Water plant coagulant adding method based on intelligent expert decision model

Abstract

The invention discloses a water plant coagulant adding method based on an intelligent expert decision model, which comprises the following steps: collecting coagulant adding historical data, and carrying out anomaly detection and missing value filling; screening out qualified samples of which the pond turbidity is close to a target value; screening high-efficiency dosing samples on the basis of a KAN-SFA model for a scene in which the dosing amount is frequently changed; aiming at a scene in which the dosage only has a limited number of grades, grouping according to the grades and screening efficient dosing samples based on fuzzy comprehensive evaluation; constructing an initial intelligent dosing decision expert database based on decision tree regression; an intelligent dosing decision expert database is expanded based on Gaussian process regression, and the method comprises an interpolation process and an extrapolation process. Under different raw water quality conditions, initial dosing amount is obtained based on the intelligent dosing decision expert database, and the dosing amount is continuously and dynamically adjusted according to a time-lagged output turbidity feedback result; keeping the structure of the decision tree unchanged, and re-estimating the optimal dosage by using new data; when new medicine adding historical data is accumulated to a certain scale, the intelligent medicine adding decision expert database is updated. According to the invention, the timeliness and reliability of coagulant addition are improved to the greatest extent, and the medicine consumption is obviously reduced.

Description

Technical Field

[0001] This invention belongs to the field of smart water technology, specifically relating to a method for adding coagulant to water plants based on an intelligent expert decision-making model. Background Technology

[0002] The dosage of coagulants in water treatment plants needs to be dynamically adjusted according to changes in raw water quality. However, raw water quality is complex and variable, making it difficult to accurately determine the appropriate dosage manually. There is a significant time lag between coagulant addition and effluent quality feedback, meaning that coagulant addition methods based solely on feedback regulation cannot respond promptly to changes in raw water quality. Therefore, a feedforward model is needed to determine the initial dosage based on raw water quality. Existing machine learning-based feedforward control models have two main drawbacks: first, while they have high accuracy for similar raw water quality parameters that have appeared in historical training data, they face difficulties in extrapolation; second, these models often cannot be locally updated using new data and require a large amount of data for global updates. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a method for adding coagulant to water plants based on an intelligent expert decision-making model, so as to overcome the shortcomings of the prior art.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0005] A method for adding coagulant to a water plant based on an intelligent expert decision-making model is provided, comprising the following steps:

[0006] S1: Collect historical data on coagulant addition and preprocess it, including data anomaly detection and missing value filling, as well as calculation of effluent lag time;

[0007] S2: Select samples that meet the dosing effect, i.e., samples whose turbidity at the outlet is close to the target value;

[0008] S3: For scenarios where the dosage changes frequently, high-efficiency dosing samples are selected based on the KAN-SFA model;

[0009] S4: For scenarios where there are only a few dosage levels, group the dosage levels and screen for efficient dosage samples based on fuzzy comprehensive evaluation;

[0010] S5: Construct an initial intelligent dosing decision expert database based on a decision tree regression model;

[0011] S6: Expanding the expert database for intelligent dosing decision-making based on Gaussian process regression model, including two processes: interpolation and extrapolation;

[0012] S7: Under different raw water quality conditions, the initial dosage is obtained based on the intelligent dosing decision expert database, and the dosage is continuously and dynamically adjusted according to the turbidity feedback results after a certain time lag.

[0013] S8: Keep the decision tree structure unchanged and re-estimate the optimal dosage using the new data;

[0014] S9: Once a certain amount of new historical data on dosing has been accumulated, update the expert database for intelligent dosing decisions.

[0015] As described in the water plant coagulant dosing method based on the intelligent expert decision-making model, in step S1, the effluent turbidity corresponding to the dosing time is calculated based on the weighted average of the effluent turbidity within twice the standard deviation of the effluent lag time, with the weights being the values ​​of the normal distribution probability density function.

[0016] As described in the water plant coagulant dosing method based on the intelligent expert decision-making model, the basic structure of the KAN-SFA model used for screening high-efficiency dosing samples in step S3 is as follows:

[0017] y=exp(KAN(x;β))·exp(ε),ε=v+u

[0018] Wherein, the independent variable x is the raw water quality condition, the dependent variable y is the chemical consumption per unit of dosing, KAN(·) is the KAN model, β is the parameter of KAN, and ε is other factors affecting the optimal dosing amount, which consists of two parts: the random error term v and the inefficiency term u. v follows a normal distribution, and u > 0 and follows a semi-normal distribution.

[0019] The water plant coagulant dosing method based on the intelligent expert decision-making model, wherein the efficient dosing sample screening method based on fuzzy comprehensive evaluation in step S4 includes the following steps:

[0020] S41: Determine the domain of factors for the evaluation object;

[0021] S42: Determine the evaluation level domain V, and the evaluation set is: V = {Excellent, Good, Average, Poor, Very Poor};

[0022] S43: Establish a single-factor fuzzy relation matrix to reflect the fuzzy relationship between each factor and the evaluation level;

[0023] S44: Establish the weights of each factor;

[0024] S45: Conduct fuzzy comprehensive evaluation to reflect the degree of membership of the evaluated object at each evaluation level;

[0025] S46: Calculate the overall score.

[0026] As described in the water plant coagulant dosing method based on the intelligent expert decision-making model, step S5, which utilizes decision tree regression to construct the initial intelligent dosing decision expert database, is carried out in three steps:

[0027] S51: Based on the selected qualified and efficient samples, with the dosage as the dependent variable and the raw water quality condition variable as the independent variable, a decision tree regression model with a small depth is constructed. The value of the leaf node is the optimal dosage. After the decision tree is constructed, the sample set corresponding to each node is cleared.

[0028] S52: Based on all samples, with the effluent turbidity as the dependent variable and the raw water quality condition variable as the independent variable, further split the decision tree obtained in step S51. After the decision tree split is completed, clear the sample set corresponding to each node.

[0029] S53: Reassign the high-efficiency samples to the decision tree in S52, calculate the optimal dosage value for each node. If some nodes are not assigned samples, use the optimal dosage of the upper-level node as the optimal dosage of this node, or use Gaussian process regression in step S6 for interpolation calculation.

[0030] As described in the water plant coagulant dosing method based on the intelligent expert decision-making model, the optimal dosage calculation steps combining feedforward control and feedback regulation in step S7 are as follows:

[0031] S71: Clear the dataset for the decision tree;

[0032] S72: Get the current time The raw water quality conditions are analyzed in real time, including anomaly detection and missing value imputation. After data cleaning, it is determined whether the data deviates significantly from historical samples. If it deviates significantly, it is considered that a suitable dosage cannot be given based on historical data, and manual dosing is switched to the next step. Otherwise, the raw water quality conditions are input into the decision tree model to obtain the leaf node to which the current raw water quality conditions belong. In order to avoid treating the "minor fluctuations" of raw water quality as "mutations", the cleaned data is smoothed to eliminate high-frequency noise, and a "mutation judgment condition" is set: only when the magnitude of the state change exceeds the threshold and the duration exceeds the set value is it judged as a mutation and a node switch is triggered.

[0033] S73: Calculate the inflow time corresponding to the current outflow based on the time delay. The current moment and the time of water intake Both are shifted backward by twice the standard deviation of the lag time distribution, and the results are obtained respectively. and Get Time The raw water quality conditions are input into the decision tree to obtain the leaf nodes to which the influent water quality conditions belong;

[0034] S74: Determine whether the leaf node corresponding to the water quality conditions in step S73 and the leaf node calculated in step S72 belong to the same entity. If so, record the water effluent time. and Then proceed to step S75; otherwise, the water outlet time will be... and the time of water intake Simultaneously move backward, while judging the water quality conditions at the time of water inflow; if it is within the upper limit of the movement length T max For the first time in the country, the same raw water quality conditions were found, and the corresponding water output time was recorded. and the time of water intake Proceed to step S75; otherwise, use the value of the leaf node to which the current water quality condition belongs in the decision tree as the current dosage and proceed to step S76.

[0035] S75: Obtain the water ingress time The dosing consumption per unit, x, is used as the starting point for adjusting the dosing dosage at the current moment, utilizing the effluent time. The weighted average turbidity NTU of the effluent is calculated from the turbidity samples before and after the effluent, and the current dosage is determined based on the NTU classification.

[0036] S76: Determine if the current dosage is consistent with the previous dosage. If consistent, maintain the current dosage. If inconsistent, the system recommends the current dosage for manual confirmation. Upon confirmation, if the intelligent dosage is accepted, it is executed; otherwise, manual dosing is used.

[0037] The effect of dosing is evaluated. If the turbidity at the outlet of the tank differs too much from the NTU in step S75, it is considered that an unobserved abnormality has occurred in the system. An alarm is triggered and intelligent dosing is stopped, and manual dosing is switched to manual dosing.

[0038] As described in the water plant coagulant dosing method based on the intelligent expert decision-making model, step S8, which involves updating the optimal dosing dosage in the expert database while maintaining the unchanged decision tree structure, is as follows:

[0039] S81: If the water outlet time If the NTU calculated in step S75 is within the normal range, then the corresponding raw water quality conditions and dosage data will be added to the decision tree sample.

[0040] S82: Calculate the optimal dosage using stochastic frontier analysis;

[0041] The intelligent dosing decision expert database in step S8 is updated online in a localized manner.

[0042] As described in the water plant coagulant dosing method based on the intelligent expert decision-making model, the update of the intelligent dosing decision expert database in step S9 will go through steps S1-S6 again. The intelligent dosing decision expert database in step S9 is updated offline globally.

[0043] The beneficial effects of the technical solution of this invention are:

[0044] This invention is an intelligent coagulant dosing model that can adapt to various raw water quality conditions, including those not seen in the training samples, and can be updated in real time and dynamically adjusted based on the quality of the effluent. This model maximizes the timeliness and reliability of coagulant dosing and significantly saves on chemical consumption. Attached Figure Description

[0045] To further illustrate the above-mentioned objectives, structural features, and effects of the present invention, the present invention will be described in detail below with reference to the accompanying drawings.

[0046] Figure 1 This is a schematic diagram illustrating the calculation of the effluent lag between the coagulation zone and the sedimentation tank in a preferred embodiment of the present invention.

[0047] Figure 2 This is a schematic diagram illustrating the calculation of effluent turbidity corresponding to a preferred embodiment of the present invention and the dosage.

[0048] Figure 3 This is a flowchart illustrating the intelligent dosing process combining feedforward control and feedback regulation, representing a preferred embodiment of the present invention.

[0049] Figure 4 This is a schematic diagram illustrating the water inflow time corresponding to the current water outflow in a preferred embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram illustrating the influent and effluent times corresponding to the same raw water quality conditions in a preferred embodiment of the present invention.

[0051] Figure 6 This is a schematic diagram showing the direction and step size of chemical dosage adjustment in different effluent turbidity ranges according to a preferred embodiment of the present invention. Detailed Implementation

[0052] The terms “invention” and “the present invention” used in this specification are intended to broadly refer to all subject matter of this specification and any of the following patent claims. Statements containing these terms should not be construed as limiting the subject matter described herein or limiting the meaning or scope of any of the following patent claims. Furthermore, this specification does not attempt to describe or limit the subject matter covered by any claim of any particular component, paragraph, statement, or drawing of this application. The subject matter should be understood with reference to the entire specification, all drawings, and any of the following claims. The invention may have other embodiments and be practiced or implemented in other ways. Moreover, it should be understood that the wording and terminology used herein are for illustrative purposes and should not be considered limiting.

[0053] The details of the invention will now be discussed with reference to the accompanying drawings, which are illustrated by way of example only. In the drawings, similar features or components may be labeled using the same reference numerals.

[0054] The use of the terms "comprising," "having," and "including," and variations thereof, herein means to include the items listed herein and their equivalents and additional items. While reference may be made in the description of the drawings to directions such as above, below, upward, downward, backward, bottom, top, front, rear, etc., for convenience, reference is made relative to the drawings. These directions are not intended to literally accept or limit the invention in any form. Furthermore, terms such as "first," "second," "third," etc., are used herein for illustrative purposes and are not intended to indicate or imply importance or significance.

[0055] The coagulant dosing model algorithm based on the expert decision-making model of this invention includes the following steps:

[0056] Step S1: Collect historical data on coagulant addition and perform preprocessing, including data anomaly detection and missing value filling, calculation of effluent lag time, and calculation of effluent turbidity at the lag time corresponding to the dosing time.

[0057] Step S2: Select samples that meet the dosing effect, i.e., samples whose turbidity at the outlet is close to the target value.

[0058] Step S3: For scenarios where the dosage changes frequently, high-efficiency dosing samples are screened based on the KAN-SFA model developed in this invention.

[0059] Step S4: For scenarios where there are only a few dosage levels, group the dosage levels and screen for efficient dosage samples based on fuzzy comprehensive evaluation.

[0060] Step S5: Construct an initial intelligent dosing decision expert database based on decision tree regression.

[0061] Step S6: Expand the expert database for intelligent dosing decision-making based on Gaussian process regression, including two processes: interpolation and extrapolation.

[0062] Step S7: In practical applications, under different raw water quality conditions, the initial dosage is obtained based on the intelligent dosing decision expert database, and the dosage is continuously and dynamically adjusted according to the feedback results of the turbidity at the outlet of the pool after a certain time lag.

[0063] Step S8: At the same time, while keeping the structure of the decision tree unchanged, re-estimate the optimal dosage using the new data.

[0064] Step S9: Once the new historical data on dosing has accumulated to a certain scale, update the intelligent dosing decision expert database.

[0065] The calculation of the effluent time delay in step S1 is shown below. Figure 1 As shown, the change in influent volume per unit time Equal to inflow rate:

[0066]

[0067] If the volume V at two adjacent measurement time points T and T+1 T and V T+1 The following two inequalities must be satisfied respectively:

[0068] and

[0069] The lag time is

[0070]

[0071] Considering the existence of measurement error, the effluent lag time should be a random variable, assumed to follow a normal distribution, with the mean calculated according to the above formula and the variance determined empirically.

[0072] See Figure 2 In step S1, the turbidity of the effluent at the lag time corresponding to the dosing time is calculated based on the weighted average of the effluent turbidity within twice the standard deviation of the effluent lag time, with the weights being the values ​​of the normal distribution probability density function.

[0073] In step S3, the basic structure of the KAN-SFA model used for screening high-efficiency drug dosing samples is as follows:

[0074] y=exp(KAN(x;β))·exp(ε), ε=v+u

[0075] Wherein, the independent variable X represents the raw water quality conditions such as influent flow rate, raw water turbidity, raw water temperature, and raw water pH; the dependent variable y represents the chemical consumption per unit dosage; KAN(·) represents the KAN model; and β represents the parameters of the KAN model. ε represents other factors affecting the optimal dosage, consisting of a random error term v and an inefficiency term u > 0. v follows a normal distribution, u follows a semi-normal distribution, and exp(-u) is a measure of the dosing efficiency. Taking the logarithm of both sides of the KAN-SFA model yields:

[0076] log(y) = KAN(x; β) + v + u

[0077] This invention uses the maximum likelihood estimation method to estimate the parameters in the KAN-SFA model, and further calculates the efficiency of each sample, selecting the top 25% of samples by efficiency value as the basis for constructing the expert database.

[0078] Step S4, the efficient dosing sample screening method based on fuzzy comprehensive evaluation, includes the following steps:

[0079] Step S41: Determine the domain of factors affecting the evaluation object. Identify the various factors influencing dosing efficiency; these factors constitute the set of evaluation indicators.

[0080] U = {Influent flow rate, raw water turbidity, raw water temperature, raw water pH, ..., effluent turbidity}.

[0081] Step S42: Determine the evaluation level domain V. Set the level or category of the evaluation result. Here, it is divided into 5 levels, and the evaluation set is as follows:

[0082] V = {Excellent, Good, Average, Poor, Very Poor}

[0083] Among them, the highest evaluation level is "excellent" and the lowest evaluation level is "very poor". The unfavorable raw water quality conditions have a positive effect on the dosing efficiency. The more unfavorable the raw water quality conditions, the higher the evaluation level. The turbidity at the outlet has a negative effect on the dosing efficiency. The greater the turbidity at the outlet, the lower the evaluation level.

[0084] Step S43: Establish a single-factor fuzzy relation matrix. For each evaluation factor, determine its membership degree of the corresponding evaluation level, thereby constructing a single-factor fuzzy relation matrix from [value] to [value]. This matrix reflects the fuzzy relationship between each factor and the evaluation level.

[0085] Step S44: Establish the weights of each factor. Based on the importance of each factor in the evaluation system, the entropy weight method is used to determine the weights.

[0086] Step S45: Perform fuzzy comprehensive evaluation. Using the synthesis operation of fuzzy matrices, the single-factor fuzzy relation matrix R and the weight set A are synthesized to obtain the comprehensive evaluation result vector B. This vector reflects the membership degree of the evaluated object at each evaluation level.

[0087] Step S46: Calculate the overall score. Set the rating values ​​for each level in the evaluation set V to form...

[0088] The scoring set G = (100, 80, 60, 40, 20) is used, and the inner product of the fuzzy comprehensive evaluation result vector B and the scoring set G is used to obtain the comprehensive score.

[0089] In step S5, the present invention utilizes decision tree regression to construct an initial intelligent dosing decision expert database, which is carried out in three steps:

[0090] Step S51: Based on the selected qualified and efficient samples, construct a shallow decision tree regression model with the dosage as the dependent variable and the raw water quality condition variable as the independent variable. The value of the leaf node is the optimal dosage. After the decision tree is constructed, clear the sample set corresponding to each node.

[0091] Step S52: Based on all samples, with the effluent turbidity as the dependent variable and the raw water quality condition variable as the independent variable, further split the decision tree obtained in step S51. After the decision tree split is completed, clear the sample set corresponding to each node.

[0092] Step S53: Reassign the high-efficiency samples to the decision tree from step two, and calculate the optimal dosage value for each node. If some nodes are not assigned samples, use the optimal dosage of the upper-level node as the optimal dosage for this node, or use the Gaussian process regression described in step S6 for interpolation calculation.

[0093] The purpose of interpolation in step S6 is to assign values ​​to the leaf nodes that were not assigned samples in step S53. The purpose of extrapolation in step S7 is to ensure that even if the raw water quality conditions exceed the historical data range by a limited margin, the expert database can still provide reasonable control decisions for the new raw water quality conditions. This invention utilizes a Gaussian regression process to construct the relationship between the optimal dosage y and the corresponding raw water quality condition segment center x, and calculates the new water quality condition segment center x. * The predicted distribution is used to further calculate the optimal dosage for that area. When the prediction variance is large, the uncertainty of the optimal dosage is also large. To avoid the water quality at the outlet of the pool failing to meet the standards, the upper bound of the confidence interval is used as the optimal dosage.

[0094] See Figure 3 As shown, the calculation steps for the optimal dosage combining feedforward control and feedback adjustment in step S7 are as follows:

[0095] Step S71: Clear the dataset of the decision tree.

[0096] Step S72: Obtain the current time The raw water quality conditions are analyzed in real-time, including anomaly detection and missing value imputation. First, anomaly detection is performed to remove outliers, which are then filled with missing values. Next, the median of the first 10 records is used to fill in the missing values. After data cleaning, it is determined whether the data deviates significantly from historical samples. If it does, it is considered that we cannot provide a suitable dosage based on historical data, and manual dosing is switched to. Otherwise, the raw water quality conditions are input into a decision tree model to obtain the leaf node to which the current raw water quality conditions belong. To avoid treating "minor fluctuations" in raw water quality as "mutations," the cleaned data is smoothed to eliminate high-frequency noise, and a "mutation judgment condition" is set: only when the magnitude of the state change exceeds a threshold and the duration exceeds a set value is it judged as a mutation, triggering a node switch.

[0097] Step S73: See Figure 4 Calculate the inflow time corresponding to the current outflow based on the time delay. The current moment and the time of water intake Both are shifted backward by twice the standard deviation of the lag time distribution, and the results are obtained respectively. and Get Time The raw water quality conditions are input into the decision tree to obtain the leaf nodes to which the raw water quality conditions belong.

[0098] Step S74: Determine whether the leaf node corresponding to the raw water quality conditions in step S73 and the leaf node calculated in step S72 belong to the same group (i.e., whether they have the same raw water quality conditions). If so, record the water output time. and Then proceed to step S75. Otherwise, the water outlet time will be... and the time of water intake Simultaneously move backward while determining the raw water quality conditions at the time of water inflow. See also Figure 5 If the maximum movement length T max For the first time within this period, identical raw water quality conditions were found, and the corresponding effluent times were recorded. and the time of water intake Proceed to step S75. Otherwise, use the value of the leaf node to which the current raw water quality condition belongs in the decision tree as the current dosage, and proceed to step S76.

[0099] Step S75: Obtain the water inlet time The dosing consumption per unit, x, is used as the starting point for adjusting the dosing dosage at the current moment. Then, the effluent time is used... The weighted average turbidity NTU of the effluent turbidity samples before and after is calculated. Determine the current chemical dosage according to the NTU classification. See Figure 6 , the step size of chemical dosage adjustment is divided into three grades: large step size (Δx L ), medium step size (Δx M ) and small step size (Δx S ).

[0100] Case 1: NTU > upper tolerance limit. The chemical dosage is seriously low, increase the chemical dosage by a large step size, and determine the final chemical dosage per unit consumption as x + Δx L .

[0101] Case 2: Upper limit of normal range < NTU <= upper tolerance limit. The chemical dosage is low, increase the chemical dosage by a medium step size, and determine the final chemical dosage as x + Δx M .

[0102] Case 3: 0 < NTU < lower limit of normal range. The chemical dosage is high, reduce the chemical dosage by a medium step size, and determine the final chemical dosage as x - Δx L .

[0103] Case 4: Lower limit of normal range <= NTU < lower limit of normal range. The chemical agent may be overdosed, reduce the chemical dosage by a small step size, and determine the final chemical dosage as x - Δx S .

[0104] Step S76: Determine whether the chemical dosage at the current moment is the same as that at the previous moment. If it is the same, maintain the original state; if it is not the same, the system recommends the chemical dosage at the current moment for manual confirmation. When confirming, if the intelligent chemical dosage is accepted, execute it; if it is not accepted, switch to manual dosing.

[0105] Evaluate the effect after dosing. As time goes by, start to calculate the lag time of the influent water at the current moment. When the lag time is calculated, evaluate the effect of intelligent dosing. If the difference between the effluent turbidity and the NTU in S75 is too large, it is considered that there is an unobserved abnormality in the system, alarm, and stop the intelligent dosing, and switch to manual dosing.

[0106] The situation where the effluent turbidity difference is too large is divided into the following cases:

[0107] Case 1: The NTU in step S75 is greater than the upper tolerance limit, while the current effluent turbidity is less than the upper limit of the normal range;

[0108] Case 2: The NTU in step S75 is greater than the upper limit of the normal range, while the current effluent turbidity is less than the lower limit of the normal range; [[ID=四十八]]

[0109] Case 3: The NTU in step S75 is less than the upper limit of the normal range, while the current effluent turbidity is greater than the upper tolerance limit;

[0110] Scenario 4: The NTU in step S75 is less than the lower limit of the normal range, while the current effluent turbidity is greater than the upper limit of the normal range.

[0111] In step S8, the steps for updating the optimal dosage in the expert database while maintaining the unchanged decision tree structure are as follows:

[0112] Step S81: If the water outlet time If the NTU calculated in step S75 is within the normal range (i.e., between the "lower limit of the normal range" and the "upper limit of the normal range"), then the corresponding raw water quality conditions and dosage data will be added to the decision tree sample.

[0113] Step S82: Calculate the optimal dosage using stochastic frontier analysis (SFA).

[0114] The update of the intelligent dosing decision expert database in step S9 will go through steps S1-S6 again.

[0115] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for adding coagulant to a water plant based on an intelligent expert decision-making model, characterized in that, Includes the following steps: S1: Collect historical data on coagulant addition and preprocess it, including data anomaly detection and missing value filling, as well as calculation of effluent lag time; S2: Select samples that meet the dosing effect, i.e., samples whose turbidity at the outlet is close to the target value; S3: For scenarios where the dosage changes frequently, high-efficiency dosing samples are selected based on the KAN-SFA model; S4: For scenarios where there are only a few dosage levels, group the dosage levels and screen for efficient dosage samples based on fuzzy comprehensive evaluation; S5: Construct an initial intelligent dosing decision expert database based on a decision tree regression model; S6: Expanding the expert database for intelligent dosing decision-making based on Gaussian process regression model, including two processes: interpolation and extrapolation; S7: Under different raw water quality conditions, the initial dosage is obtained based on the intelligent dosing decision expert database, and the dosage is continuously and dynamically adjusted according to the turbidity feedback results after a certain time lag. S8: Keep the decision tree structure unchanged and re-estimate the optimal dosage using the new data; S9: Once the new historical data on dosing has accumulated to a certain scale, update the intelligent dosing decision expert database using steps S1-S6.

2. The method for adding coagulant to a water plant based on an intelligent expert decision-making model as described in claim 1, characterized in that, In step S1, the turbidity of the effluent at the lag time corresponding to the dosing time is calculated based on the weighted average of the effluent turbidity within twice the standard deviation of the effluent lag time, with the weights being the values ​​of the normal distribution probability density function.

3. The method for adding coagulant to a water plant based on an intelligent expert decision-making model as described in claim 1, characterized in that, The basic structure of the KAN-SFA model used for screening high-efficiency dosing samples in step S3 is as follows: y=exp(KAN(x;β))·exp(ε), ε=v+u in, The independent variable X represents the raw water quality conditions, the dependent variable y represents the chemical consumption per unit of dosage, KAN(·) represents the KAN model, |β represents the parameters of KAN, and ε represents other factors affecting the optimal dosage, consisting of a random error term v and an inefficiency term u. v follows a normal distribution, and u > 0 and follows a semi-normal distribution.

4. The method for adding coagulant to a water plant based on an intelligent expert decision-making model as described in claim 1, characterized in that, Step S4, the efficient dosing sample screening method based on fuzzy comprehensive evaluation, includes the following steps: S41: Determine the domain of factors for the evaluation object; S42: Determine the evaluation level domain V, and the evaluation set is: V = {Excellent, Good, Average, Poor, Very Poor}; S43: Establish a single-factor fuzzy relation matrix to reflect the fuzzy relationship between each factor and the evaluation level; S44: Establish the weights of each factor; S45: Conduct fuzzy comprehensive evaluation to reflect the degree of membership of the evaluated object at each evaluation level; S46: Calculate the overall score.

5. The method for adding coagulant to a water plant based on an intelligent expert decision-making model as described in claim 1, characterized in that, Step S5 involves constructing the initial intelligent dosing decision expert database using decision tree regression, which is carried out in three steps: S51: Based on the selected qualified and efficient samples, with the dosage as the dependent variable and the raw water quality condition variable as the independent variable, a decision tree regression model with a small depth is constructed. The value of the leaf node is the optimal dosage. After the decision tree is constructed, the sample set corresponding to each node is cleared. S52: Based on all samples, with the effluent turbidity as the dependent variable and the raw water quality condition variable as the independent variable, further split the decision tree obtained in step S51. After the decision tree split is completed, clear the sample set corresponding to each node. S53: Reassign the high-efficiency samples to the decision tree in S52, calculate the optimal dosage value for each node. If some nodes are not assigned samples, use the optimal dosage of the upper-level node as the optimal dosage of this node, or use Gaussian process regression in step S6 for interpolation calculation.

6. The method for adding coagulant to a water plant based on an intelligent expert decision-making model as described in claim 1, characterized in that, The calculation steps for the optimal dosage combining feedforward control and feedback regulation in step S7 are as follows: S71: Clear the dataset for the decision tree; S72: Get the current time The raw water quality conditions are analyzed in real time, including anomaly detection and missing value imputation. After data cleaning, it is determined whether the data deviates significantly from historical samples. If it deviates significantly, it is considered that an appropriate dosage cannot be given based on historical data, and manual dosing is switched to the next step. Otherwise, the raw water quality conditions are input into the decision tree model to obtain the leaf node to which the current raw water quality conditions belong. In order to avoid treating "minor fluctuations" in raw water quality as "mutations", the cleaned data is smoothed to eliminate high-frequency noise, and "mutation judgment conditions" are set: only when the magnitude of the state change exceeds the threshold and the duration exceeds the set value is it judged as a mutation and a node switch is triggered. S73: Calculate the inflow time corresponding to the current outflow based on the time delay. The current moment and the time of water intake Both are shifted backward by twice the standard deviation of the lag time distribution, and the results are obtained respectively. and Get Time The raw water quality conditions are input into the decision tree to obtain the leaf nodes to which the raw water quality conditions belong; S74: Determine whether the leaf node corresponding to the water quality conditions in step S73 and the leaf node calculated in step S72 belong to the same entity. If so, record the water effluent time. and Then proceed to step S75; otherwise, the water outlet time will be... and water intake time Simultaneously move backward, while judging the water quality conditions at the time of water inflow; if it is within the upper limit of the movement length T max For the first time in the country, similar water quality conditions were found, and the corresponding water outflow time was recorded. and water intake time Proceed to step S75; otherwise, use the value of the leaf node to which the current water quality condition belongs in the decision tree as the current dosage and proceed to step S76. S75: Obtain the water ingress time The dosing order | consumption x is used as the starting point for adjusting the dosing dosage at the current moment, utilizing the effluent time. The weighted average turbidity NTU of the effluent is calculated from the turbidity samples before and after the effluent, and the current dosage is determined based on the NTU level. S76: Determine if the current dosage is consistent with the previous dosage; if consistent, maintain the current state; if inconsistent, the system recommends the current dosage, pending manual confirmation. Upon confirmation, if the intelligent dosage is accepted, it will be executed accordingly. If not accepted, then manual application will be used; The effect of dosing is evaluated. If the turbidity at the outlet of the tank differs too much from the NTU in step S75, it is considered that an unobserved abnormality has occurred in the system. An alarm is triggered and intelligent dosing is stopped, and manual dosing is switched to manual dosing.

7. The method for adding coagulant to a water plant based on an intelligent expert decision-making model as described in claim 6, characterized in that, In step S8, the steps for updating the optimal dosage in the expert database while maintaining the decision tree structure are as follows: S81: If the water outlet time If the NTU calculated in step S75 is within the normal range, then the corresponding raw water quality conditions and dosage data will be added to the decision tree sample. S82: Calculate the optimal dosage using stochastic frontier analysis; The intelligent dosing decision expert database in step S8 is updated online in a localized manner.

8. The method for adding coagulant to a water plant based on an intelligent expert decision-making model as described in claim 1, characterized in that, The update of the intelligent dosing decision expert database in step S9 will go through steps S1-S6 again. The intelligent dosing decision expert database in step S9 is updated offline globally.

Images