A method for optimizing the compound ratio of active coke based on machine learning

Abstract

This invention discloses a machine learning-based method for optimizing the blending ratio of activated carbon, relating to the field of intelligent adsorption blending technology. The method includes: collecting batch attribute information of partially polar and partially non-polar activated carbon; performing feature normalization and sample labeling on the batch attribute information to generate batch learning features; reading the blending ratios that are ranked high and where polar and non-polar residues converge evenly based on the blending ratio suitability ranking, and determining the layered filling structure in conjunction with the pollutant polarity distribution to generate a blending implementation plan; executing the blending of partially polar and partially non-polar activated carbon according to the blending implementation plan, collecting adsorption operation feedback, obtaining feedback correction features, correcting the blending ratio prediction direction based on the feedback correction features, determining the corrected blending ratio and blending method, and generating an optimized carbon blending scheme. This invention improves blending suitability, residue control stability, and continuous adsorption effect.

Description

Technical Field

[0001] This invention relates to the field of intelligent adsorption compounding technology, and in particular to a method for optimizing the compounding ratio of activated char based on machine learning. Background Technology

[0002] Activated carbon, with its tunable pore structure, abundant surface functional groups, and wide range of adsorption applications, has been widely used in flue gas purification, waste gas treatment, advanced wastewater treatment, and synergistic removal of multi-component pollutants. Among existing methods, a common approach is to combine activated carbon with different modification directions based on pollutant type, polarity differences, operating temperature, humidity conditions, and batch indicators of activated carbon, combined with machine learning to analyze historical adsorption data and batch characteristics, in order to balance the adsorption capacity for both polar and non-polar pollutants.

[0003] Existing activated carbon compounding optimization processes still rely heavily on empirical ratios, single-round verification, or evaluation by a single adsorption index when facing batch differences between polar and non-polar activated carbon, changes in pollutant polarity distribution, and fluctuations in adsorption residues. They lack continuous deduction and feedback correction for the equilibrium convergence relationship between polar and non-polar residues for different compounding ratios. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a machine learning-based method for optimizing the ratio of activated carbon compounding, which solves the problems of difficulty in dynamically optimizing the ratio of activated carbon compounding and difficulty in controlling the adsorption residue in a balanced manner.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a machine learning-based method for optimizing the blending ratio of activated carbon, comprising: collecting batch attribute information of partially polar activated carbon and partially non-polar activated carbon; performing feature normalization and sample labeling on the batch attribute information to generate batch learning features; labeling the batch learning features with historical adsorption performance to obtain perturbation learning features; performing machine learning ratio prediction on the perturbation learning features to deduce the adsorption adaptation performance of different blending ratios and generate a candidate blending ratio sequence; predicting the fluctuation of polar and non-polar residues item by item according to the candidate blending ratio sequence, retaining the blending ratio with the minimum fluctuation and eliminating the blending ratio with a sudden increase in fluctuation, and generating a blending ratio adaptation ranking; reading the blending ratios that are ranked first and have balanced convergence of polar and non-polar residues based on the blending ratio adaptation ranking, and determining the layered filling structure in combination with the polarity distribution of pollutants to generate a blending implementation plan; executing the blending of partially polar activated carbon and partially non-polar activated carbon according to the blending implementation plan, collecting adsorption operation feedback, obtaining feedback correction features, correcting the blending ratio prediction direction according to the feedback correction features, determining the corrected blending ratio and blending method, and generating an optimized carbon blending scheme.

[0008] As a preferred embodiment of the active coke blending ratio optimization method based on machine learning described in this invention, the specific steps for generating coke batch learning features are as follows:

[0009] Field checks and anomaly removal are performed on batch attribute information and partially nonpolar active focal groups to obtain valid batch information. The valid batch information is then categorized and scaled to generate normalized batch features.

[0010] Samples are labeled with the activated coke category, historical adsorption performance, and batch stability of the normalized batch characteristics to generate coke batch learning features.

[0011] As a preferred embodiment of the machine learning-based active coke blending ratio optimization method of the present invention, the specific steps of obtaining perturbation learning features by labeling the coke batch learning features with historical adsorption performance are as follows:

[0012] Based on the active coke category label and batch stability status label in the coke batch learning features, the historical adsorption performance under the same adsorption conditions is linked to the coke batch learning features item by item according to batch number, pollutant polarity type and adsorption operation stage to generate a coke batch adsorption correspondence table.

[0013] The corresponding changes of polar activated coke and non-polar activated coke in adsorption perturbation index were extracted from the coke batch adsorption correspondence table, and the occupancy difference and residual change of the two types of activated coke were coupled and labeled to generate competitive residual coupling labels.

[0014] Perturbation-graded samples are identified from competing residual coupling markers and written back to the char batch adsorption correspondence table to generate a perturbation-labeled sample table.

[0015] Based on the perturbation-labeled sample table, low-perturbation samples with continuous adsorption performance and competing perturbation samples are retained, while high-fluctuation perturbation samples with sudden increases in residuals and decreased adsorption are removed. The foci-batch learning features, perturbation types and historical adsorption performance in the retained samples are merged to generate perturbation learning features.

[0016] As a preferred embodiment of the machine learning-based activated carbon compound ratio optimization method of the present invention, the specific steps of performing machine learning ratio prediction on perturbation learning features, deducing the adsorption adaptation performance of different compound ratios, and generating candidate ratio sequences are as follows:

[0017] The perturbation learning features are hierarchically merged to obtain samples of the same layer, and the ratio response labels in the samples of the same layer are bound to generate a ratio learning sample table.

[0018] Based on the proportional learning sample table, the perturbation type of the current coke batch is matched with the perturbation type of the historical samples, the historical samples with opposite perturbation directions are removed, and the position of the change of the compound ratio in the historical samples with the synchronous convergence of polar and non-polar residues is identified by machine learning. The proportional convergence center of the current coke batch is predicted, and the proportional neighborhood is expanded around the proportional convergence center to generate a set of ratios to be deduced.

[0019] Based on the set of ratios to be deduced, each ratio is backfilled into the polar adsorption response and non-polar adsorption response of the current batch of coke, and the corresponding polar residue, non-polar residue, residue fluctuation and adsorption retention change are predicted using the historical samples after gating and matching, and a ratio adsorption deduction table is generated.

[0020] The ratios of polar and non-polar residues that decrease simultaneously, whose residue fluctuations continuously converge and whose adsorption remains unchanged are identified from the ratio adsorption deduction table. These ratios are then arranged in order of adsorption compatibility from best to worst to generate candidate ratio sequences.

[0021] As a preferred embodiment of the activated carbon compound ratio optimization method based on machine learning described in this invention, the specific steps of predicting the fluctuations of polar and non-polar residues item by item according to the candidate ratio sequence, retaining the ratio with the minimum fluctuation and meeting the residue standard, and eliminating the ratio with a sudden increase in fluctuation are as follows:

[0022] Based on the candidate ratio sequence, the adsorption adaptation performance corresponding to different compound ratios is read one by one, and the different compound ratios are input into the ratio perturbation prediction rule to predict polar and non-polar residues, and generate a ratio residue prediction set.

[0023] The ratio residual prediction set is continuously arranged to obtain the information of adjacent variables of dual residuals, and the difference between the adjacent variables of dual residuals is compared before and after to generate a dual residual fluctuation sequence.

[0024] The fluctuation surge position is identified by the dual-residue fluctuation sequence, where the non-polar residue suddenly increases and the subsequent compound ratio does not fall back to correct it. The compound ratio corresponding to the fluctuation surge position is removed from the candidate ratio sequence to generate a stable candidate ratio set.

[0025] As a preferred embodiment of the active coke blend ratio optimization method based on machine learning described in this invention, the specific steps for generating the blend ratio adaptation ranking are as follows:

[0026] The stable candidate formulation set is sorted by dual residual fluctuation gating, the formulation fit value is calculated, and the stable candidate formulation set is sorted from high to low according to the formulation fit value. The formulation ratios that meet the control requirements for both polar and non-polar residues and have the highest formulation fit value are retained to generate the compliant stable formulation set.

[0027] The adjacent order of the stable ratio set that meets the standard is checked. The ratio of the compound with the decrease in the ratio fit value and the increase in the fluctuation of the two residues is moved to the back, and the ratio of the compound with the stable ratio fit value and the small fluctuation of the two residues is moved to the front, so as to generate the ratio fit ranking.

[0028] As a preferred embodiment of the activated carbon compound ratio optimization method based on machine learning described in this invention, the specific steps of reading the compound ratios that are ranked first and have a balanced convergence of polar and non-polar residues based on the ratio adaptation sorting, and determining the layered filling structure in combination with the polarity distribution of pollutants, are as follows:

[0029] Based on the matching ratio, read the top-ranked stable compound ratios that meet the standards, and extract the compound ratios from the stable compound ratios where both polar residual changes and non-polar residual changes converge continuously, to generate balanced convergent ratios.

[0030] Based on the balanced convergence ratio, the proportions of polar active coke and non-polar active coke are read, and the stable segment of dual residue change is confirmed by combining the dual residue fluctuation sequence corresponding to the balanced convergence ratio, and the target compound ratio is generated.

[0031] The polarity migration characteristics in the target wastewater detection data are read according to the target compounding ratio, and the distribution of the polarity migration characteristics is analyzed according to the strength of pollutant polarity and the order of influent migration to generate the pollutant polarity distribution.

[0032] Polar gradient stratified filling analysis was performed on the polarity distribution of pollutants to determine the filling tendency of predominantly polar activated carbon and predominantly non-polar activated carbon in different filling layers, and to generate stratified filling tendency information.

[0033] Based on the information on the layered filling tendency, the inlet section is configured as a polarity-enhanced layer, the filling position corresponding to the stable section of dual residual changes is configured as a balanced mixing layer, and the filling position with a high non-polar penetration tendency is configured as a non-polar interception layer, thus generating a layered filling structure.

[0034] As a preferred embodiment of the active coke blending ratio optimization method based on machine learning described in this invention, the blending implementation scheme is obtained by reading the layered configuration features corresponding to each filling layer according to the layered filling structure, performing an overall check on the layered configuration features, obtaining blending execution elements, determining the blending execution elements of biased polar active coke and biased non-polar active coke, and performing hierarchical arrangement and parameter adjustment of the blending execution elements.

[0035] As a preferred embodiment of the activated carbon blending ratio optimization method based on machine learning described in this invention, the steps of performing a blending of biased polar activated carbon and biased non-polar activated carbon according to the blending implementation scheme, and collecting adsorption operation feedback to obtain feedback correction features are as follows:

[0036] According to the compounding implementation plan, the compounding execution elements are read, and the compounding layering treatment of biased polar activated carbon and biased non-polar activated carbon is determined according to the compounding execution elements to generate the actual compounding layer group.

[0037] The actual compound layer group was monitored for operation. The adsorption operation status at the outlet of each filling layer was collected along the water inlet direction. Adsorption operation feedback was obtained. The adsorption operation feedback was checked against the compound implementation plan, and adsorption offset characteristics were extracted.

[0038] The adsorption offset features are merged in time sequence. Residual offsets and early penetration changes with consistent direction within continuous running segments are merged into the same correction segment, short-term isolated offsets are eliminated, and feedback correction features are generated.

[0039] As a preferred embodiment of the active coke blending ratio optimization method based on machine learning described in this invention, the specific steps for correcting the blending ratio prediction direction based on feedback correction features, determining the corrected blending ratio and blending method, and generating the coke blending optimization scheme are as follows:

[0040] Feedback offset correction is performed on the feedback correction features to determine the correction direction of the proportion of polarized active focal groups and the proportion of non-polarized active focal groups, and proportional correction direction information is generated.

[0041] Based on the proportional correction direction information, the target compound ratio is corrected in direction, and the filling amount in the biased reinforcement layer, the balanced mixing layer and the non-polar retention layer is adjusted in a coordinated manner to generate the corrected compound ratio.

[0042] Based on the verification of the interlayer connection relationship in the layered filling structure by correcting the compounding ratio, the interlayer transition method and the requirement for uniform mixing are adjusted to generate a corrected compounding method;

[0043] Based on the modified blending ratio and modified blending method, the prediction direction of the blending ratio in the ratio disturbance prediction rule is written back and corrected to generate the coking optimization scheme.

[0044] The beneficial effects of this invention are as follows: by constructing a perturbation learning basis through the learning characteristics of coke batches and historical adsorption performance, and using machine learning to predict the compounding ratio, a stable ratio with balanced convergence of dual residues is screened, and a layered filling structure is formed by combining the polarity distribution of pollutants. Based on the adsorption operation feedback, the ratio direction and compounding method are corrected, and the activated coke compounding is transformed from experience-based selection to prediction, implementation, and feedback closed-loop optimization, which improves the compatibility of compounding, the stability of residue control, and the continuous adsorption effect. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of the overall process for optimizing the blending ratio of activated coke based on machine learning.

[0047] Figure 2 This is a schematic diagram of the process for optimizing the blending ratio of activated coke.

[0048] Figure 3 This is a flowchart for the collaborative optimization of compounding ratio and layered filling.

[0049] Figure 4 The flowchart shows the proportional perturbation prediction rule and the dual-residual gating sorting process.

[0050] Figure 5 This is a comparison chart of equilibrium convergence under different compound ratios.

[0051] Figure 6 This is a comparison chart of pollutant penetration time under different compounding methods. Detailed Implementation

[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0053] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0054] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0055] Reference Figures 1-6 This is one embodiment of the present invention, which provides a method for optimizing the ratio of active coke blends based on machine learning, including the following steps:

[0056] S1. Collect batch attribute information of partially polar activated coke and partially non-polar activated coke, perform feature normalization and sample labeling on the batch attribute information, and generate coke batch learning features.

[0057] S1.1 Perform field verification and anomaly removal on batch attribute information and partially nonpolar active focal groups to obtain valid batch information. Then, perform category binding and scale unification on the valid batch information to generate normalized batch features.

[0058] It should be noted that each batch field element in the batch attribute information is checked item by item, and the batch field elements are verified to correspond with the preset batch field table to confirm the record status of the batch attribute information in terms of field completeness, source consistency, value standardization, and time correspondence. Abnormal batch items in the batch attribute information are removed. Batch abnormal items that can be recovered by supplementing with adjacent records in the same batch or by format conversion are corrected and retained. Batch abnormal items whose source cannot be confirmed, whose values ​​cannot be recovered, or that conflict with category records are deleted to obtain valid batch information.

[0059] Based on the activated coke category identifier in the batch validity information, the batch validity information corresponding to the partially polar activated coke and the partially non-polar activated coke are respectively assigned to their corresponding categories, completing the category binding; the coke batch attribute elements in the batch validity information after category binding are scaled uniformly, and different dimensions, different value directions, and different record orders are converted into a unified comparison benchmark; the scaled batch validity information is arranged according to the same field position, the same value range, and the same category order, so that the different batch attributes of partially polar activated coke and partially non-polar activated coke form comparable normalized features, generating normalized batch features.

[0060] It should also be noted that batch attribute information refers to the basic recorded information used to characterize the differences between different production batches of partially polar activated carbon and partially non-polar activated carbon. This typically includes activated carbon category, batch number, pore structure, surface functional groups, particle size distribution, moisture content, ash characteristics, modification method, production source, detection time, and corresponding initial adsorption performance. By collecting batch attribute information, the differences in adsorption capacity, stability, and compatibility between different batches of activated carbon can be determined, providing a data foundation for subsequent generation of batch learning features and machine learning-based proportion prediction.

[0061] S1.2. Sample annotation is performed on the activated coke category, historical adsorption performance, and batch stability status of the normalized batch characteristics to generate coke batch learning features.

[0062] It should be noted that when labeling samples for the normalized batch characteristics, first read the activated carbon category identifier in the normalized batch characteristics, and establish category labels according to the category affiliation of polar activated carbon and non-polar activated carbon; then read the adsorption performance elements in the normalized batch characteristics, organize the adsorption performance elements according to the time sequence, and identify the adsorption performance type according to the change relationship of the adsorption performance elements in the continuous records, forming adsorption performance labels.

[0063] The batch stability elements in the normalized batch features are read, and the consistency of the batch stability elements with the continuous records of the same category batches is compared. The batch stability type is determined based on the comparison difference, and a batch stability label is formed. Finally, the category label, adsorption performance label and batch stability label are integrated into sample annotation information, and the sample annotation information is written into the batch position corresponding to the normalized batch features to generate the focused batch learning features.

[0064] It should also be noted that coke batch learning features refer to feature data that can be used for machine learning processing, formed by processing batch attribute information through field verification, anomaly removal, category binding, scale unification, and sample labeling. These features are primarily used to characterize the category affiliation, attribute differences, historical adsorption performance, and batch stability of partially polar and partially nonpolar activated coke in different batches. By generating coke batch learning features, scattered batch records can be transformed into a comparable, trainable, and predictable data foundation, providing support for subsequent perturbation learning feature construction, compounding ratio prediction, and candidate compounding sequence generation.

[0065] S2. Label the learning features of the coke batch with the historical adsorption performance, obtain the perturbation learning features, perform machine learning ratio prediction on the perturbation learning features, deduce the adsorption adaptation performance of different compound ratios, and generate candidate ratio sequences.

[0066] S2.1 Based on the active coke category label and batch stability status label in the coke batch learning characteristics, the historical adsorption performance under the same adsorption conditions is linked to the coke batch learning characteristics item by item according to the batch number, pollutant polarity type and adsorption operation stage to generate a coke batch adsorption correspondence table.

[0067] It should be noted that, based on the activated carbon category label and batch stable state label in the batch learning characteristics, the batch learning characteristics corresponding to polar activated carbon and non-polar activated carbon are first distinguished according to the activated carbon category label. Then, the batch learning characteristics with continuous and comparable batch records are screened out according to the batch stable state label. The screened batch learning characteristics are used as the linking objects. Adsorption records corresponding to the same batch are found from the historical adsorption performance according to the batch number. The adsorption performance of polar pollutants and non-polar pollutants are further classified into the corresponding batch positions according to the pollutant polarity type. After the batch number and pollutant polarity type are consistent, the historical adsorption performance is split into stages according to the adsorption operation stage, so that the initial adsorption stage, stable adsorption stage and breakthrough change stage correspond to the batch positions in the batch learning characteristics. The historical adsorption performance with the completed stage split is linked to the batch learning characteristics item by item. The consistency of the linked batch number, pollutant polarity type, adsorption operation stage, activated carbon category label and batch stable state label is checked. Records that cannot form a corresponding relationship for the same adsorption condition are deleted. Records with complete corresponding relationships are retained and arranged in batch order to generate a batch adsorption correspondence table.

[0068] S2.2 Extract the corresponding changes of polar activated coke and non-polar activated coke in the adsorption disturbance index from the coke batch adsorption correspondence table, and couple the occupancy difference and residual change of the two types of activated coke to generate competitive residual coupling markers.

[0069] It should be noted that, based on the batch adsorption correspondence table, records for partially polar activated carbon and partially non-polar activated carbon were located according to the same adsorption conditions, the same adsorption operation stage, and the same pollutant polarity type. The adsorption disturbance indicators in both types of records were arranged side-by-side in chronological order of batch numbers to form comparable disturbance positions. Within these comparable disturbance positions, the direction, magnitude, and residual change of the occupancy of partially polar and partially non-polar activated carbon were compared to determine the corresponding changes of the two types of activated carbon in the adsorption of polar and non-polar pollutants. The results showed that an increase in the occupancy of partially polar activated carbon led to a decrease in polar residue and a decrease in non-polar residue. Records showing increased residue are marked as biased polarity competitive occupancy. Records showing decreased nonpolar residue and increased polar residue after increased nonpolarity activated carbon occupancy are marked as biased nonpolarity competitive occupancy. The two types of competitive occupancy are then bound to the corresponding residue change locations. Next, the binding competitive occupancy is checked according to the adsorption operation stage to see if it continuously affects subsequent residue changes. Occupancy differences and residue changes that can form a sequential relationship are retained, while occasional changes that only appear in a single record and cannot be continued are eliminated. Finally, the retained occupancy differences, residue changes, pollutant polarity type, and adsorption operation stage are coupled and marked to generate competitive residue coupling marks.

[0070] S2.3 Identify perturbation-graded samples from the competing residual coupling markers and write them back to the coke adsorption correspondence table to generate a perturbation-labeled sample table.

[0071] It should be noted that the coupling relationship between site differences and residue changes is read from the competitive residue coupling markers according to batch number, pollutant polarity type, and adsorption operation stage. The coupling relationships of stable residue changes, unilateral residue increases, simultaneous fluctuations of both residues, and decreased adsorption retention within the same batch are respectively assigned to different disturbance judgment positions. At each disturbance judgment position, the site difference direction, polar residue change direction, non-polar residue change direction, and adsorption retention change of the partially polar activated coke and partially non-polar activated coke are continuously compared to determine the disturbance classification samples corresponding to low disturbance, competitive disturbance, and high fluctuation disturbance. The disturbance classification samples are then reverse-located to the same record position in the coke batch adsorption correspondence table according to batch number, pollutant polarity type, and adsorption operation stage. The disturbance level, site difference direction, residue change direction, and adsorption retention change in the disturbance classification samples are written into the corresponding record positions. Finally, the complete record integrity of the rewritten coke batch adsorption correspondence table is checked, retaining records where the disturbance level corresponds to historical adsorption performance, and generating a disturbance labeling sample table.

[0072] S2.4. Based on the perturbation-labeled sample table, retain low-perturbation samples and competing perturbation samples with continuous adsorption performance, remove high-fluctuation perturbation samples with a sudden increase in residuals and a decrease in adsorption, and merge the foci-batch learning features, perturbation types and historical adsorption performance in the retained samples to generate perturbation learning features.

[0073] It should be noted that, based on the perturbation-labeled sample table, low-perturbation samples, competing perturbation samples, and high-fluctuation perturbation samples are retrieved sequentially according to batch number, pollutant polarity type, and adsorption operation stage. The polarity residual change, non-polarity residual change, and adsorption retention change corresponding to each type of sample are placed in the same screening position. Within the same screening position, samples with stable residual changes and continuous adsorption retention changes are identified as low-perturbation samples with continuous adsorption performance. Samples with a decrease in residue on one side and a controllable increase in residue on the other side with uninterrupted adsorption retention changes are identified as competing perturbation samples with continuous adsorption performance. Low-perturbation samples and competing perturbation samples are retained as samples with polar or non-polar residues. Samples with sudden increases in current or residual levels that do not show a correction after a decline and whose adsorption remains stable are identified as high-fluctuation disturbance samples and removed from the disturbance labeling sample table. Subsequently, based on the batch number of the retained samples, the corresponding active coke category label, batch stable state label, and batch attribute change information are matched from the coke batch learning features. The matched coke batch learning features are then merged with the disturbance type and historical adsorption performance in the retained samples and rearranged according to pollutant polarity type and adsorption operation stage. This allows the coke batch learning features, disturbance type, and historical adsorption performance to form a continuous learning record that can be used for subsequent hierarchical merging and machine learning proportion prediction, generating disturbance learning features.

[0074] S2.5. Perform hierarchical merging of the perturbation learning features to obtain samples in the same layer, and bind the ratio response labels in the samples in the same layer accordingly to generate a ratio learning sample table.

[0075] It should be noted that when stratifying and merging perturbation learning features, stratification conditions are first established according to activated coke category labeling, batch stable state labeling, perturbation type, pollutant polarity type, and adsorption operation stage. Learning records that meet the same stratification conditions in the perturbation learning features are grouped into the same stratum. Within the same stratum, the continuity of polar residual changes, non-polar residual changes, and adsorption retention changes is compared. Learning records with consistent change directions, similar fluctuation states, and continuous adsorption performance are grouped into samples of the same stratum. Subsequently, the polar residual response, non-polar residual response, residual fluctuation state, and adsorption retention change corresponding to the blending ratio are extracted from the samples of the same stratum, and a ratio response label is formed according to the position of the blending ratio. The ratio response label is then bound to the coke batch learning features, perturbation type, and historical adsorption performance in the samples of the same stratum, so that each group of samples of the same stratum corresponds to a clear blending ratio response result. Finally, the samples are arranged according to the corresponding order of stratification conditions, samples of the same stratum, and ratio response labels to generate a ratio learning sample table.

[0076] S2.6 Based on the proportional learning sample table, the perturbation type of the current coke batch is matched with the perturbation type of the historical samples, and the historical samples with opposite perturbation directions are removed. The position of the change of the compound ratio in the historical samples with the synchronous convergence of polar and non-polar residues is identified by machine learning, the proportional convergence center of the current coke batch is predicted, and the proportional neighborhood is expanded around the proportional convergence center to generate the set of ratios to be deduced.

[0077] It should be noted that, based on the proportional learning sample table, the disturbance type, polar residue change direction, non-polar residue change direction, and adsorption retention change of the current coke batch are first read. The disturbance type of the current coke batch is then compared item by item with the disturbance types of historical samples in the proportional learning sample table. If the disturbance type of the historical samples is consistent with the disturbance type of the current coke batch in terms of the direction of occupancy difference, residue change direction, and adsorption retention change, the historical sample is marked as a gated retention sample. If the disturbance type of the historical samples is contrary to the disturbance type of the current coke batch in terms of residue rise / fall direction or adsorption retention change, the historical sample is determined to be a historical sample with the opposite disturbance direction and is removed. The gated retention samples are arranged continuously according to the compounding ratio and processed by a machine. The instrument learns to read the learning features, perturbation types, historical adsorption performance, and ratio response labels of the coke batches in the gated retained samples. It identifies the positions where polar residues gradually approach the control requirements and non-polar residues synchronously approach the control requirements while the adsorption and retention changes do not decrease during the process of compound ratio changes. The compound ratio position with the most synchronous convergence is predicted as the ratio convergence center of the current coke batch. Adjacent compound ratios are expanded around the ratio convergence center in the direction of increasing polar active coke ratio and increasing non-polar active coke ratio, respectively. Compound ratios that exceed the total compound constraint, deviate from the direction of residue change, or have interrupted adsorption and retention changes are eliminated. Compound ratios that can form a continuous deduction relationship around the ratio convergence center are retained, generating a set of ratios to be deduced.

[0078] S2.7. Based on the set of ratios to be deduced, each compound ratio is backfilled into the polar adsorption response and non-polar adsorption response of the current batch of coke, and the corresponding polar residue, non-polar residue, residue fluctuation and adsorption retention change are predicted using the historical samples after gating and matching, and a ratio adsorption deduction table is generated.

[0079] It should be noted that, based on the set of formulations to be derived, each formulation ratio is retrieved sequentially according to its arrangement, and each formulation ratio is broken down into the proportion of partially polar activated coke and the proportion of partially non-polar activated coke. The proportion of partially polar activated coke is then filled into the partially polar adsorption response of the current coke batch to obtain the adsorption response position corresponding to polar pollutants, and the proportion of partially non-polar activated coke is filled into the non-polar adsorption response of the current coke batch to obtain the adsorption response position corresponding to non-polar pollutants. Based on the historical samples after gating and matching, and following the principles of consistent disturbance type, similar batch stable state, and corresponding adsorption operation stage, historical responses matching the adsorption response positions are searched. Record the changes in polar residue, non-polar residue, residue fluctuation, and adsorption retention in the historical response records as a prediction reference; simultaneously extrapolate the partial polar adsorption response and non-polar adsorption response for each compound ratio to predict the changes in polar residue and non-polar residue, and continue to compare the residual rise and fall and the adsorption retention change transition between adjacent compound ratios to obtain the residue fluctuation and adsorption retention change; finally, write each compound ratio, partial polar adsorption response, non-polar adsorption response, polar residue, non-polar residue, residue fluctuation, and adsorption retention change in the order of the set of ratios to be extrapolated to generate a ratio adsorption extrapolation table.

[0080] S2.8 Identify from the ratio adsorption deduction table the ratios of polar and non-polar residues that decrease simultaneously, have continuous convergence of residue fluctuations, and whose adsorption does not decrease, and arrange them in order of adsorption suitability from best to worst to generate candidate ratio sequences.

[0081] It should be noted that, according to the arrangement order of the set of ratios to be derived, the polar residue, non-polar residue, residue fluctuation, and adsorption retention changes corresponding to each compound ratio should be read item by item from the ratio adsorption derivation table, and each compound ratio should be placed in the same comparison position as adjacent compound ratios. In the same comparison position, the direction of change of polar residue and non-polar residue should be compared first. Compound ratios in which both polar residue and non-polar residue decrease or synchronously approach the control requirements should be marked as dual-residue synchronous decrease ratios. Then, the residue fluctuation corresponding to the dual-residue synchronous decrease ratios should be read. If the residue fluctuation decreases from large to small or remains stable between adjacent compound ratios, then the dual-residue ratios should be marked as dual-residue synchronous decrease ratios. The proportion of synchronous decrease is marked as the proportion of continuous convergence of residual fluctuations; then the adsorption retention change corresponding to the proportion of continuous convergence of residual fluctuations is read. If the adsorption retention change does not decrease or is interrupted, the proportion of continuous convergence of residual fluctuations is determined as the proportion of the blend that can be retained; the degree of polar residual control, the degree of non-polar residual control, the degree of residual fluctuation convergence, and the degree of adsorption retention stability of the proportion of the blend that can be retained are checked in tandem, and the adsorption fit performance is determined in the order of more sufficient dual residual control, more stable residual fluctuation, and more continuous adsorption retention change; finally, the proportions of the blend that can be retained are arranged from best to worst in terms of adsorption fit performance to generate a candidate blend ratio sequence.

[0082] S3. Based on the candidate ratio sequence, predict the fluctuations of polar and non-polar residues one by one, retain the ratios with the minimum residue and the smallest fluctuation, remove the ratios with sudden increases in fluctuation, and generate the ratio matching ranking.

[0083] like Figure 5 As shown, the polar and non-polar residue curves under varying proportions of polarized activated carbon demonstrate the influence of different compounding ratios on the two types of residues. Polar residues gradually decrease as the proportion of polarized activated carbon increases, while non-polar residues gradually increase after the proportion of non-polarized activated carbon becomes insufficient. The magnified area shows that the two curves tend to converge and form a balanced convergence zone around 55%–60%, which intuitively illustrates that the present invention screens out fluctuating ratios by sorting through dual residue fluctuations and determines a more stable target compounding ratio from the candidate ratio sequence.

[0084] S3.1. Read the adsorption adaptation performance corresponding to different compound ratios one by one according to the candidate ratio sequence, and input the different compound ratios into the ratio perturbation prediction rule to predict polar residues and non-polar residues, and generate a ratio residue prediction set.

[0085] It should be noted that different compound ratios are read one by one according to the order of the candidate compound ratio sequence, and the adsorption adaptation performance corresponding to different compound ratios is retrieved. The adsorption adaptation performance is used as the prediction reference for different compound ratios to enter the ratio perturbation prediction rule.

[0086] Different blending ratios are sequentially input into the proportional perturbation prediction rule. Based on the adsorption coupling relationship already formed in the proportional perturbation prediction rule, the residual responses of partially polar activated carbon and partially non-polar activated carbon under different blending ratios are deduced to obtain polar and non-polar residues respectively. The positions of polar and non-polar residues with different blending ratios are written into the corresponding positions, and the polar and non-polar residues after corresponding writing are collected in the order of the candidate blending ratio sequence to generate a blending residue prediction set.

[0087] S3.2. The ratio residual prediction set is continuously arranged to obtain the information of adjacent variables of the two residuals, and the difference between the adjacent variables of the two residuals is compared before and after to generate the two residual fluctuation sequence.

[0088] It should be noted that the predicted set of residual ratios is arranged continuously according to the different compound ratios in the candidate compound ratio sequence, and the polar and non-polar residuals corresponding to adjacent compound ratios are placed in the same adjacent variable comparison position to obtain the adjacent variable information of dual residuals.

[0089] When comparing the differences between adjacent compounding ratios with dual residues, the polar and non-polar residues corresponding to the previous compounding ratio are read first, and then the polar and non-polar residues corresponding to the next compounding ratio are read. The amount of residue change between the previous and next compounding ratios is calculated. Based on the amount of residue change, it is determined whether there is a synchronous increase, synchronous decrease, or one increase and one decrease relationship between adjacent compounding ratios. The relationship of change is continuously written to the corresponding compounding ratio position. Finally, the amount of residue change and the relationship of change between each adjacent compounding ratio are merged according to the order of the candidate compounding ratio sequence to generate a dual residue fluctuation sequence.

[0090] It should also be noted that dual-residue adjacent variation information refers to the residue change information formed by placing the polar and non-polar residues corresponding to adjacent compound ratios in the same comparison position according to the compound ratio arrangement order of the candidate compound ratio sequence. It is mainly used to reflect the rising and falling direction, change magnitude, and continuity state of the two types of residues between the previous and subsequent compound ratios. By obtaining dual-residue adjacent variation information, it is possible to determine whether there are changes such as synchronous decrease, one increase and one decrease, synchronous increase, or unilateral sudden increase between different compound ratios, providing a basis for subsequent generation of dual-residue fluctuation sequences, elimination of fluctuation and sudden increase ratios, and determination of stable compound ratios.

[0091] The expression for calculating the residual change in the compounding ratio before and after is:

[0092] ;

[0093] in, For the first The compound ratio and the first The vector of dual residual changes between each blending ratio; This represents the change in polarity residue. This represents the change in nonpolar residues; For the first The predicted polar residual value corresponding to each compound ratio; For the first The predicted polar residual value corresponding to each compound ratio; For the first The predicted nonpolar residual values ​​corresponding to each compound ratio; For the first The predicted nonpolar residual values ​​corresponding to each compound ratio; This represents the position of the compound ratio arrangement in the candidate matching sequence.

[0094] S3.3 Identify the position of sudden increase in nonpolar residue based on the double residue fluctuation sequence, where the subsequent compound ratio does not fall back to correct the fluctuation, and remove the compound ratio corresponding to the fluctuation surge position from the candidate ratio sequence to generate a stable candidate ratio set.

[0095] It should be noted that, based on the dual-residue fluctuation sequence, the adjacent change states of nonpolar residues are read one by one according to the arrangement order of the candidate ratio sequence, and the position where the nonpolar residue changes from a steady change to a rapid increase between adjacent compound ratios is identified as the fluctuation surge position.

[0096] Continue reading the continuous compound ratios after the fluctuation surge position. If the non-polar residue corresponding to the continuous compound ratio does not show a downward trend, and the improvement of polar residue cannot offset the increase of non-polar residue, then the fluctuation surge position is judged to be a non-correction position. Remove the compound ratio corresponding to the non-correction position from the candidate ratio sequence, and retain the compound ratio with stable non-polar residue changes or that can be corrected downward in subsequent compound ratios. Finally, reconnect the retained compound ratios according to the original arrangement order of the candidate ratio sequence to generate a stable candidate ratio set.

[0097] S3.4 Perform dual-residue fluctuation gating sorting on the stable candidate formulation set, calculate the formulation fit value, and sort the stable candidate formulation set from high to low according to the formulation fit value. Retain the compound ratios that meet the control requirements for both polar and non-polar residues and have the highest formulation fit value to generate the compliant stable formulation set.

[0098] It should be noted that, following the order of the compounding ratios in the stable candidate ratio set, the polar residue, non-polar residue, and dual residue fluctuation sequences corresponding to each compounding ratio are read sequentially, and these sequences are placed in the same gated verification position. Within the same gated verification position, the polar residue corresponding to the current compounding ratio is first compared with the polar residue control requirements, and then the non-polar residue corresponding to the current compounding ratio is compared with the non-polar residue control requirements. If the polar residue does not exceed the polar residue control requirements and the non-polar residue does not exceed the non-polar residue control requirements, then the current compounding ratio is determined to meet the dual residue control requirements; if any residue exceeds the corresponding control requirements, then the current compounding ratio is determined not to meet the dual residue control requirements. Taking the current compounding ratio as the center, compare the dual-residue fluctuation state corresponding to the previous compounding ratio with the current compounding ratio, and compare the dual-residue fluctuation state corresponding to the next compounding ratio. If the change in polar residue and the change in non-polar residue continuously increase between adjacent compounding ratios, it is determined that there is continuous amplification. If one side of the change in polar residue or the change in non-polar residue suddenly increases, while the other side does not form a synchronous compensating decrease, it is determined that there is a unilateral sudden increase. If the direction of residue change before and after the current compounding ratio changes from stable convergence to reverse divergence, or if the previous compounding ratio, the current compounding ratio, and the next compounding ratio cannot form a continuous decrease, a stable approach, or a return, it is determined that there is a continuous increase. If the relationship is corrected, it is determined that there is an interruption in the continuity between the preceding and following ratios. Only when the dual residue control requirements are met, and there is no continuous amplification, unilateral surge, or interruption in the continuity between the preceding and following ratios, will the current compound ratio enter the gated retention range. When both polar and non-polar residues meet the control requirements, and the dual residue fluctuation sequence maintains a stable continuity between the preceding and following compound ratios, the corresponding compound ratio will be marked as the gated retention ratio. When polar or non-polar residues do not meet the control requirements, or when the dual residue fluctuation sequence shows continuous amplification, unilateral surge, or interruption in the continuity between the preceding and following ratios, the corresponding compound ratio will be marked as the gated rejection ratio, thereby completing the gated screening of the stable candidate ratio set.

[0099] When calculating the fit value for the gated retention ratio, the degree of residue control is first determined based on the proximity of polar and non-polar residues to the control requirements. The stability of fluctuations is determined based on the magnitude and direction of change of adjacent blending ratios in the dual-residue fluctuation sequence. The degree of disturbance deviation is determined based on the disturbance deviation change of the disturbance learning characteristics under the corresponding blending ratio. The degree of residue control, the stability of fluctuations, and the degree of disturbance deviation are converted to the same evaluation scale so that the blending ratio with more sufficient residue control, more stable dual-residue fluctuations, and smaller disturbance deviations achieves a higher fit value. Finally, the gated retention ratios are rearranged from high to low according to the fit value, and the blending ratios with the highest fit values ​​that meet the control requirements for both polar and non-polar residues are retained to generate a set of compliant and stable blending ratios.

[0100] S3.5. Perform adjacent order verification on the stable ratio set that meets the standard, move the ratio of the compound with decreasing ratio fit value and increased double residue fluctuation to the back, and move the ratio of the compound with stable ratio fit value and small double residue fluctuation to the front, to generate the ratio fit ranking.

[0101] It should be noted that adjacent compound ratios are read one by one according to the current arrangement order of the compliant stable ratio set, and the ratio adaptation value and dual residual fluctuation state corresponding to the adjacent compound ratios are retrieved.

[0102] The preceding and following compound ratios are compared adjacently. If the matching value of the preceding compound ratio is lower than that of the following compound ratio and the fluctuation of the two residues is higher than that of the following compound ratio, the preceding compound ratio is shifted to the next position. If the matching value of the following compound ratio remains stable and the fluctuation of the two residues is lower than that of the preceding compound ratio, the following compound ratio is shifted to the next position. The adjacent order of the stable compound ratios after shifting is checked again until a sorting relationship is formed between the adjacent compound ratios with priority given to the matching value and stable fluctuation of the two residues, thus generating the matching ranking.

[0103] S4. Based on the ratio matching sorting, read the compound ratios that are ranked first and where polar and non-polar residues converge evenly, and determine the layered filling structure in combination with the polarity distribution of pollutants to generate a compound implementation plan.

[0104] like Figure 6 As shown, by comparing the penetration times of polar and non-polar pollutants under Scheme A (empirical compounding method), Scheme B (ordinary machine learning compounding method), and Scheme C (the coke compounding optimization scheme of the present invention), the advantages of the coke compounding optimization scheme of the present invention in delaying the penetration of two types of pollutants are demonstrated. Among them, the penetration times of polar and non-polar pollutants in Scheme C are higher than those in other schemes, indicating that the layered filling structure can form a continuous adsorption and support relationship between the polarity enhancement layer, the balanced mixing layer, and the non-polar retention layer, and the adsorption operation feedback can further correct the compounding ratio and compounding method, thereby improving the long-term operational stability.

[0105] Option A, the experience-based compounding method, refers to determining the compounding relationship between polar activated carbon and non-polar activated carbon based on existing engineering experience, historical addition ratios, or routine test results. The compounding is usually completed in a fixed ratio, without continuously predicting the synchronous changes in polar and non-polar residues.

[0106] Option B, the conventional machine learning compounding method, involves inputting the batch attributes of activated carbon, pollutant composition, and historical adsorption performance into a conventional machine learning algorithm to predict the overall adsorption performance under different compounding ratios and select the optimal compounding ratio accordingly. However, it usually focuses on the overall removal effect and does not further combine the dual residual fluctuations, layered filling structure, and operational feedback for linkage correction.

[0107] S4.1. Based on the matching ratio, read the top-ranked stable compound ratios that meet the standards, and extract the compound ratios from the stable compound ratios where both polar residual changes and non-polar residual changes converge continuously, generating balanced convergent ratios.

[0108] It should be noted that, according to the matching ratio, the top-ranked compound ratios are read from front to back, and the top-ranked compound ratios are taken as the compliant and stable compound ratios; residual convergence elements are extracted from the dual residual fluctuation sequence corresponding to the compliant and stable compound ratios, and the continuous change relationship of the residual convergence elements is checked according to the arrangement order of the compliant and stable compound ratios.

[0109] If the residual convergence factor corresponding to the compliant stable compound ratio remains synchronously declining or stably close to the control requirements between adjacent positions, and there is no continuous rebound of unilateral residue, then it is determined that the compliant stable compound ratio meets the requirement that both polar residue changes and non-polar residue changes are continuously converged; the compliant stable compound ratios that meet the continuous convergence requirements are retained and aggregated according to their order of position in the ratio matching ranking to generate balanced converged ratios.

[0110] It should also be noted that the change in polarity residue refers to the rise and fall of the portion of polar pollutants that have not been adsorbed and removed under different compounding ratios or continuous adsorption operation stages, which is used to reflect the control effect of the polarized activated coke on polar pollutants.

[0111] Non-polar residue variation refers to the rise and fall of the portion of non-polar pollutants that have not been adsorbed and removed, reflecting the control effect of partially non-polar activated carbon on non-polar pollutants. Continuous convergence means that both polar and non-polar residues in adjacent blending ratios gradually decrease and synchronously approach the control requirements, or that residues on one side fluctuate slightly but can be corrected by falling back in subsequent blending ratios, without any continuous increase in residues on one side, synchronous rebound of both residues, or interruption of the direction of change.

[0112] S4.2. Based on the balanced convergence ratio, read the proportion of partially polar active coke and partially non-polar active coke, and combine the dual-residue fluctuation sequence corresponding to the balanced convergence ratio to confirm the stable segment of dual-residue change, and generate the target compound ratio.

[0113] It should be noted that, based on the balanced convergence ratio, the proportions of partially polar active slag and partially non-polar active slag are read item by item, and the proportions of partially polar active slag and partially non-polar active slag are correlated with the position of the balanced convergence ratio in the ratio matching order; the residual stable elements are read by combining the dual residual fluctuation sequence corresponding to the balanced convergence ratio, and the continuous change state of the residual stable elements is checked along the arrangement order of the balanced convergence ratio.

[0114] If the residual stabilizing elements remain stable between adjacent equilibrium convergence ratios, and changes in the proportions of polar and non-polar activated coke do not cause a rebound in residues, then the corresponding position is confirmed to belong to the stable segment of dual residue changes. From the stable segment of dual residue changes, select the equilibrium convergence ratio with the highest matching order and the smallest fluctuation in residual stabilizing elements, and determine the selected proportions of polar and non-polar activated coke as the target blending ratio.

[0115] S4.3. Read the polar migration characteristics in the target wastewater detection data according to the target compounding ratio, and analyze the distribution of the polar migration characteristics according to the strength of pollutant polarity and the order of influent migration to generate the pollutant polarity distribution.

[0116] It should be noted that the target wastewater detection data is read according to the target compounding ratio, and polar migration characteristics that reflect the changes of pollutants along the influent direction are extracted from the target wastewater detection data; the polar migration characteristics are arranged chronologically according to the influent migration order to form a polar migration sequence; when performing distribution analysis on the polar migration sequence, the polar migration characteristics are first divided into polarity levels according to the strength of pollutant polarity, and then the migration connection relationship between different polarity levels in the preceding and following positions is checked along the influent migration order.

[0117] If the same polarity level remains concentrated in consecutive influent locations, it is determined that a polarity concentration zone is formed at the corresponding influent location. If different polarity levels transition continuously in adjacent influent locations, it is determined that a polarity transition zone is formed at the corresponding influent location. The polarity concentration zone and the polarity transition zone are merged according to the influent migration sequence to generate a pollutant polarity distribution.

[0118] It should also be noted that polarity migration characteristics refer to the distribution changes of different polar pollutants with varying positions, times, and adsorption stages in the target contaminated medium along the influent direction or flow path. This mainly reflects the enrichment, transition, and penetration trends of polar and non-polar pollutants in the initial, middle, and final stages. By extracting polarity migration characteristics, the migration and transfer relationships of pollutant polarity at different filling locations can be determined, providing a basis for subsequent generation of pollutant polarity distribution, determination of polarity-enhanced layers, balanced mixing layers, and non-polar retention layers.

[0119] S4.4 Perform polar gradient stratified filling analysis on the polarity distribution of pollutants to determine the filling tendency of predominantly polar activated carbon and predominantly non-polar activated carbon in different filling layers, and generate stratified filling tendency information.

[0120] It should be noted that the polarity concentration zone and polarity transition zone in the polarity distribution of pollutants should be read according to the influent migration sequence, and the polarity concentration zone and polarity transition zone should be converted into the basis for dividing the filling layer; the polarity receiving position of different filling layers should be determined according to the filling layer dividing basis, and the connection between the polarity receiving positions of different filling layers should be checked.

[0121] If the polar receiving position corresponding to the filling layer is mainly a polar concentration area, then the filling tendency of the partially polar activated carbon is determined to be higher. If the polar receiving position corresponding to the filling layer is mainly a non-polar enrichment change, then the filling tendency of the partially non-polar activated carbon is determined to be higher. If the polar receiving position corresponding to the filling layer is in a polar transition zone, then the balanced filling tendency of partially polar activated carbon and partially non-polar activated carbon is determined. The filling tendencies corresponding to different filling layers are sorted according to the influent migration sequence to generate stratified filling tendency information.

[0122] S4.5 Based on the layered filling tendency information, the inlet section is configured as a polarity-enhanced layer, the filling position corresponding to the stable section of dual residual changes is configured as a balanced mixing layer, and the filling position with a high non-polar penetration tendency is configured as a non-polar interception layer, thus generating a layered filling structure.

[0123] It should be noted that, based on the stratified filling tendency information, the filling tendency corresponding to different filling layers is read according to the influent migration sequence, and the filling position with a higher tendency of polar activated carbon in the influent front section is configured as a polarized reinforcement layer; the filling position corresponding to the stable section of dual residue changes is read, and the filling position in the stable section of dual residue changes where polar activated carbon and non-polar activated carbon show a balanced filling tendency is configured as a balanced mixing layer.

[0124] Read the filling positions with a high non-polar penetration tendency from the layered filling tendency information, and configure the filling positions with a high non-polar activated coke filling tendency and located at the subsequent interception position as non-polar interception layers; connect the layers according to the water inlet receiving sequence of the polar reinforcement layer, the balanced mixing layer and the non-polar interception layer, and check the filling tendency transition relationship between adjacent filling layers to form a continuous adsorption receiving relationship between the front and back filling layers, thus generating a layered filling structure.

[0125] S4.6. Based on the layered filling structure, read the layered configuration features corresponding to each filling layer, perform an overall check on the layered configuration features, obtain the compounding execution elements, determine the compounding execution elements of biased polar active coke and biased non-polar active coke, and perform hierarchical arrangement and parameter adjustment of the compounding execution elements to generate a compounding implementation plan.

[0126] It should be noted that the layer configuration characteristics should be read sequentially according to the water intake sequence of the polarity enhancement layer, the balanced mixing layer, and the non-polar interception layer. The filling tendency, compound ratio, intralayer distribution mode, and interlayer transition requirements in the layer configuration characteristics should be checked item by item against the target compound ratio and layer filling structure.

[0127] When the polarized activated carbon configuration in the polarized enhancement layer corresponds to the polar pollutant interception requirements in the upstream section of the influent, the polarized and non-polarized activated carbon configuration in the balanced mixing layer corresponds to the stable section of dual residual changes, and the non-polarized activated carbon configuration in the non-polar interception layer corresponds to the downstream non-polar penetration control requirements, the corresponding stratified configuration characteristics are retained. When the stratified configuration characteristics of any filling layer are inconsistent with the target compounding ratio or stratified filling structure, the compounding ratio and intra-layer distribution of the corresponding filling layer are adjusted according to the connection relationship between adjacent filling layers. Subsequently, according to the polarized enhancement... The order of addition, intralayer laying method, and interlayer connection method of each filling layer are sorted out according to the positions of the layer, the balanced mixing layer, and the non-polar interception layer. The compounding execution elements are obtained, and the compounding layering position, compounding ratio distribution, and interlayer transition requirements of the polar activated carbon and the non-polar activated carbon in each filling layer are determined according to the compounding execution elements of the layered filling structure. Finally, the compounding execution elements are arranged hierarchically and the parameters are adjusted according to the water inlet receiving sequence of the layered filling structure, so that the compounding execution elements of each filling layer are consistent with the target compounding ratio and the layered filling structure, and a compounding implementation plan is generated.

[0128] S5. Perform the blending of polar activated coke and non-polar activated coke according to the blending implementation plan, collect adsorption operation feedback, obtain feedback correction characteristics, correct the direction of blending ratio prediction based on feedback correction characteristics, determine the corrected blending ratio and blending method, and generate coke blending optimization plan.

[0129] S5.1. Read the compounding execution elements according to the compounding implementation plan, and determine the compounding layering treatment of biased polar activated carbon and biased non-polar activated carbon according to the compounding execution elements to generate the actual compounding layer group.

[0130] It should be noted that, according to the compounding implementation plan, the compounding execution elements are read in the order of water intake of the layered filling structure, and the compounding execution elements are matched with the polarized strengthening layer, the balanced mixing layer, and the non-polarized interception layer layer by layer; the layering position, compounding order and transition requirements of the polarized activated carbon and the non-polarized activated carbon in each filling layer are determined according to the compounding execution elements, forming the basis for compounding layering.

[0131] Based on the compounding and layering criteria, the polarized activated carbon and the non-polarized activated carbon are laid layer by layer and the interlayer connection is processed. After each filling layer is completed, the correspondence between the filling layer and the compounding execution elements is checked. The filling layers that have been checked and found to be consistent are connected continuously according to the water inlet receiving sequence to generate the actual compounding layer group.

[0132] S5.2. Monitor the actual compound layer group, collect the adsorption operation status at the outlet of each filling layer along the water inlet direction, obtain adsorption operation feedback, verify the corresponding relationship between the adsorption operation feedback and the compound implementation plan, and extract adsorption offset characteristics.

[0133] It should be noted that the adsorption operation status at the outlet of each filling layer is read layer by layer according to the actual water inlet receiving sequence of the compound layer group, and the adsorption operation status is recorded in correspondence with the layer position of the corresponding filling layer to obtain adsorption operation feedback; when verifying the correspondence between the adsorption operation feedback and the compound implementation plan, the compound execution elements in the compound implementation plan are read, and the adsorption operation status at the outlet of each filling layer is compared layer by layer with the layer adsorption performance required by the compound execution elements to determine whether the actual compound layer group maintains the adsorption receiving relationship required by the compound implementation plan during operation.

[0134] If the adsorption operation feedback shows a deviating level of adsorption performance from the requirements of the compounding execution elements, then the adsorption offset elements are extracted according to the location of the filling layer where the deviation occurs and the deviation continuation process. The adsorption offset elements are then merged with the corresponding filling layer location in the adsorption operation feedback to generate adsorption offset features.

[0135] It should also be noted that adsorption shift characteristics refer to the characteristic information formed when the adsorption operation state at the outlet of each filling layer deviates from the actual compounding scheme during operation. This mainly reflects higher polar residue, higher non-polar residue, earlier breakthrough, decreased adsorption retention, and the continuation of the shift between different filling layers. By extracting adsorption shift characteristics, it is possible to determine whether the adsorption anomaly mainly originates from the polar enhancement layer, the balanced mixing layer, or the non-polar retention layer, providing a basis for subsequent generation of feedback correction characteristics, determination of proportional correction direction information, and correction of the compounding ratio.

[0136] S5.3 Perform time-series merging of adsorption offset features, merge residual offsets and early penetration changes with consistent direction within continuous running segments into the same correction segment, eliminate short-term isolated offsets, and generate feedback correction features.

[0137] It should be noted that the adsorption offset features are arranged continuously according to the collection time of the adsorption operation feedback, and the adsorption offset features in adjacent operation segments are checked and verified.

[0138] If the adsorption shift characteristics in adjacent running segments continue in the same direction, and the corresponding filling layer positions are connected in the actual compound layer group, then the adsorption shift characteristics are classified into the same correction segment. If the adsorption shift characteristics only appear in a single running segment, and there is no continuation in the same direction in the preceding and following running segments, then the adsorption shift characteristics are judged to be short-term isolated shifts and are removed. Finally, the remaining correction segments are sorted according to time sequence and filling layer position, so that the residual shifts and penetration changes with consistent direction in consecutive running segments form a feedback basis that can be used for subsequent correction, and feedback correction characteristics are generated.

[0139] S5.4 Perform feedback offset correction on the feedback correction feature, determine the correction direction of the proportion of polarized active focal groups and the proportion of non-polarized active focal groups, and generate proportional correction direction information.

[0140] It should be noted that the feedback correction features are read sequentially according to the time sequence of adsorption operation feedback and the actual filling layer position of the compound layer group. The residual offset direction, residual offset duration, and penetration advance position in the feedback correction features are written into the same offset correction position. The offset correction position is then correlated with the proportion of polar activated carbon and the proportion of non-polar activated carbon in the target compound ratio. The distribution of the offset correction position in the polar enhancement layer, the balanced mixing layer, and the non-polar interception layer is checked along the water intake sequence. If the residual offset direction is consistent with the main adsorbent of the corresponding filling layer, and the residual offset duration continues in the same direction in adjacent operation segments, then the corresponding filling layer is marked as the offset-dominant layer.

[0141] If the penetration advance position extends from the previous filling layer to the next filling layer, the offset starting point is determined according to the starting layer of the penetration advance position, and the offset diffusion direction is determined according to the extension layer of the penetration advance position. The offset source is determined according to the offset dominant layer, the offset starting point, and the offset diffusion direction. When the offset source is concentrated in the polarity-enhanced layer and the corresponding polarity residue is continuously high, the polarity-active-foam ratio is determined as the upward adjustment direction. When the offset source is concentrated in the non-polarity-retaining layer and the corresponding non-polarity residue is continuously high, the non-polarity-active-foam ratio is determined as the upward adjustment direction. When the offset source is concentrated in the balanced mixing layer and the polarity residue and non-polarity residue alternately increase during the preceding and following running periods, the main correction direction is determined according to the side with the larger residue offset, and the ratio of the other side is corrected accordingly. Finally, the correction directions of the polarity-active-foam ratio and the non-polarity-active-foam ratio, the corresponding offset dominant layer, and the offset diffusion direction are adjusted to generate proportional correction direction information.

[0142] S5.5. Based on the proportional correction direction information, the target compound ratio is corrected in direction, and the filling amount in the biased reinforcement layer, the balanced mixing layer and the non-polar retention layer is adjusted in conjunction to generate the corrected compound ratio.

[0143] It should be explained that, based on the proportional correction direction information, the correction direction of the proportions of polar activated carbon and non-polar activated carbon is read, and the correction direction is connected with the corresponding proportion in the target compounding ratio; the proportion of activated carbon on the side with insufficient adsorption is increased according to the correction direction, while the proportion of activated carbon on the side with redundant adsorption is simultaneously reduced, so that the proportions of polar activated carbon and non-polar activated carbon are kept consistent in the total compounding amount; according to the target compounding ratio after direction correction, the filling and receiving relationship in the polar enhancement layer, the balanced mixing layer and the non-polar interception layer are checked respectively, and the changes in filling amount are adjusted in a linkage manner along the influent receiving sequence; the consistency between the filling amount of each filling layer after linkage adjustment and the target compounding ratio after direction correction is checked, and the corrected compounding ratio is generated.

[0144] S5.6. Based on the modified compounding ratio, check the interlayer connection relationship in the layered filling structure, and adjust the interlayer transition method and the requirement for uniform mixing to generate a modified compounding method.

[0145] It should be noted that the adjusted relationship between the proportion of biased active coke and the proportion of biased non-polar active coke is read based on the modified compounding ratio, and the adjusted relationship is mapped to the biased strengthening layer, the balanced mixing layer and the non-polar retention layer in the layered filling structure.

[0146] According to the water intake sequence, check the interlayer connection between adjacent filling layers to determine whether the filling tendency of adjacent filling layers can be continuously transitioned. If there is an abrupt change in filling tendency between adjacent filling layers, adjust the interlayer transition method to make the polar activated carbon and the non-polar activated carbon form a gradual connection between adjacent filling layers. If the filling distribution in the balanced mixing layer is inconsistent with the corrected compounding ratio, adjust the mixing uniformity requirements to make the distribution of polar activated carbon and the non-polar activated carbon in the balanced mixing layer consistent with the corrected compounding ratio. The adjusted interlayer transition method and mixing uniformity requirements are then integrated according to the layered filling structure to generate a corrected compounding method.

[0147] S5.7. Based on the modified compounding ratio and modified compounding method, the prediction direction of the compounding ratio in the ratio disturbance prediction rule is written back and corrected to generate the coking optimization scheme.

[0148] It should be noted that the proportion of biased active coke, the proportion of biased non-polar active coke, and the stratification adjustment relationship are read according to the corrected compounding ratio and the corrected compounding method, and the stratification adjustment relationship is checked against the compounding ratio prediction direction in the ratio disturbance prediction rule.

[0149] If the modified blending ratio changes direction relative to the target blending ratio, the direction of the blending ratio prediction is adjusted according to the adsorption feedback corresponding to the modified blending ratio, so that the ratio perturbation prediction rule subsequently points to the blending ratio with more stable residual control. If the modified blending method changes the interlayer transition relationship in the layered filling structure, the change in interlayer connection corresponding to the modified blending method is written into the ratio perturbation prediction rule, so that the ratio perturbation prediction rule simultaneously retains the modified correlation between the blending ratio and the blending method. The modified ratio perturbation prediction rule, the modified blending ratio, and the modified blending method are then integrated to generate the coke blending optimization scheme.

[0150] In summary, this invention constructs a perturbation learning foundation by learning the characteristics of activated carbon batches and historical adsorption performance, and uses machine learning to predict the compounding ratio, screens stable ratios with balanced convergence of dual residues, and forms a layered filling structure by combining the polarity distribution of pollutants. Based on the adsorption operation feedback, the ratio direction and compounding method are corrected, thus completing the transformation of activated carbon compounding from experience-based selection to prediction, implementation, and feedback closed-loop optimization, improving the compatibility of compounding, the stability of residue control, and the continuous adsorption effect.

[0151] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing the blending ratio of active coke based on machine learning, characterized in that, include: Batch attribute information of partially polar activated coke and partially non-polar activated coke is collected, and the batch attribute information is normalized and sample labeled to generate coke batch learning features. By labeling the learning features of the coke batch with the historical adsorption performance, perturbation learning features are obtained. Machine learning is used to predict the proportion of perturbation learning features, and the adsorption adaptation performance of different compound ratios is deduced to generate candidate ratio sequences. Based on the candidate ratio sequence, predict the fluctuation of polar and non-polar residues one by one, retain the ratio with the minimum fluctuation and the minimum residue, remove the ratio with a sudden increase in fluctuation, and generate the ratio matching ranking. Based on the ratio matching sorting, read the compound ratio that ranks first and has balanced convergence of polar and non-polar residues, and determine the layered filling structure in combination with the polarity distribution of pollutants to generate a compound implementation plan. According to the compounding implementation plan, the partial polar activated carbon and partial non-polar activated carbon are compounded, and the adsorption operation feedback is collected to obtain the feedback correction characteristics. Based on the feedback correction characteristics, the predicted direction of the compounding ratio is corrected, the corrected compounding ratio and compounding method are determined, and the optimized carbon compounding scheme is generated.

2. The method for optimizing the blending ratio of active coke based on machine learning as described in claim 1, characterized in that, The specific steps for generating the focus batch learning features are as follows: Field checks and anomaly removal are performed on batch attribute information and partially nonpolar active focal groups to obtain valid batch information. The valid batch information is then categorized and scaled to generate normalized batch features. Samples are labeled with the activated coke category, historical adsorption performance, and batch stability of the normalized batch characteristics to generate coke batch learning features.

3. The method for optimizing the blending ratio of active coke based on machine learning as described in claim 2, characterized in that, The specific steps for mapping the learned features of the focal batch to historical adsorption performance and obtaining perturbation learning features are as follows: Based on the active coke category label and batch stability status label in the coke batch learning features, the historical adsorption performance under the same adsorption conditions is linked to the coke batch learning features item by item according to batch number, pollutant polarity type and adsorption operation stage to generate a coke batch adsorption correspondence table. The corresponding changes of polar activated coke and non-polar activated coke in adsorption perturbation index were extracted from the coke batch adsorption correspondence table, and the occupancy difference and residual change of the two types of activated coke were coupled and labeled to generate competitive residual coupling labels. Perturbation-graded samples are identified from competing residual coupling markers and written back to the char batch adsorption correspondence table to generate a perturbation-labeled sample table. Based on the perturbation-labeled sample table, low-perturbation samples with continuous adsorption performance and competing perturbation samples are retained, while high-fluctuation perturbation samples with sudden increases in residuals and decreased adsorption are removed. The foci-batch learning features, perturbation types and historical adsorption performance in the retained samples are merged to generate perturbation learning features.

4. The method for optimizing the blending ratio of active coke based on machine learning as described in claim 1, characterized in that, The specific steps for performing machine learning ratio prediction on the perturbation learning features, inferring the adsorption adaptation performance of different compound ratios, and generating candidate ratio sequences are as follows: The perturbation learning features are hierarchically merged to obtain samples of the same layer, and the ratio response labels in the samples of the same layer are bound to generate a ratio learning sample table. Based on the proportional learning sample table, the perturbation type of the current coke batch is matched with the perturbation type of the historical samples, the historical samples with opposite perturbation directions are removed, and the position of the change of the compound ratio in the historical samples with the synchronous convergence of polar and non-polar residues is identified by machine learning. The proportional convergence center of the current coke batch is predicted, and the proportional neighborhood is expanded around the proportional convergence center to generate a set of ratios to be deduced. Based on the set of ratios to be deduced, each ratio is backfilled into the polar adsorption response and non-polar adsorption response of the current batch of coke, and the corresponding polar residue, non-polar residue, residue fluctuation and adsorption retention change are predicted using the historical samples after gating and matching, and a ratio adsorption deduction table is generated. The ratios of polar and non-polar residues that decrease simultaneously, whose residue fluctuations continuously converge and whose adsorption remains unchanged are identified from the ratio adsorption deduction table. These ratios are then arranged in order of adsorption compatibility from best to worst to generate candidate ratio sequences.

5. The method for optimizing the blending ratio of active coke based on machine learning as described in claim 4, characterized in that, The specific steps for predicting the fluctuations of polar and non-polar residues item by item based on the candidate ratio sequence, retaining the ratios that meet the residue standards and have the smallest fluctuations, and eliminating ratios with sudden increases in fluctuations are as follows: Based on the candidate ratio sequence, the adsorption adaptation performance corresponding to different compound ratios is read one by one, and the different compound ratios are input into the ratio perturbation prediction rule to predict polar and non-polar residues, and generate a ratio residue prediction set. The ratio residual prediction set is continuously arranged to obtain the information of adjacent variables of dual residuals, and the difference between the adjacent variables of dual residuals is compared before and after to generate a dual residual fluctuation sequence. The fluctuation surge position is identified by the dual-residue fluctuation sequence, where the non-polar residue suddenly increases and the subsequent compound ratio does not fall back to correct it. The compound ratio corresponding to the fluctuation surge position is removed from the candidate ratio sequence to generate a stable candidate ratio set.

6. The method for optimizing the blending ratio of active coke based on machine learning as described in claim 1, characterized in that, The specific steps for generating the ratio adaptation sort are as follows: The stable candidate formulation set is sorted by dual residual fluctuation gating, the formulation fit value is calculated, and the stable candidate formulation set is sorted from high to low according to the formulation fit value. The formulation ratios that meet the control requirements for both polar and non-polar residues and have the highest formulation fit value are retained to generate the compliant stable formulation set. The adjacent order of the stable ratio set that meets the standard is checked. The ratio of the compound with the decrease in the ratio fit value and the increase in the fluctuation of the two residues is moved to the back, and the ratio of the compound with the stable ratio fit value and the small fluctuation of the two residues is moved to the front, so as to generate the ratio fit ranking.

7. The method for optimizing the blending ratio of active coke based on machine learning as described in claim 6, characterized in that, The specific steps for determining the layered filling structure based on the ratio matching and sorting of the top-ranked compounding ratios with balanced convergence of polar and non-polar residues, combined with the polarity distribution of pollutants, are as follows: Based on the matching ratio, read the top-ranked stable compound ratios that meet the standards, and extract the compound ratios from the stable compound ratios where both polar residual changes and non-polar residual changes converge continuously, to generate balanced convergent ratios. Based on the balanced convergence ratio, the proportions of polar active coke and non-polar active coke are read, and the stable segment of dual residue change is confirmed by combining the dual residue fluctuation sequence corresponding to the balanced convergence ratio, and the target compound ratio is generated. The polarity migration characteristics in the target wastewater detection data are read according to the target compounding ratio, and the distribution of the polarity migration characteristics is analyzed according to the strength of pollutant polarity and the order of influent migration to generate the pollutant polarity distribution. Polar gradient stratified filling analysis was performed on the polarity distribution of pollutants to determine the filling tendency of predominantly polar activated carbon and predominantly non-polar activated carbon in different filling layers, and to generate stratified filling tendency information. Based on the information on the layered filling tendency, the inlet section is configured as a polarity-enhanced layer, the filling position corresponding to the stable section of dual residual changes is configured as a balanced mixing layer, and the filling position with a high non-polar penetration tendency is configured as a non-polar interception layer, thus generating a layered filling structure.

8. The method for optimizing the blending ratio of active coke based on machine learning as described in claim 1, characterized in that, The compounding implementation scheme is obtained by reading the layered configuration features corresponding to each filling layer according to the layered filling structure, performing an overall check on the layered configuration features, obtaining the compounding execution elements, determining the compounding execution elements of biased polar activated carbon and biased non-polar activated carbon, and performing hierarchical arrangement and parameter adjustment of the compounding execution elements.

9. The method for optimizing the blending ratio of active coke based on machine learning as described in claim 8, characterized in that, The process involves blending partially polar activated carbon and partially non-polar activated carbon according to the blending implementation scheme, collecting adsorption operation feedback, and obtaining feedback correction characteristics. The specific steps are as follows: According to the compounding implementation plan, the compounding execution elements are read, and the compounding layering treatment of biased polar activated carbon and biased non-polar activated carbon is determined according to the compounding execution elements to generate the actual compounding layer group. The actual compound layer group was monitored for operation. The adsorption operation status at the outlet of each filling layer was collected along the water inlet direction. Adsorption operation feedback was obtained. The adsorption operation feedback was checked against the compound implementation plan, and adsorption offset characteristics were extracted. The adsorption offset features are merged in time sequence. Residual offsets and early penetration changes with consistent direction within continuous running segments are merged into the same correction segment, short-term isolated offsets are eliminated, and feedback correction features are generated.

10. The method for optimizing the blending ratio of active coke based on machine learning as described in claim 9, characterized in that, The steps for correcting the predicted direction of the blending ratio based on feedback correction characteristics, determining the corrected blending ratio and blending method, and generating an optimized blending scheme are as follows: Feedback offset correction is performed on the feedback correction features to determine the correction direction of the proportion of polarized active focal groups and the proportion of non-polarized active focal groups, and proportional correction direction information is generated. Based on the proportional correction direction information, the target compound ratio is corrected in direction, and the filling amount in the biased reinforcement layer, the balanced mixing layer and the non-polar retention layer is adjusted in a coordinated manner to generate the corrected compound ratio. Based on the verification of the interlayer connection relationship in the layered filling structure by correcting the compounding ratio, the interlayer transition method and the requirement for uniform mixing are adjusted to generate a corrected compounding method; Based on the modified blending ratio and modified blending method, the prediction direction of the blending ratio in the ratio disturbance prediction rule is written back and corrected to generate the coking optimization scheme.

Images