhanced dominant wave pattern according to the degree of energy concentration. Energy calculation employs a sliding window method with a window length of 32 sampling points, and energy is defined as the sum of squared signal amplitudes within the window. Peak detection of the energy distribution curve uses a local maximum algorithm, with a peak spacing threshold of 16 sampling points. Based on the peak energy magnitude, the distribution curve is divided into high-energy, medium-energy, and low-energy layers, with stratification thresholds set at 80% and 40% of the maximum peak energy, respectively. Hierarchical recombination uses a weighted average method to recombine the energy distribution of each layer. The weighting coefficients are determined based on the energy proportion of each layer: 0.6 for the high-energy layer, 0.3 for the medium-energy layer, and 0.1 for the low-energy layer.

[0083] The baseline evolution path for pollutant accumulation is extracted from the reconstituted energy distribution. The baseline path is defined as the envelope of the reconstituted energy distribution, and the instantaneous amplitude is obtained using the Hilbert transform method. A moving average filter is used for envelope smoothing, with a filter length of 8 sampling points. The filtered envelope serves as the baseline evolution path. The sampling interval of the baseline path remains consistent with the original data, and the path numerical range is limited to 0 to 1 using min-max normalization. The baseline path is stored as a one-dimensional array, with the array length equal to the number of sampling points within the time window.

[0084] In the secondary wave pattern, the transformation of local perturbation features is achieved using modulation and demodulation techniques. These features are separated from the secondary wave pattern using a high-pass filter with a cutoff frequency set to 10% of the sampling frequency. The filter is an 8th-order Butterworth filter. The modulation function is constructed by modulating the local perturbation features with the instantaneous phase of the reference evolution path. The phase information is obtained through Hilbert transform. The amplitude of the modulation function is determined by the envelope of the local perturbation features, and the phase is determined by the instantaneous phase of the reference path. Normalization of the modulation function ensures its amplitude ranges from -0.5 to +0.5, avoiding excessive perturbation to the reference path.

[0085] The deep fusion of the reference evolution path and modulation function in the time-frequency joint domain employs a weighted synthesis method in the short-time Fourier transform domain. Both the reference path and the modulation function undergo short-time Fourier transforms, using a Hanning window with a length of 128 sampling points and an overlap rate of 75%. The time-frequency domain representation is a complex matrix, with rows corresponding to frequencies and columns corresponding to times. The fusion process is performed at each time-frequency point, and the fusion weights are determined based on the energy ratio of the reference path and the modulation function at that point. The energy ratio is calculated by dividing the energy of the reference path by the sum of the energies of the two functions, and the fusion coefficient is the square root of the energy ratio.

[0086] The adaptive adjustment of time-varying weight coefficients in the fusion process is based on the rate of change of pollution state. This rate of change is calculated using the absolute value of the difference in pollution concentration between adjacent time points. When the rate of change exceeds a preset threshold, the fusion weight of the modulation function is increased. The threshold is set to twice the standard deviation of the historical rate of change, and the weight adjustment range is limited to 0.1 to 0.9. The update frequency of the time-varying weight coefficients is synchronized with the data acquisition frequency. Weight changes are smoothed using a first-order inertial loop, and the time constant is set to 10 sampling intervals. The weight coefficients are stored as a time series array for subsequent pattern reconstruction verification and parameter optimization.

[0087] The reconstruction of the pollutant accumulation rate evolution model is achieved through inverse time-frequency joint domain transform. The fused time-frequency domain complex matrix is ​​converted back to a time-domain signal using inverse short-time Fourier transform. The inverse transform employs an overlapping addition method to ensure signal continuity. The length of the reconstructed signal remains consistent with the original time window, and the signal amplitude range is limited to a reasonable range through dynamic gain control. The quality assessment of the evolution model uses the correlation coefficient between the reconstructed signal and the original pollution accumulation rate; a correlation coefficient greater than 0.85 is considered a successful reconstruction.

[0088] In one optional implementation, based on the pollutant accumulation rate and diffusion characteristics in the zonal pollution assessment results, and combined with the system backwashing resource constraints, a multi-objective optimization function is constructed to generate initial pulse sequence schemes for zonals with different pollution levels, including:

[0089] The pollutant accumulation rate and pollutant diffusion characteristics of each zone are analyzed from the pollution assessment results of the zones. Based on the pollutant accumulation rate, the pollutant removal demand intensity of each zone is decomposed. Based on the pollutant diffusion characteristics, the pollutant migration influence coefficient between each zone and adjacent zones is restored.

[0090] Based on the intensity of pollutant removal demand and the pollutant migration influence coefficient, a multi-objective optimization function is established with the joint optimization objectives of maximizing oil removal efficiency and minimizing backwashing energy consumption.

[0091] Import backwashing resource constraints, transform the backwashing resource constraints into the constraint boundary of the multi-objective optimization function, solve the multi-objective optimization function, and obtain the pulse parameter combination that maximizes the oil removal efficiency and minimizes the backwashing energy consumption to achieve Pareto optimality under the backwashing resource constraints.

[0092] The pulse energy allocation scheme and pulse duration allocation scheme for different pollution levels in the pulse parameter combination are combined in a gradient to construct an initial pulse sequence scheme.

[0093] The initial pulse sequence scheme was generated based on an in-depth analysis of pollutant accumulation rates and diffusion characteristics from the zonal pollution assessment results. The zonal pollution assessment results were provided in tabular form, including key parameters such as pollutant concentration, accumulation rate, diffusion coefficient, and boundary flux for each of the nine membrane zones. The pollutant accumulation rate was calculated as the ratio of the concentration difference at consecutive time points to the time interval, with a calculation accuracy of four decimal places, in milligrams per liter per second. The accumulation rate data window length was set to 300 seconds, with a sliding step size of 30 seconds to ensure data smoothness and real-time performance. The pollutant diffusion characteristics were obtained through numerical solutions to Fick's diffusion law. The diffusion coefficient was calculated as the ratio of the concentration gradient to the flux density, and the diffusion directionality was represented by the direction angle of the gradient vector.

[0094] The decomposition of pollutant removal demand intensity for each zone is based on a quantitative grading process using the cumulative rate. The cumulative rate is divided into five levels according to its numerical value, with grading thresholds set at 0.05, 0.15, 0.35, 0.65, and 1.0 mg / L / s, respectively. Level 1 corresponds to slight pollution, with a removal demand intensity coefficient set at 0.2; Level 2 corresponds to mild pollution, with an intensity coefficient of 0.4; Level 3 corresponds to moderate pollution, with an intensity coefficient of 0.6; Level 4 corresponds to severe pollution, with an intensity coefficient of 0.8; and Level 5 corresponds to severe pollution, with an intensity coefficient of 1.0. The time decay of removal demand intensity adopts an exponential decay model, with a decay time constant set at 180 seconds, and the decay factor calculated using the natural logarithm function. The intensity coefficient is stored as a 9-dimensional vector, with each element corresponding to the removal demand intensity of a zone. The vector update frequency is synchronized with the pollution assessment frequency.

[0095] The reconstruction of the pollutant migration impact coefficient is based on a coupled analysis of diffusion characteristics and the geometric relationship between adjacent zones. Adjacent zones are defined as pairs of zones sharing a common boundary. The adjacency relationship formed by the nine zones is represented by an adjacency matrix with dimensions 9 x 9. Elements corresponding to adjacent relationships are set to 1, and non-adjacent relationships are set to 0. The migration impact coefficient is calculated by multiplying the diffusion coefficient by the boundary length and dividing by the zone area. The calculation results are normalized to ensure the coefficient ranges between 0 and 1. The migration directionality is determined by the sign of the concentration gradient: a positive gradient indicates outward migration, and a negative gradient indicates inward migration. The impact coefficient matrix is ​​a 9 x 9 symmetric matrix; diagonal elements represent the influence within a zone, and off-diagonal elements represent the mutual influence between zones.

[0096] The multi-objective optimization function is established by using a weighted sum method to transform the maximization of oil removal efficiency and the minimization of backwash energy consumption into a single-objective function. Oil removal efficiency is defined as the ratio of pollutant removal amount to unit time, calculated by multiplying the concentration difference before and after backwashing by the partition volume. Backwash energy consumption includes three parts: pulse generator power consumption, water pump power consumption, and control power consumption. Power consumption is calculated based on the product of equipment power and operating time. The weighting coefficient for oil removal efficiency is set to 0.7, and the weighting coefficient for energy consumption is set to 0.3. These weighting coefficients can be adjusted according to actual operating requirements, ranging from 0.1 to 0.9. The optimization function adopts a linear combination form, maximizing the positive value of the oil removal efficiency term and minimizing the negative value of the energy consumption term.

[0097] The backwash resource constraints include four categories: total power consumption limit, water resource limit, pulse frequency limit, and equipment load limit. The total power consumption limit is set at no more than 15 kWh per hour, and the water resource limit is no more than 200 liters of water used per backwash. The pulse frequency limit is no more than 20 pulses per hour for each zone, and the equipment load limit is no more than 5 kW of power per pulse. The constraints are transformed into inequality constraints of the optimization function, which is processed using a penalty function method with a penalty factor set to 100. Solutions exceeding the constraints are subject to a significant penalty. Constraint boundary checks employ a real-time monitoring mechanism; when constraints are violated, warnings are triggered and parameters are automatically adjusted to within the feasible region.

[0098] The multi-objective optimization function is solved using a genetic algorithm to search for the Pareto optimal solution set. The population size is set to 100 individuals, each containing impulse parameters for 9 partitions, encoded using real numbers. The crossover probability is set to 0.8, the mutation probability to 0.1, and tournament selection is used with a tournament size of 5. The genetic algorithm iterates for 200 generations, and the convergence criterion is that the change in the optimal solution is less than 0.001 over 10 consecutive generations. The Pareto front is identified using a non-dominated sorting algorithm, and the number of front solutions is typically between 15 and 25. The solution set quality is evaluated using a hypervolume index, with the worst combination of objective function values ​​as the reference point.

[0099] The pulse parameter combination includes four dimensions: pulse energy, pulse duration, pulse interval, and pulse waveform. The pulse energy range is set to 100 to 2000 joules, with an energy distribution accuracy of 10 joules. The pulse duration ranges from 0.1 to 5.0 seconds, with a duration accuracy of 0.1 seconds. The pulse interval ranges from 10 to 300 seconds, with an interval accuracy of 1 second. Three pulse waveform types are used: square wave, triangle wave, and sine wave, with waveform parameters represented by coded values ​​0, 1, and 2. The parameter combination is validated through simulation to verify its feasibility and effectiveness under actual working conditions, with simulation accuracy required to reach at least 95% of the actual effect.

[0100] The pulse energy allocation scheme is based on a linear allocation strategy that removes the required intensity from each zone. The total available energy is allocated according to the proportion of the intensity coefficient of each zone. The allocation formula is: the energy allocated to a zone equals the total energy multiplied by the intensity coefficient of that zone, divided by the sum of the intensity coefficients of all zones. The energy allocation result is rounded to the nearest accurate energy value, with the allocation error controlled within 2%. The energy allocation matrix is ​​a 9-dimensional column vector, and the sum of the vector elements equals the total available energy. The allocation result is updated in real time and stored in the parameter database.

[0101] The pulse duration allocation scheme employs a proportional allocation method similar to that used for energy allocation. The baseline duration is set as the pollutant diffusion time constant, calculated as the ratio of the diffusion coefficient to the characteristic length. Duration allocation for each zone is determined based on the ratio of the zone's pollution level to the average pollution level; zones with higher pollution levels are allocated longer durations. The duration allocation range is limited to 50% to 200% of the baseline duration, with an allocation accuracy of 0.1 seconds. The duration allocation vector has the same dimension as the energy allocation vector, and the two vectors together constitute the complete pulse parameter allocation scheme.

[0102] Gradient combination processing rearranges and combines the energy allocation scheme and the duration allocation scheme in an increasing gradient manner. Gradient calculation is based on the spatial distribution of contamination levels in each partition, with higher contamination areas corresponding to larger gradient values. The combination strategy adopts a high-gradient-first, then-low-gradient order to ensure that severely contaminated areas receive pulse cleaning first. The gradient threshold is set to 1.5 times the average gradient; areas exceeding the threshold are classified as high-gradient groups, and the rest as low-gradient groups. The combination result forms a time-series pulse sequence, the length of which is equal to the number of partitions participating in the cleaning.

[0103] The initial pulse sequence construction uses a time-axis expansion method to convert the gradient combination results into an executable control sequence. The time axis precision is set to 0.1 seconds, and the total sequence length is calculated and determined based on the maximum pulse duration and interval time. Each time step corresponds to a control state vector with a dimension of 9, and the vector elements represent the pulse intensity of that partition at that time. The pulse sequence uses a sparse storage method, storing only the time and amplitude information of non-zero control signals. The sequence format is a triple array containing three fields: time, partition number, and pulse intensity.

[0104] In one optional implementation, based on the spatial distribution characteristics and time-varying patterns of pollutants in the zonal pollution assessment results, the initial pulse sequence scheme is dynamically reconstructed. By adjusting the temporal relationship and range of action of each pulse unit, an optimized pulse sequence with spatial adaptability is formed, including:

[0105] The pollution assessment results of the zoning are subjected to multidimensional profile separation, and the pollutant concentration distribution data and concentration change rate data of each zoning are purified. A spatial coupling matrix that characterizes the correlation of pollutant concentration between zoning ...

[0106] Based on the concentration change rate data, we deepen the temporal extension analysis, track the pollutant concentration evolution trajectory of each zone in the future time window, and combine the gradient difference between the pollutant concentration evolution trajectory and the current pollutant concentration to identify the rapid deterioration area of ​​pollutants where the concentration increment exceeds the preset increment threshold.

[0107] The rapidly deteriorating region of the pollutant is projected onto the spatial propagation spectrum, and the rapidly deteriorating region located at the propagation source is selected as the intervention region. Based on the hierarchical relationship of the intervention region in the spatial propagation spectrum, the priority execution order of each pulse unit in the initial pulse sequence scheme is reconstructed.

[0108] Around the pollutant diffusion gradient between adjacent partitions in the spatial propagation spectrum, a domain expansion function is constructed for each pulse unit. Based on the domain expansion function, the radiation range of the pulse unit acting on the intervention area is progressively extended to multiple consecutive partitions in the spatial propagation spectrum to form an optimized pulse sequence.

[0109] The optimized pulse sequence was generated using a multidimensional profile separation technique based on the regional pollution assessment results. These results included pollutant concentration data, spatial coordinates, and timestamps for nine membrane zones. Data was acquired at a frequency of 5 times per second, with storage precision to three decimal places. The multidimensional profile separation employed principal component analysis (PCA) to decompose the original 3D data space into several orthogonal profiles, each corresponding to a principal component direction. The separation process treated pollutant concentration distribution data and concentration change rate data as two independent dimensions. The concentration distribution data was extended from discrete sampling points to a continuous distribution field using spatial interpolation techniques. Kriging interpolation was used, with an interpolation grid resolution of 0.1 m x 0.1 m. The concentration change rate data was calculated by differencing concentration values ​​between adjacent time points, with a difference step size of 0.2 seconds. The change rate was limited to a range of -10 to +10 mg / L / s.

[0110] The spatial coupling matrix for pollutant concentration correlation is constructed based on inter-regional concentration correlation analysis. The coupling matrix has a dimension of 9x9, and the matrix elements are calculated using the Pearson correlation coefficient. The correlation coefficient calculation window is set to 300 sampling points, corresponding to 60 seconds of historical data. Diagonal elements are set to 1, indicating that the region is perfectly correlated with itself. Off-diagonal elements range from -1 to +1, and elements with an absolute value greater than 0.3 are considered to have a valid coupling relationship. The construction of the spatial coupling matrix considers the geometric positional relationship of the regions. The coupling strength of adjacent regions is weighted and adjusted by the length of their common boundary, with a boundary length weighting coefficient set to 0.2. The matrix is ​​updated once per minute, and the matrix is ​​stored using a compressed sparse row format, with a compression ratio typically exceeding 60%.

[0111] The eigenvalue decomposition of the spatial coupling matrix is ​​implemented using the Jacobi iterative algorithm, with an iterative convergence precision set to 10⁻⁶. Eigenvalues ​​are sorted from largest to smallest, and the top few eigenvectors with eigenvalues ​​greater than 5% of the total energy are retained as dominant distribution directions. Eigenvector normalization ensures a vector magnitude of 1, and the signs of vector elements are uniformly adjusted to positive values ​​for the dominant elements. The eigenvectors representing the dominant distribution directions of pollutants are interpreted physically through spatial gradient analysis; changes in the signs of the eigenvector elements reflect the inflow and outflow relationships of pollutants. A dominant distribution direction typically contains 3 to 5 eigenvectors, each corresponding to a typical pollutant distribution pattern. The eigenvectors are stored as a 9-dimensional column vector array, with array indices sorted according to the magnitude of the eigenvalues.

[0112] The spatial propagation spectrum of pollutants is reconstructed based on the eigenvectors of the dominant distribution direction. The reconstruction process rearranges the eigenvectors according to spatial adjacency relationships, forming a spectrum structure reflecting the pollutant propagation path. Spectrum nodes correspond to membrane partitions, and the node weight is equal to the absolute value of the coefficient of that partition in the dominant eigenvector. Spectrum edges represent the propagation relationship between adjacent partitions, and the edge weight is calculated by multiplying the coefficients of the eigenvectors of adjacent partitions. The directionality of the propagation relationship is determined by the difference in the signs of the eigenvector coefficients; the same sign indicates propagation in the same direction, and opposite signs indicate convective propagation. The propagation spectrum is represented by a directed graph, with the adjacency matrix storing the propagation relationships. The matrix density is typically around 30%. The hierarchical structure of the spectrum is determined using a depth-first search algorithm, with the node with the largest weight selected as the root node for propagation.

[0113] The temporal extrapolation analysis of concentration change rate data was performed using an autoregressive moving average model. The model order was selected using the Akaike Information Criterion, and the optimal parameters were determined by comparing the fitting effects of models of different orders. The model parameters were trained using maximum likelihood estimation, with a training data window length of 1200 sampling points, corresponding to 4 minutes of historical data. The extrapolation prediction time window was set to the next 120 seconds, with a prediction step size of 1 second, and a prediction accuracy requirement of over 90% of the actual value. The evolution trajectory of pollutant concentrations in each zone was obtained through forward modeling, considering the influence of random disturbances, with the disturbance variance set to 80% of the historical residual variance. The trajectory data was stored as a time series array, with the array length equal to the prediction time window length. The confidence interval of the evolution trajectory was estimated using Monte Carlo simulation, with 1000 simulations and a confidence level of 95%.

[0114] The gradient difference between the pollutant concentration evolution trajectory and the current concentration is calculated using a numerical differential method. The current gradient is calculated based on the spatial derivatives of the concentration values ​​of adjacent zones, and the derivatives are approximately obtained using the central difference formula, maintaining a calculation accuracy to four decimal places. The predicted gradient is calculated by dividing the difference between the concentration at the end of the evolution trajectory and the current concentration by the prediction time window length. The gradient difference is measured using Euclidean distance, with each component of the difference vector corresponding to gradient changes in different spatial directions. The concentration increment is calculated based on the integral of the evolution trajectory, and the integral is numerically calculated using Simpson's rule. The preset increment threshold is set to 2.5 times the standard deviation of historical increments, and the threshold adaptive adjustment mechanism dynamically adjusts the threshold range between 1.5 and 4.0 times the standard deviation according to the pollution deterioration rate. Zones where the increment exceeds the threshold are marked as rapidly deteriorating pollutant areas, and the timestamps are recorded for subsequent tracking and analysis. The identification of rapidly deteriorating areas uses a sliding window mechanism with a window length of 30 seconds, an overlap of 50%, and the detection frequency is synchronized with the data acquisition frequency.

[0115] The projection of rapidly deteriorating pollutant areas onto the spatial propagation spectrum is achieved through a node matching algorithm. A one-to-one match is made between the partition number of the rapidly deteriorating area and the node number in the propagation spectrum; successfully matched nodes are marked as potential intervention points. The projection process establishes a mapping relationship between the regional pollution state and the propagation spectrum structure, with the mapping weight determined based on the region's importance within the spectrum. Propagation source identification is based on the in-degree and out-degree analysis of nodes in the propagation spectrum. In-degree represents the number of paths receiving pollutants, and out-degree represents the number of paths propagating outwards. Nodes with an in-degree less than 2 and an out-degree greater than or equal to 3 are identified as propagation sources; these nodes are typically located at the starting point of pollution diffusion. Intervention area selection prioritizes rapidly deteriorating areas located at propagation sources. When the number of source areas exceeds three, the top three areas are selected as primary intervention areas, sorted from highest to lowest pollution deterioration rate. The selection results are stored as an intervention area list, with each element containing three fields: area number, deterioration rate, and intervention priority.

[0116] The initial pulse sequence scheme reconstructs the priority order of pulse units based on the hierarchical relationship of the spatial propagation spectrum. This hierarchy is determined using a graph topological sorting algorithm, and the sorting result reflects the temporal relationship and causal chain of pollutant propagation. Pulse units corresponding to the intervention area are assigned the highest priority, with a priority value of 1. The priorities of other pulse units are determined based on their shortest path distance to the intervention area within the propagation spectrum. Distance calculation uses Dijkstra's algorithm, where the path weight is equal to the reciprocal of the edge weight. Pulse units with smaller distances have higher priorities, and the priority adjustment range is limited to 1 to 9. Pulse units with the same priority maintain their relative positions according to the original sequence order to avoid unnecessary adjustment overhead. The reconstructed execution order is obtained through priority sorting, and the sorting result is stored as a pulse unit index array.

[0117] A physical diffusion model was used to construct the domain expansion function of the pollutant diffusion gradient between adjacent segments in the spatial propagation spectrum. The diffusion gradient was calculated using Fick's first law, where the gradient value equals the diffusion coefficient multiplied by the concentration gradient. The diffusion coefficient was determined based on the membrane material properties and operating conditions, typically ranging from 0.05 to 0.25 m² / s. The domain expansion function adopted a Gaussian function form, with the function center located at the geometric center of the intervention region. The function width parameter was determined based on the variance of the local diffusion gradient. Boundaries with gradients greater than the average gradient corresponded to larger expansion coefficients, calculated by dividing the gradient value by the maximum gradient value, ranging from 0.1 to 1.0. The anisotropy of the expansion function was achieved using an elliptic Gaussian function, where the ratio of the major and minor axes of the ellipse reflected the differences in diffusion intensity in different directions. Normalization of the expansion function ensured that the sum of the expansion coefficients of all segments equaled 1, and the normalized coefficients served as the spatial distribution weights for the pulse energy.

[0118] The progressive penetration of the pulse unit radiation range within the intervention area is implemented using a breadth-first search algorithm. The search starts from a node in the intervention area and expands layer by layer to adjacent nodes according to the edge connections of the propagation spectrum. The penetration intensity of each layer decreases according to the distance attenuation law, with an attenuation factor set to 0.7, which physically corresponds to the loss of pulse energy during propagation. The maximum number of expansion layers is limited to 3, corresponding to the effective influence range of the pulse effect. Within the penetration range, partitions are assigned corresponding pulse effect intensities based on their graph distance from the intervention area, using an exponential attenuation mode with an attenuation exponent of 1.5. During penetration, the directionality of the propagation spectrum is considered; the penetration intensity is maintained along the forward propagation path, while the penetration intensity is halved along the reverse propagation path. When the penetration ranges of multiple intervention areas overlap, the effect intensity of the overlapping areas is calculated using the superposition principle, with the total intensity not exceeding the maximum intensity of a single pulse.

[0119] The domain expansion function progressively extends the radiation range of the pulsed unit acting on the intervention area to multiple consecutive partitions along the spatial propagation spectrum. The infiltration process establishes an energy transfer matrix from the intervention area to the affected area, with matrix elements representing the energy transfer coefficient from the source partition to the target partition. The transfer coefficient comprehensively considers three factors: diffusion gradient, geometric distance, and propagation path impedance, and the final coefficient value is calculated using a weighted average method. The determination of multiple consecutive partitions is based on connectivity analysis of the propagation spectrum, selecting partitions with direct or indirect connections to the intervention area as infiltration targets. The infiltration depth is dynamically adjusted according to the severity of pollution and available pulse energy; severely polluted areas correspond to greater infiltration depths, and infiltration over longer distances is permitted when available energy is sufficient.

[0120] The optimized pulse sequence formation is achieved through a combined process of temporal rearrangement and spatial expansion. Temporal rearrangement, based on the reconstructed priority execution order, rearranges pulse units according to their priority from high to low, and further sorts those with the same priority according to their spatial adjacency. Spatial expansion expands pulse units that originally acted on a single partition into composite pulses acting on multiple partitions. The composite pulses consist of two levels: a main pulse and auxiliary pulses. The main pulse acts directly on the intervention area, while the auxiliary pulses act on the surrounding influence areas through a penetration mechanism. The total energy of the composite pulses is equal to the energy of the original pulses. The energy distribution among multiple partitions is determined based on the scope expansion function, with the main pulse allocating 60% to 80% of the energy and the auxiliary pulses allocating the remaining energy. The total execution time of the optimized sequence is compressed through pulse overlap optimization, allowing pulses in non-adjacent partitions to execute in parallel, with the overlap controlled within 30% to avoid mutual interference. Pulse timing coordination employs an event-driven scheduling mechanism, dynamically adjusting the execution time based on the readiness status and resource occupancy of each pulse unit.

[0121] In one optional implementation, during the backwashing operation of the optimized pulse sequence, a tiered evaluation criterion for the backwashing effect is established. Based on the differences in contamination characteristics in different regions of the micro-interface, a zonal quantitative evaluation of the actual effect of the backwashing process is performed, including:

[0122] During the backwashing operation of the optimized pulse sequence, the backwash flow fluctuation pattern of different partitions of the micro-interface is dynamically identified. Combined with the distribution of abrupt change points of the differential pressure response characteristic curve, a spatiotemporal evolution spectrum characterizing the intensity of the backwashing action is constructed. The spatiotemporal evolution spectrum is deeply integrated with the real-time collected pollutant concentration change trajectory to form a multidimensional evaluation matrix reflecting the transient effect of the backwashing process.

[0123] Based on the flow-pressure difference response characteristics and pollutant migration characteristics in the multidimensional evaluation matrix, feature decomposition is performed. By tracking the principal components of the multidimensional feature space and tracing the information source, the dominant change pattern of the backwashing effect is restored. The dominant change pattern is correlated and mapped with the pollutant accumulation rate spectrum of each partition of the micro-interface to construct a hierarchical evaluation system of backwashing effect that reflects different pollution levels of partitions.

[0124] Based on the dynamic response characteristics in the backwash effect hierarchical evaluation system, the backwash response characteristics of each zone are decoupled in multiple levels according to pollutant removal rate, backwash intensity utilization rate and energy conversion efficiency. By quantitatively evaluating the decoupled indicators at each level, a backwash effect evaluation result with zone differences is formed.

[0125] The quantitative evaluation of backwashing effect by zone is achieved through data acquisition via a multi-dimensional sensor array during the optimized pulse sequence execution. Each of the nine zones in the micro-interface is equipped with a flow sensor, differential pressure sensor, and concentration sensor. The sensor sampling frequency is set to 10 times per second, and data transmission uses the CAN bus protocol with a communication rate of 1 Mbps.

[0126] Dynamic identification of backflow flow fluctuation patterns is based on spectral analysis of real-time flow data, employing a short-time Fourier transform method with a transform window length of 128 sampling points and an overlap rate of 75%. Fluctuation pattern identification is achieved through a peak detection algorithm, with a detection threshold set at 2.5 times the signal standard deviation and a peak interval threshold of 0.5 seconds. The flow fluctuation feature vector contains four parameters: peak amplitude, peak frequency, fluctuation duration, and fluctuation attenuation coefficient. The feature vector is stored as a 4-dimensional array, with array elements maintained to three decimal places. Fluctuation pattern classification uses a pattern recognition algorithm to identify three basic patterns: continuous fluctuation, impulsive fluctuation, and attenuated fluctuation, with a classification accuracy requirement of over 95%.

[0127] The identification of abrupt change points in the differential pressure response characteristic curve employs a joint judgment mechanism using first and second derivatives. The first derivative is calculated through the difference between adjacent sampling points, and the second derivative is obtained through the difference of the first derivative. The criteria for identifying an abrupt change point are that the absolute value of the first derivative is greater than three times the historical mean and the sign of the second derivative changes. The recording accuracy of abrupt change points is 0.1 seconds, and the intensity of the change is quantified using the logarithm of the absolute value of the derivative. The differential pressure response curve is smoothed using cubic spline interpolation with a smoothing parameter set to 0.01 to ensure curve continuity while preserving abrupt change characteristics. The statistical analysis of abrupt change point distribution includes three dimensions: abrupt change frequency, abrupt change intensity distribution, and abrupt change time interval. The statistical results are stored as a histogram data structure. Cluster analysis of abrupt change points identifies three levels: strong, moderate, and weak abrupt changes, with level thresholds set at the 90%, 60%, and 30% quantiles of the abrupt change intensity, respectively.

[0128] The spatiotemporal evolution spectrum characterizing the backwashing effect intensity is constructed by spatiotemporally coupling the flow fluctuation pattern with the distribution of pressure drop abrupt changes. The time dimension uses a fixed time window segmentation with a window length of 5 seconds and a window overlap rate of 20%. The spatial dimension is divided according to membrane surface partitions, with each partition corresponding to a spatial node in the evolution spectrum. The effect intensity is calculated using a weighted combination of the flow fluctuation amplitude and the pressure drop abrupt change intensity, with weighting coefficients set to 0.6 and 0.4, respectively. The spatiotemporal evolution spectrum is represented by a three-dimensional matrix, with the matrix dimension being the number of time windows multiplied by the number of spatial partitions multiplied by the number of intensity components. Evolution spectrum normalization ensures that the intensity values ​​range from 0 to 1, and the normalization method uses minimum-maximum standardization. Spatiotemporal interpolation of the evolution spectrum uses a bilinear interpolation method, maintaining an interpolation accuracy of over 95% of the original sampling accuracy.

[0129] Real-time acquisition of pollutant concentration change trajectories was based on a combination of high-precision turbidity sensors and online chromatographs. The turbidity sensor had a measurement accuracy of 0.01 NTU, and the online chromatograph had a detection accuracy of 0.1 mg / L. The two measurement methods were fused using calibration curves. Noise suppression was achieved by using a moving average filter on the concentration change trajectory, with a filter window length of 5 sampling points. Deep fusion of the trajectory data and the spatiotemporal evolution spectrum employed a multiple linear regression method to establish a quantitative relationship model between concentration changes and influence intensity. The fusion weights were determined using least squares fitting, with a required fitting accuracy of at least a coefficient of determination of 0.85. The deep fusion algorithm used a recursive least squares method, with a forgetting factor set at 0.99 to ensure real-time updating of model parameters.

[0130] A multidimensional evaluation matrix reflecting the transient effects of the backwashing process is formed by reorganizing the fused data according to time and space dimensions. Matrix rows correspond to time sampling points, columns to spatial partitions, and matrix elements are the transient effect indicators of backwashing at that spatiotemporal location. The indicators include four components: flow response intensity, pressure difference change rate, concentration reduction magnitude, and energy consumption density. The matrix update frequency is synchronized with the data acquisition frequency, and the matrix storage adopts a circular buffer mechanism with a buffer depth of 3600 time points, corresponding to 6 minutes of historical data. The matrix data type is double-precision floating-point numbers, with a memory footprint of approximately 500MB. Singular value decomposition of the multidimensional evaluation matrix is ​​used for data dimensionality reduction and noise suppression, retaining over 95% of the main components of the singular value energy.

[0131] Independent component analysis (ICA) was used to decompose the flow-pressure differential response and pollutant migration characteristics in the multidimensional evaluation matrix. ICA separated the multidimensional signal into several independent source signals, each representing a specific physical process. The flow-pressure differential response characteristics were extracted using a fast ICA algorithm, with an iteration limit of 100 and a convergence threshold of 10⁻⁶. Pollutant migration characteristics were obtained through spatial gradient analysis, with gradient calculation using the finite difference method and a second-order central difference precision. The ICA results contained 3 to 5 principal components, each corresponding to a typical backwash response mode. The cumulative contribution rate of the explained variance was required to reach over 90% to ensure the completeness of feature extraction.

[0132] Principal component tracking in the multidimensional feature space employs Kalman filtering for dynamic tracking. The state vector contains three state variables: principal component amplitude, phase, and frequency. The state transition matrix is ​​constructed based on a linear time-invariant model. The diagonal elements of the process noise covariance matrix are set to 0.01, and the observation noise covariance is set to 0.005. The filter prediction step uses the state transition equation for forward derivation, and the update step uses measurement data for posterior correction. Principal component tracking accuracy is evaluated using tracking error, with the error threshold set at 5% of the theoretical value. The stability of the tracking results is evaluated using the Lyapunov exponent; an exponent less than 0 indicates stable convergence of the tracking process.

[0133] Information tracing is based on the causal relationship analysis between the principal component and the original physical quantity. The tracing process establishes a mapping relationship from changes in the principal component to specific physical phenomena, and the mapping weights are determined through correlation analysis. The strength of the causal relationship is evaluated using the Granger causality test, with a significance level set at 0.05. The tracing results form a causal relationship graph, where nodes represent physical quantities, edges represent causal relationships, and edge weights represent causal strength. The dominant change pattern of the backwashing effect is identified through graph connectivity analysis; nodes with high connectivity correspond to the dominant pattern. The time complexity of the tracing algorithm is O(n squared), where n is the number of physical quantities, and the computation time is controlled within 1 second.

[0134] Cross-correlation analysis was used to correlate the dominant change pattern with the pollutant accumulation rate spectra of each micro-interface partition. The accumulation rate spectra were obtained through spectral analysis of pollutant concentration time series, with a spectral resolution set to 0.1 Hz. The correlation mapping established correlation coefficients between the dominant pattern time series and each frequency component of the accumulation rate spectrum, calculated using the Pearson correlation coefficient. The mapping results were stored as an correlation matrix, with rows corresponding to the dominant pattern and columns corresponding to the frequency components. A correlation strength threshold of 0.3 was set; correlations exceeding this threshold were considered valid mappings. The time delay analysis of the correlation mapping was determined by the peak position of the cross-correlation function, with a time delay range between -5 seconds and +5 seconds.

[0135] A hierarchical evaluation system for backwashing effectiveness, reflecting different pollution levels in different zones, is constructed based on a multi-level structure designed using fuzzy hierarchical analysis. The evaluation system comprises three levels: a target layer, a criterion layer, and an indicator layer. The target layer provides a comprehensive evaluation of the backwashing effectiveness; the criterion layer includes three criteria: efficiency, intensity, and stability; and the indicator layer contains nine specific indicators. The weights of the hierarchical structure are determined through expert scoring and consistency checks, with a consistency ratio requirement of less than 0.1. The weight vectors of the evaluation system are stored as a tree-like data structure, and node weights support dynamic adjustment. Pollution level classification employs cluster analysis, dividing the nine zones into three levels—heavy pollution, moderate pollution, and light pollution—based on pollution characteristics, with each level corresponding to a different evaluation weight configuration.

[0136] In the hierarchical evaluation system for backwashing effectiveness, the dynamic response characteristic analysis is based on the time series characteristics of each indicator. The response delay is determined by the peak position of the cross-correlation function, with a delay time accuracy of 0.1 seconds. The response amplitude is calculated by the peak value of the signal envelope, and the envelope extraction uses the Hilbert transform method. Response stability is evaluated using the coefficient of variation, defined as the ratio of the standard deviation to the mean. Dynamic response characteristic parameters are stored as characteristic vectors, with the vector dimension equal to the number of evaluation indicators. The trend analysis of response characteristics uses a linear regression method, with the slope parameter reflecting the changing trend of the response characteristics. The p-value threshold for the slope significance test is set at 0.05.

[0137] The backwash response characteristics of each zone are decoupled in multiple levels using blind source separation technology, based on pollutant removal rate, backwash intensity utilization rate, and energy conversion efficiency. The decoupling process separates the mixed observation signals into independent source signals, each corresponding to a single physical process. The pollutant removal rate is calculated as the ratio of the concentration difference before and after backwashing to the initial concentration, with the calculation accuracy maintained to four decimal places. The backwash intensity utilization rate is defined as the ratio of the actual backwash flow rate to the theoretical maximum flow rate, which is determined based on the pump's rated parameters. The energy conversion efficiency is calculated as the ratio of effective cleaning energy to the total input energy, with the effective cleaning energy calculated based on the pollutant removal amount and separation power consumption. The multi-level decoupling includes three levels: coarse separation, fine separation, and optimized separation, each employing different algorithm parameters and convergence criteria.

[0138] After decoupling, the quantitative evaluation of each level of indicators adopts the fuzzy comprehensive evaluation method. The evaluation indicator values ​​are converted into fuzzy evaluation values ​​through membership functions, which adopt a trapezoidal distribution. The function parameters are determined based on historical data statistics. The fuzzy evaluation vector is synthesized using a weighted average method, and the weight vector is determined based on the importance of the indicators. The quantitative evaluation results include four levels: excellent, good, average, and poor, with level thresholds set at 0.8, 0.6, 0.4, and 0.2, respectively. The evaluation accuracy is verified using cross-validation, with a required accuracy rate of over 90%. The uncertainty analysis of the quantitative evaluation adopts the Monte Carlo method, with 1000 simulations and a confidence interval of 95%.

[0139] The evaluation results of backwashing effectiveness with regional differences are based on cluster analysis and difference testing. Cluster analysis uses the K-means algorithm, with three clusters corresponding to three effectiveness levels: high efficiency, medium efficiency, and low efficiency. Difference testing uses analysis of variance to examine the significant differences in evaluation results across different sub-regions. The evaluation results are output in a regional evaluation table, containing four fields: regional number, overall score, effectiveness level, and improvement suggestions. Results are updated once after each backwashing operation, and historical results are saved up to 30 operation records. Regional differences are quantified using the coefficient of variation and coefficient of dispersion; a coefficient of variation greater than 0.2 indicates a significant difference.

[0140] In one optional implementation, feature decomposition is performed based on the flow-pressure difference response characteristics and pollutant migration characteristics in the multidimensional evaluation matrix. By tracking the principal components and tracing the information sources in the multidimensional feature space, the dominant change patterns of the backwashing effect are reconstructed, including:

[0141] Dynamic fluctuation information of flow-pressure difference response characteristics and pollutant migration characteristics is extracted from the multidimensional evaluation matrix. The dynamic fluctuation information is hierarchically separated according to different time scales. The separated hierarchical features are recombined and mapped in the time-frequency coupling domain to construct a multidimensional feature space that reflects the transient response of the system.

[0142] In the multidimensional feature space, the source analysis of the coordinated change pattern of pollutant concentration and flow-pressure difference response characteristics is performed. By deeply analyzing the propagation path and diffusion intensity of the coordinated change pattern, the driving factors and hindering factors in the evolution of backwashing effect are identified. Based on the spatiotemporal distribution law of the driving factors and the hindering factors, the dominant change pattern of backwashing effect is reconstructed.

[0143] like Figure 2 As shown, the method includes:

[0144] Dynamic fluctuation information of flow-pressure differential response characteristics and pollutant migration characteristics was extracted from the multidimensional evaluation matrix. This extraction process employed a sliding window technique, dividing the backwash cycle into several continuous time periods, each set to 30 seconds, with a window interval of 5 seconds, ensuring continuous data acquisition coverage. For the flow-pressure differential response characteristics, the amplitude of flow rate changes and pressure differential changes within each time window were recorded. The process of flow rate changing from an initial value of 1.2 cubic meters per hour to a peak value of 2.8 cubic meters per hour was quantified as a flow response sequence, and the process of pressure differential changing from 0.15 MPa to 0.42 MPa was quantified as a pressure differential response sequence. For the pollutant migration characteristics, time-series data of suspended solids concentration, turbidity, and organic matter content in the backwash effluent were collected. The entire process of suspended solids concentration gradually decreasing from 850 mg / L at the beginning of backwashing to 120 mg / L at the end of backwashing was numerically characterized.

[0145] The acquired dynamic fluctuation information was hierarchically separated according to time scale. This separation process employed a multi-resolution analysis method to decompose the original time-series data into components of different frequencies. For the fast fluctuation component, the time scale was set to 5 to 30 seconds, mainly reflecting the transient impact effects during backwashing, including the immediate disturbance of the filter media by the backwash water flow and the rapid release of contaminants. For the medium-speed fluctuation component, the time scale was set to 30 to 180 seconds, mainly reflecting the transitional response characteristics during the adjustment of backwash intensity, including the gradual change in pressure difference caused by flow regulation and the phased changes in contaminant concentration. For the slow fluctuation component, the time scale was set to 180 seconds to the entire backwash cycle, mainly reflecting the overall evolution trend of the backwash effect, including the cumulative improvement in the cleanliness of the filter media and the gradual optimization of the system's operating status. Using a practical example, when the backwash cycle was 600 seconds, the fast fluctuation component captured 17 transient impact events, the medium-speed fluctuation component identified 4 transitional response stages, and the slow fluctuation component showed a monotonically decreasing overall trend.

[0146] In the time-frequency coupled domain, the separated features at each level are reconstructed and mapped. This reconstruction process establishes a joint representation space of the time and frequency dimensions, linking the feature information at each time point with its corresponding frequency distribution characteristics. For the flow-pressure differential response characteristics, a two-dimensional distribution map of the response intensity is constructed in the coupled domain. The horizontal axis represents the temporal evolution of the backwashing process, and the vertical axis represents the frequency range of the response signal. The value at each position in the map represents the energy intensity of a specific frequency component at a specific time. For example, at 240 seconds into the backwashing process, the energy intensity of the fast fluctuation band is 0.68 units, the medium fluctuation band is 0.42 units, and the slow fluctuation band is 0.29 units. For the pollutant migration characteristics, a two-dimensional distribution map of the migration rate is constructed in the coupled domain. By tracking the rate of change of pollutant concentration at different frequency components, the spatiotemporal evolution of pollutant release is identified. When backwashing reached 360 seconds, the migration rate of suspended solids decreased by 3.2 mg / L in the fast release band, 1.5 mg / L in the medium release band, and 0.4 mg / L in the slow release band.

[0147] Based on a comprehensive analysis of the above-mentioned recombinant mapping results, a multidimensional feature space reflecting the system's transient response is constructed. This feature space adopts a three-dimensional coordinate system. The first dimension is the time process dimension, with values ​​ranging from the beginning to the end of the backwash cycle. The second dimension is the response intensity dimension, with values ​​normalized from the minimum to the maximum value of the flow-pressure difference response characteristic. The third dimension is the contaminant migration dimension, with values ​​normalized from the highest to the lowest value of the contaminant concentration. In this feature space, each data point represents the system's operating state at a specific moment, and the spatial distribution trajectory of the data points reflects the dynamic evolution of the backwash process. Taking actual measurement data as an example, the coordinates of the state point at the initial stage of backwashing are 0 seconds in the time dimension, 0.35 in the response intensity dimension, and 0.92 in the contaminant migration dimension; the coordinates of the state point in the middle stage of backwashing are 300 seconds in the time dimension, 0.68 in the response intensity dimension, and 0.48 in the contaminant migration dimension; and the coordinates of the state point at the end of backwashing are 600 seconds in the time dimension, 0.41 in the response intensity dimension, and 0.15 in the contaminant migration dimension.

[0148] In a constructed multidimensional feature space, the synergistic variation pattern of pollutant concentration and flow-pressure difference response characteristics was analyzed. This analysis identified the dynamic correlation between the two types of features, quantifying the tightness of the synergistic variation by calculating the correlation between the rate of change of pollutant concentration and the rate of change of flow-pressure difference response intensity at different times. When the flow rate increased from 1.8 m³ / h to 2.4 m³ / h, the pressure difference correspondingly increased from 0.28 MPa to 0.36 MPa, and simultaneously, the suspended solids concentration decreased rapidly from 560 mg / L to 320 mg / L, indicating that the increase in flow rate directly drove the accelerated migration of pollutants. When the pressure difference was maintained at a stable level of 0.38 MPa, the rate of decrease in suspended solids concentration slowed from 2.8 mg / L / s to 1.1 mg / L / s, indicating that the driving effect of simply maintaining the pressure difference on the continuous removal of pollutants weakened.

[0149] A deep analysis of the propagation path of the co-change pattern was conducted to trace the transmission process of pollutant release signals within the system. This propagation path analysis began at the initial position of the backwash water entering the filter media layer and traced the movement of the peak pollutant concentration along the water flow direction. Within the first 90 seconds after backwashing started, the peak pollutant concentration appeared at the bottom of the filter media layer, reaching 920 mg / L. As the backwashing process progressed to 180 seconds, the peak concentration moved upwards to the middle region of the filter media layer, reaching 680 mg / L. Continuing to advance to 270 seconds, the peak concentration reached the upper region of the filter media layer, decreasing to 410 mg / L. The propagation path movement speed varied at different stages: initially at 1.8 cm / s, accelerating to 2.5 cm / s in the middle stage, and slowing to 0.9 cm / s in the later stage.

[0150] The diffusion intensity of the synergistic change pattern was quantitatively assessed to describe the spatial distribution characteristics of the pollutant release process. This diffusion intensity was characterized by measuring the degree of difference in pollutant concentration distribution at different spatial locations within the filter media layer. In the initial stage of backwashing, the suspended solids concentration at the bottom of the filter media layer was 850 mg / L, in the middle region it was 620 mg / L, and in the upper region it was 380 mg / L, with a concentration gradient of 15.6 mg / L per centimeter. In the middle stage of backwashing, the concentration at the bottom decreased to 480 mg / L, in the middle region to 390 mg / L, and in the upper region to 240 mg / L, with the concentration gradient decreasing to 8.0 mg / L per centimeter. In the later stage of backwashing, the concentrations in each region tended to homogenize, and the concentration gradient further decreased to 2.3 mg / L per centimeter.

[0151] Based on in-depth analysis of propagation paths and diffusion intensity, driving factors in the evolution of backwashing effectiveness were identified. These driving factors include the mechanical scouring effect of the backwash water flow, the expansion and loosening effect of the filter media, and the adsorption and dissociation process of pollutants with the filter media. The driving intensity of the mechanical scouring effect was quantified by the product of flow velocity and pressure difference; when this product increased from an initial 0.18 to 0.54, the pollutant removal rate increased by 2.6 times. The driving intensity of the filter media expansion and loosening effect was quantified by the expansion rate of the filter media thickness; when the expansion rate increased from 15% to 35%, the unobstructedness of the pollutant migration channels significantly improved, and the migration resistance decreased by 42%. The driving intensity of the pollutant dissociation process was related to the turbulence intensity of the backwash water flow; when the turbulence intensity increased from 0.32 to 0.71, the pollutant dissociation rate accelerated by 1.9 times.

[0152] The hindering factors in the evolution of backwashing effectiveness were identified. These factors included localized clogging residue in the filter media, contaminant redeposition, and uneven hydraulic distribution during backwashing. The hindering effect of localized clogging residue was quantified by the proportion of residual contaminants; when the residue proportion reached 18%, the efficiency of subsequent backwashing processes decreased by 23%. The hindering effect of contaminant redeposition was assessed by monitoring the localized rebound in contaminant concentration during the later stages of backwashing; when the concentration rebound reached 12% of the original decrease, the actual removal efficiency was reduced by 9%. The hindering effect of uneven hydraulic distribution was quantified by measuring the coefficient of variation of flow velocity in different areas; when the coefficient of variation was 0.26, the overall backwashing efficiency decreased by 16%.

[0153] The spatiotemporal distribution patterns of driving and hindering factors were analyzed, and descriptive models of factor intensity variations with time and spatial location were established. In the temporal dimension, the intensity of the driving factor rapidly increased within the first 120 seconds after backwashing started, rising from an initial value of 0.22 to a peak of 0.79, then remained relatively stable between 120 and 480 seconds, gradually decreasing to 0.35 after 480 seconds. The intensity of the hindering factor showed a trend of initially increasing slowly and then rapidly decreasing throughout the backwashing process, reaching a peak of 0.48 at 300 seconds, and then gradually decreasing to 0.11 as pollutants continued to be removed. In the spatial dimension, the driving factor showed the greatest intensity at the bottom of the filter media (0.83), followed by the middle region (0.61) and the upper region (0.44). The hindering factor showed an intensity of 0.52 at the bottom of the filter media, 0.39 in the middle region, and 0.27 in the upper region.

[0154] Based on the spatiotemporal distribution patterns of driving and inhibiting factors, the dominant change pattern of backwashing effect is reconstructed. This reconstruction process spatiotemporally couples the combined effects of driving and inhibiting factors to generate a dominant trend line for the evolution of backwashing effect. The dominant change pattern exhibits three typical stages: The first stage is the rapid purification stage, lasting from 0 to 180 seconds after backwashing starts. In this stage, driving factors dominate, with the pollutant removal rate reaching 4.1 mg / L, and the cumulative removal accounting for 56% of the total removal. The second stage is the stable removal stage, lasting from 180 to 420 seconds. In this stage, driving and inhibiting factors mutually constrain each other to reach a dynamic equilibrium, with the pollutant removal rate maintained at 1.8 mg / L, and the cumulative removal accounting for 32% of the total removal. The third stage is the finishing optimization stage, lasting from 420 seconds until the end of backwashing. In this stage, inhibiting factors gradually diminish, the pollutant removal rate drops to 0.6 mg / L, and the cumulative removal accounts for 12% of the total removal. The overall trajectory of the dominant change pattern shows an orderly migration path from regions with high pollutant concentration and high response intensity to regions with low pollutant concentration and low response intensity in the multidimensional feature space. The spatial geometric features of this path reflect the physical nature of the backwashing process from violent disturbance to smooth convergence.

[0155] A second aspect of the present invention provides an intelligent control system for a micro-interface oil removal device based on backwashing circulation, comprising:

[0156] The acquisition module is used to acquire real-time operating status data and historical backwash data of the micro-interface oil removal device. Based on the operating status data, a pollution state evolution model considering the spatiotemporal coupling effect is constructed. The pollution state evolution model is used to predict the accumulation trend and diffusion law of pollutants in each region of the micro-interface, and obtain the zonal pollution assessment results with time-series characteristics.

[0157] The module is used to construct a multi-objective optimization function based on the pollutant accumulation rate and diffusion characteristics in the zoning pollution assessment results, combined with the system backwashing resource constraints, and generate initial pulse sequence schemes for zoning with different pollution levels.

[0158] The determination module is used to dynamically reconstruct the initial pulse sequence scheme based on the spatial distribution characteristics and time-varying patterns of pollutants in the zonal pollution assessment results, and to form an optimized pulse sequence with spatial adaptability by adjusting the temporal relationship and range of action of each pulse unit.

[0159] The optimization module is used to establish a graded evaluation criterion for the backwashing effect during the execution of the optimized pulse sequence backwashing operation. Based on the differences in pollution characteristics in different regions of the micro-interface, it performs a zonal quantitative evaluation of the actual effect of the backwashing process. Based on the results of the zonal quantitative evaluation, it selectively adjusts the parameters of the pollution state evolution model for different pollution level regions to achieve optimization and improvement of the backwashing strategy.

[0160] A third aspect of the present invention provides an electronic device, comprising:

[0161] processor;

[0162] Memory used to store processor-executable instructions;

[0163] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0164] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0165] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent control method for a micro-interface oil removal device based on backwashing circulation, characterized in that, include: Real-time operating status data and historical backwash data of the micro-interface oil removal device are obtained. Based on the operating status data, a pollution state evolution model considering the spatiotemporal coupling effect is constructed. The pollution state evolution model is used to predict the pollutant accumulation trend and diffusion law of each region of the micro-interface, and obtain the zonal pollution assessment results with time-series characteristics. Based on the pollutant accumulation rate and diffusion characteristics in the zonal pollution assessment results, and combined with the system backwashing resource constraints, a multi-objective optimization function is constructed to generate initial pulse sequence schemes for zonals with different pollution levels. Based on the spatial distribution characteristics and time-varying patterns of pollutants in the zonal pollution assessment results, the initial pulse sequence scheme is dynamically reconstructed. By adjusting the temporal relationship and range of action of each pulse unit, an optimized pulse sequence with spatial adaptability is formed. During the backwashing operation of the optimized pulse sequence, a graded evaluation criterion for the backwashing effect is established. Based on the differences in contamination characteristics in different regions of the micro-interface, the actual effect of the backwashing process is evaluated quantitatively by region. Based on the results of the quantitative evaluation of the zoning, the parameters of the pollution state evolution model are selectively adjusted for areas with different pollution levels to optimize and improve the backwashing strategy.

2. The method according to claim 1, characterized in that, Based on the operational status data, a pollution state evolution model considering spatiotemporal coupling effects is constructed. This model predicts the pollutant accumulation trends and diffusion patterns in each region of the micro-interface, yielding temporal-series-based zonal pollution assessment results, including: Multidimensional spectral decomposition is performed on the pressure difference characteristics, flow characteristics and pollutant accumulation characteristics in the operation status data. The decomposed feature information is then finely reconstructed according to the spatial structure spectrum of the micro-interface to construct a pollution state evolution model that reflects the pollution state of each region. By tracing the inter-regional pollutant migration relationship in the pollution state evolution model, a spatial correlation spectrum is established. Based on the spatiotemporal structural characteristics of the pollution state evolution model, the pollution state data at different times are expanded in the time domain. By hierarchically stripping and progressively recombining the dominant features of the expanded sequence, the evolution pattern of pollutant accumulation rate is restored. The migration path and diffusion intensity information in the spatial correlation spectrum are deeply coupled with the evolution pattern to construct a state evolution function. Based on the spatiotemporal coupling relationship in the state evolution function, the trajectory of pollutant concentration change in each region of the micro-interface is dynamically tracked and predicted. By partitioning and analyzing the spatiotemporal characteristics of the pollutant concentration change trajectory, an assessment result reflecting the pollution evolution law of each region is generated.

3. The method according to claim 2, characterized in that, Based on the spatiotemporal structural characteristics of the pollution state evolution model, pollution state data at different times are expanded in the time domain. By hierarchically stripping and progressively recombining the dominant features of the expanded sequence, the evolution pattern of pollutant accumulation rate is restored, including: Based on the spatiotemporal structure spectrum in the pollution state evolution model, the pollution state information at different times is separated into multiple scales in the time domain to construct a feature spectrum matrix that reflects the progressive evolution of pollutants in the time dimension. The dominant fluctuation mode and secondary fluctuation mode of the pollutant accumulation process are extracted from the feature spectrum matrix. Feature enhancement is performed based on the main changing trends in the dominant fluctuation mode. By hierarchically deconstructing and sequentially reorganizing its temporal energy distribution, the baseline evolution path of pollutant accumulation is extracted, and the local perturbation features in the secondary fluctuation mode are transformed into modulation functions of the baseline evolution path. In the time-frequency joint domain, the reference evolution path and the correction effect of the modulation function are deeply integrated. By introducing time-varying weight coefficients, the fusion process is adaptively adjusted to reconstruct the evolution mode of pollutant accumulation rate.

4. The method according to claim 1, characterized in that, Based on the pollutant accumulation rate and diffusion characteristics in the zoning pollution assessment results, and combined with the system backwashing resource constraints, a multi-objective optimization function is constructed to generate initial pulse sequence schemes for zoning with different pollution levels, including: The pollutant accumulation rate and pollutant diffusion characteristics of each zone are analyzed from the pollution assessment results of the zones. Based on the pollutant accumulation rate, the pollutant removal demand intensity of each zone is decomposed. Based on the pollutant diffusion characteristics, the pollutant migration influence coefficient between each zone and adjacent zones is restored. Based on the intensity of pollutant removal demand and the pollutant migration influence coefficient, a multi-objective optimization function is established with the joint optimization objectives of maximizing oil removal efficiency and minimizing backwashing energy consumption. Import backwashing resource constraints, transform the backwashing resource constraints into the constraint boundary of the multi-objective optimization function, solve the multi-objective optimization function, and obtain the pulse parameter combination that maximizes the oil removal efficiency and minimizes the backwashing energy consumption to achieve Pareto optimality under the backwashing resource constraints. The pulse energy allocation scheme and pulse duration allocation scheme for different pollution levels in the pulse parameter combination are combined in a gradient to construct an initial pulse sequence scheme.

5. The method according to claim 1, characterized in that, Based on the spatial distribution characteristics and time-varying patterns of pollutants in the zonal pollution assessment results, the initial pulse sequence scheme is dynamically reconstructed. By adjusting the temporal relationship and range of action of each pulse unit, an optimized pulse sequence with spatial adaptability is formed, including: The pollution assessment results of the zoning are subjected to multidimensional profile separation, and the pollutant concentration distribution data and concentration change rate data of each zoning are purified. A spatial coupling matrix that characterizes the correlation of pollutant concentration between zoning ... Based on the concentration change rate data, we deepen the temporal extension analysis, track the pollutant concentration evolution trajectory of each zone in the future time window, and combine the gradient difference between the pollutant concentration evolution trajectory and the current pollutant concentration to identify the rapid deterioration area of ​​pollutants where the concentration increment exceeds the preset increment threshold. The rapidly deteriorating region of the pollutant is projected onto the spatial propagation spectrum, and the rapidly deteriorating region located at the propagation source is selected as the intervention region. Based on the hierarchical relationship of the intervention region in the spatial propagation spectrum, the priority execution order of each pulse unit in the initial pulse sequence scheme is reconstructed. Around the pollutant diffusion gradient between adjacent partitions in the spatial propagation spectrum, a domain expansion function is constructed for each pulse unit. Based on the domain expansion function, the radiation range of the pulse unit acting on the intervention area is progressively extended to multiple consecutive partitions in the spatial propagation spectrum to form an optimized pulse sequence.

6. The method according to claim 1, characterized in that, During the backwashing operation of the optimized pulse sequence, a graded evaluation criterion for the backwashing effect is established. Based on the differences in contamination characteristics in different regions of the micro-interface, the actual effect of the backwashing process is quantitatively evaluated by region, including: During the backwashing operation of the optimized pulse sequence, the backwash flow fluctuation pattern of different partitions of the micro-interface is dynamically identified. Combined with the distribution of abrupt change points of the differential pressure response characteristic curve, a spatiotemporal evolution spectrum characterizing the intensity of the backwashing action is constructed. The spatiotemporal evolution spectrum is deeply integrated with the real-time collected pollutant concentration change trajectory to form a multidimensional evaluation matrix reflecting the transient effect of the backwashing process. Based on the flow-pressure difference response characteristics and pollutant migration characteristics in the multidimensional evaluation matrix, feature decomposition is performed. By tracking the principal components of the multidimensional feature space and tracing the information source, the dominant change pattern of the backwashing effect is restored. The dominant change pattern is correlated and mapped with the pollutant accumulation rate spectrum of each partition of the micro-interface to construct a hierarchical evaluation system of backwashing effect that reflects different pollution levels of partitions. Based on the dynamic response characteristics in the backwash effect hierarchical evaluation system, the backwash response characteristics of each zone are decoupled in multiple levels according to pollutant removal rate, backwash intensity utilization rate and energy conversion efficiency. By quantitatively evaluating the decoupled indicators at each level, a backwash effect evaluation result with zone differences is formed.

7. The method according to claim 6, characterized in that, Based on the flow-pressure difference response characteristics and pollutant migration characteristics in the multidimensional evaluation matrix, feature decomposition is performed. By tracing the principal components and information sources in the multidimensional feature space, the dominant change patterns of the backwashing effect are reconstructed, including: Dynamic fluctuation information of flow-pressure difference response characteristics and pollutant migration characteristics is extracted from the multidimensional evaluation matrix. The dynamic fluctuation information is hierarchically separated according to different time scales. The separated hierarchical features are recombined and mapped in the time-frequency coupling domain to construct a multidimensional feature space that reflects the transient response of the system. In the multidimensional feature space, the source analysis of the coordinated change pattern of pollutant concentration and flow-pressure difference response characteristics is performed. By deeply analyzing the propagation path and diffusion intensity of the coordinated change pattern, the driving factors and hindering factors in the evolution of backwashing effect are identified. Based on the spatiotemporal distribution law of the driving factors and the hindering factors, the dominant change pattern of backwashing effect is reconstructed.

8. An intelligent control system for a micro-interface oil removal device based on backwashing circulation, used to implement the method of any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire real-time operating status data and historical backwash data of the micro-interface oil removal device. Based on the operating status data, a pollution state evolution model considering the spatiotemporal coupling effect is constructed. The pollution state evolution model is used to predict the accumulation trend and diffusion law of pollutants in each region of the micro-interface, and obtain the zonal pollution assessment results with time-series characteristics. The module is used to construct a multi-objective optimization function based on the pollutant accumulation rate and diffusion characteristics in the zoning pollution assessment results, combined with the system backwashing resource constraints, and generate initial pulse sequence schemes for zoning with different pollution levels. The determination module is used to dynamically reconstruct the initial pulse sequence scheme based on the spatial distribution characteristics and time-varying patterns of pollutants in the zonal pollution assessment results, and to form an optimized pulse sequence with spatial adaptability by adjusting the temporal relationship and range of action of each pulse unit. The optimization module is used to establish a graded evaluation criterion for the backwashing effect during the backwashing operation of the optimized pulse sequence, and to perform a zonal quantitative evaluation of the actual effect of the backwashing process based on the differences in contamination characteristics in different regions of the micro-interface. Based on the results of the quantitative evaluation of the zoning, the parameters of the pollution state evolution model are selectively adjusted for areas with different pollution levels to optimize and improve the backwashing strategy.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images