An intelligent modeling method for coagulant dosing in water treatment plants based on big data analysis

Abstract

The present invention provides an intelligent modeling method for coagulant addition in a water treatment plant based on big data analysis. Based on the historical big data analysis of raw water quality, coagulant dosage and sedimentation tank effluent turbidity, long-term and short-term models of coagulant dosage under corresponding water quality conditions are constructed respectively according to the characteristics of periodic seasonal changes and non-periodic dynamic changes in raw water quality; in actual operation control, the weights of the long-term and short-term models are adjusted online according to the control effect of the effluent turbidity of the sedimentation tank at the current moment; the outputs of the long-term and short-term models are weighted and calculated to obtain the predicted output of the coagulant dosage model under the current water quality conditions, and the coagulant dosage is adjusted online. The present invention adopts a big data analysis method to intelligently and real-timely control the coagulant dosage, effectively improve the treatment effect of the coagulation process with complex nonlinear and large hysteresis characteristics, stabilize and improve the quality of drinking water, and ensure drinking water safety.

Description

Technical Field

[0001] The present invention relates to an intelligent modeling method, and in particular to an intelligent modeling method for coagulant addition in a water purification plant based on big data analysis. Background Art

[0002] In the water treatment process of the water treatment plant, turbidity removal and clarification are one of the main goals. The main purpose of adding coagulants is to make the turbid substances in the raw water aggregate to form flocs with a certain particle size and surface characteristics, creating good conditions for precipitation or filtration removal. Since a large number of microorganisms such as bacteria and viruses are attached to the surface of the flocs, the removal of microorganisms such as bacteria and viruses in the water is also very obvious in this process. Therefore, strengthening the coagulation effect and strictly controlling the turbidity of the effluent from the sedimentation tank and sand filter are conducive to the removal of viruses and ensuring the safety of the effluent water quality of the water treatment process.

[0003] Under certain process conditions, the coagulation effect is determined by the coagulant dosage control. On the one hand, the raw water quality has a significant impact on the coagulant consumption, but the current raw water quality big data information of the water treatment plant is still in the data collection / display stage, and has not been fully and timely involved in the coagulant dosage control; on the other hand, there is a large lag time of at least 2 hours from coagulant addition to water discharge from the sedimentation tank. Relying on traditional feedback control methods, it is difficult to timely and effectively overcome the impact of raw water quality changes on coagulant consumption. Therefore, it is very important to fully explore the raw water quality big data information and adjust the coagulant dosage in real time according to the current raw water quality.

[0004] The coagulant dosage control of the water plant, the research object of this invention, Xiangcheng Water Plant of Suzhou Water Supply Co., Ltd., still adopts the traditional feedback control method. It is impossible to adjust the coagulant dosage in real time according to the current raw water quality, and it is difficult to effectively ensure the coagulation effect and control the turbidity of the effluent, which seriously affects the safety of the water quality out of the factory. Summary of the invention

[0005] Purpose of the invention:

[0006] According to the process conditions of the water plant under study, the present invention proposes an intelligent modeling method that can calculate the coagulant dosage in real time according to the periodic seasonal changes and non-periodic dynamic changes of raw water quality, solves the optimization control problem of the coagulation and sedimentation process with complex nonlinear and large hysteresis characteristics, and effectively improves the coagulation effect.

[0007] Technical solution:

[0008] The present invention adopts the following technical solutions to solve the above technical problems:

[0009] A method for intelligent modeling of coagulant addition in a water treatment plant based on big data analysis comprises the following steps:

[0010] In step (I), the steps for establishing the long-term model of coagulant dosage are as follows:

[0011] Step 1: extract, clean and analyze the big data of raw water quality indicators (pH, water temperature, dissolved oxygen, oxygen consumption, turbidity), sedimentation tank inlet flow, coagulant dosage and sedimentation tank effluent turbidity history (past 3 years and above) in the production and operation database of the water treatment plant, screen out the big data sample set of raw water quality indicators, sedimentation tank inlet flow, and coagulant dosage history (past 3 years and above) at the corresponding time (i.e. before the large lag time of the coagulation / sedimentation process) that meets the sedimentation tank effluent turbidity requirements, and divide them into 12 sub-sample sets by month;

[0012] In step 2, the raw water quality indicators (pH, water temperature, dissolved oxygen, oxygen consumption, turbidity) and sedimentation tank inlet flow rate are used as input, and the coagulant dosage is used as output. The random forest algorithm is used to train 12 long-term models of coagulant dosage (1 to 12 months) based on the 12 sub-sample sets obtained in step 1.

[0013] Step (II), the steps for establishing the short-term model of coagulant dosage are as follows:

[0014] Step 1: extract, clean and analyze the big data of raw water quality indicators (pH, water temperature, dissolved oxygen, oxygen consumption, turbidity), sedimentation tank inlet flow, coagulant dosage and sedimentation tank effluent turbidity history (past 15 days) in the water treatment plant production and operation database, and select the big data sample set of raw water quality indicators, sedimentation tank inlet flow, and coagulant dosage history (past 15 days) at the corresponding time (i.e. before the large lag time of the coagulation / sedimentation process) that meets the sedimentation tank effluent turbidity requirements;

[0015] In step 2, the raw water quality indicators (pH, water temperature, dissolved oxygen, oxygen consumption, turbidity) and sedimentation tank inlet flow rate are used as input, and the coagulant dosage is used as output. The random forest algorithm is used to train the short-term model of coagulant dosage based on the big data sample set obtained in step 1.

[0016] Step (III), the intelligent weighting steps of the long-term and short-term models of coagulant dosage are as follows:

[0017] Step 1: Calculate the deviation between the turbidity of the sedimentation tank effluent at the current moment and the set value. If the deviation is greater than a certain threshold, it means that the actual coagulant dosage at the corresponding moment (i.e. before the large lag time of the coagulation / sedimentation process) is unreasonable, and then start step 2; otherwise, it means that the dosage is reasonable, and the current weight remains unchanged;

[0018] Step 2: If the current turbidity of the sedimentation tank effluent is greater than the set value, it means that the actual coagulant dosage at the corresponding moment is too small. At this time, the current long-term model and short-term model weights are adjusted based on the principle of increasing the coagulant dosage; otherwise, it means that the actual coagulant dosage at the corresponding moment is too large. The current long-term model and short-term model weights are adjusted based on the principle of reducing the coagulant dosage;

[0019] If the current turbidity of the sedimentation tank effluent is greater than the set value, it means that the actual coagulant dosage at the corresponding moment (i.e. before the long lag time of the coagulation / sedimentation process) is too small, then it is judged that the weight of the model with the larger long and short-term coagulant dosage output at the corresponding moment (long-term model or short-term model) is too small, so the weight of the model at the current moment is increased (the adjustment step is 5% / time, the adjustment cycle is 1 hour / time), and the weight of the other model at the current moment is reduced (the adjustment step is 5% / time, the adjustment cycle is 1 hour / time); If the current turbidity of the sedimentation tank effluent is less than the set value, it means that the actual coagulant dosage at the corresponding moment (that is, before the long lag time of the coagulation / sedimentation process) is too large. Then it is judged that the weight of the model with the larger long and short-term coagulant dosage output (long-term model or short-term model) is too large, so the weight of the model at the current moment is reduced (the adjustment step is 5% / time, and the adjustment cycle is 2 hours / time), and the weight of the other model at the current moment is increased (the adjustment step is 5% / time, and the adjustment cycle is 2 hours / time).

[0020] Step 3, using the long-term model of the current month and the short-term model of the past 15 days, obtain the coagulant dosage under the current raw water quality and sedimentation tank inlet flow conditions through weighted calculation.

[0021] Beneficial effects:

[0022] The present invention establishes a coagulant dosing model based on the analysis of raw water quality big data, and intelligently controls the coagulant dosage in real time, effectively improving the coagulation effect. The method of the present invention is scientific, reasonable, and feasible, does not require any additional hardware costs, and is particularly suitable for coagulation dosing control in water production processes with obvious raw water quality mutation phenomena. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is the process flow chart of coagulant addition in water treatment plants;

[0024] Figure 2 It is a flow chart of the intelligent modeling method for coagulant addition in water treatment plants;

[0025] Figure 3 It is a schematic diagram of the long-term and short-term model training of coagulant dosage;

[0026] Figure 4 It is a schematic diagram of the long-term and short-term model prediction of coagulant dosage;

[0027] Figure 5 This is a schematic diagram of random forest model construction. DETAILED DESCRIPTION

[0028] The technical solution of the present invention is described in detail below in conjunction with the actual online test at the Xiangcheng Water Plant of Suzhou Water Supply Co., Ltd.:

[0029] (I) Obtaining the coagulant dosage FD for the long-term model of coagulant dosage 1 , the specific steps are as follows:

[0030] Step 1: Through data preprocessing, the big data sample set of raw water quality indicators, sedimentation tank inflow flow, and coagulant dosage history (past 3 years and above) at the corresponding time (i.e. before the long lag time of the coagulation / sedimentation process) that meets the sedimentation tank effluent turbidity requirements in the water treatment plant production and operation database is screened out, and the big data sample set is divided into 12 sub-sample sets by month;

[0031] Step 2: Using raw water quality indicators (pH, water temperature, dissolved oxygen, oxygen consumption, turbidity) and sedimentation tank inlet flow as input and coagulant dosage as output, the random forest algorithm is used to train 12 long-term models of coagulant dosage (1 to 12 months) based on the 12 subsample sets obtained in step 1;

[0032] Step 3: Input the real-time data of the current raw water quality indicators (pH, water temperature, dissolved oxygen, oxygen consumption, turbidity) and the sedimentation tank inlet flow into the long-term model of coagulant dosage for the current month to calculate the corresponding coagulant dosage FD. 1 .

[0033] (II) Obtaining the coagulant dosage FD of the short-term model of coagulant dosage 2 , the specific steps are as follows:

[0034] Step 1: Through data preprocessing, filter out the big data sample set of raw water quality indicators, sedimentation tank inflow flow, and coagulant dosage history (past 15 days) at the corresponding time (i.e. before the long lag time of the coagulation / sedimentation process) that meet the sedimentation tank effluent turbidity requirements in the water treatment plant production and operation database;

[0035] In step 2, the raw water quality indicators (pH, water temperature, dissolved oxygen, oxygen consumption, turbidity) and sedimentation tank inlet flow rate are used as input, and the coagulant dosage is used as output. The random forest algorithm is used to train the short-term model of coagulant dosage based on the big data sample set obtained in step 1.

[0036] Step 3: Input the current raw water quality indicators (pH, water temperature, dissolved oxygen, oxygen consumption, turbidity) and the real-time data of the sedimentation tank inlet flow into the current coagulant dosage short-term model to calculate the corresponding coagulant dosage FD. 2 .

[0037] Supplementary explanation: The update cycle of the short-term model of coagulant dosage can be determined according to the actual situation (such as changes in raw water quality). Taking the research object water plant as an example, the short-term model is retrained and updated once a day.

[0038] (III) Obtaining the calculated value FD of the coagulant dosage at the current moment, the specific steps are as follows:

[0039] Step 1: Calculate the deviation between the turbidity TU of the sedimentation tank effluent at the current moment and the set value REF. If the deviation is greater than a certain threshold, it means that the actual coagulant dosage FD' at the corresponding moment is unreasonable, and then judge the size relationship between the turbidity TU of the sedimentation tank effluent at the current moment and the set value REF, and start step 2 or step 3; otherwise, it means that the actual coagulant dosage FD' at the corresponding moment is reasonable, and keep the current weight unchanged;

[0040] Step 2, if the current sedimentation tank effluent turbidity TU is greater than the set value REF, it means that the actual coagulant dosage FD' at the corresponding moment is too small, so it is judged that the weight of the larger model in the long-term and short-term model outputs of the coagulant dosage at the corresponding moment is too small, so the weight of the model at the current moment is increased and the weight of the other model at the current moment is reduced (the adjustment step is 5% / time, and the adjustment cycle is 1 hour / time);

[0041] Step 3. If the current sedimentation tank effluent turbidity TU is less than the set value REF, it means that the actual coagulant dosage FD′ at the corresponding moment is too large. Then it is judged that the weight of the larger model in the long-term and short-term model outputs of the coagulant dosage at the corresponding moment is too large. Therefore, the weight of the model at the current moment is reduced and the weight of the other model at the current moment is increased (the adjustment step is 2% / time, and the adjustment cycle is 2 hours / time).

[0042] Step 4: Coagulant dosage FD output based on the long-term model of the current month and the short-term model of the past 15 days 1 and FD 2 , and the corresponding model weight ω 1 and ω 2 , the current coagulant dosage is obtained by weighted calculation: FD = ω 1 ·FD 1 +ω 2 ·FD 2 .

[0043] (IV) Regarding the construction process of the random forest model, the specific steps are as follows:

[0044] Step 1: randomly extract m groups of data samples with replacement from the large data sample set (assuming there are n groups of data samples) as training subsets, and the remaining nm groups of data samples as test subsets;

[0045] Step 2: Each training subset is grown into a regression decision tree without leaf pruning. At each node of the decision tree, k features are randomly selected from the K features (k≤K). At each node, the optimal feature is selected from the k features for branch growth. This decision tree grows fully without pruning, so that the impurity of each node is minimized.

[0046] Step 3, according to the nm group data samples of the test subset, respectively establish nm decision tree models, and take the average value of the outputs of the nm decision tree models as the output of the random forest model;

[0047] Step 4, calculate the deviation between the random forest model output of the test subset and the sample output. If the deviation is greater than the threshold, repeat steps 1 to 4; otherwise, the training ends, and the random forest model obtained from the last training is the long / short-term model of coagulant dosage under the big data sample set.

[0048] (V) The calculation of the maximum lag time of the coagulation / sedimentation process is explained in detail as follows:

[0049] The maximum lag time τ of the coagulation / sedimentation process is inversely proportional to the influent flow rate Q of the sedimentation tank. Taking the water plant as an example, it is determined that the influent flow rate of the sedimentation tank is 3000 cubic meters / hour, and the corresponding maximum lag time of the coagulation / sedimentation process is 2.1 hours. Therefore, the corresponding relationship between τ and Q is:

Claims

1. An intelligent modeling method for coagulant dosing in water treatment plants based on big data analysis, It is characterized in that The specific steps include: Step 1: Based on the periodic seasonal variation characteristics of raw water quality, the raw water quality indicators and sedimentation tank inflow flow in the past three years or more of historical big data are used as input, and the coagulant dosage is used as output. A long-term model of coagulant dosage is established based on the random forest algorithm; Step 2: In view of the non-periodic dynamic change characteristics of raw water quality, the raw water quality indicators and sedimentation tank inflow flow in the past 15 days of historical big data are used as input, and the coagulant dosage is used as output. Based on the random forest algorithm, a short-term model of coagulant dosage is established; Step 3: According to the turbidity control effect of the sedimentation tank effluent at the current moment, the rationality of the actual coagulant dosage at the corresponding moment is evaluated, and the weights of the long-term model and the short-term model at the current moment are adjusted online accordingly, including: Step 31, calculate the deviation between the turbidity of the effluent from the sedimentation tank at the current moment and the set threshold value. If the deviation is greater than the set threshold value, it means that the actual dosage of the coagulant at the corresponding moment is unreasonable, and then start step 2; otherwise, it means that the dosage is reasonable, and the current weight value remains unchanged; the corresponding moment is before the large lag time of the coagulation / sedimentation process; Step 32, if the current turbidity of the effluent from the sedimentation tank is greater than the set value, it means that the actual coagulant dosage at the corresponding moment is too small, and the current long-term model and short-term model weights are adjusted based on the principle of increasing the coagulant dosage. The weight of the model with a larger output in the long-term model and short-term model of coagulant dosage at the corresponding moment is too small, so the weight of the model with a larger output in the long-term model and short-term model at the current moment is increased, and the weight of the other model at the current moment is reduced; on the contrary, it means that the actual coagulant dosage at the corresponding moment is too large, and the current long-term model and short-term model weights are adjusted based on the principle of reducing the coagulant dosage, and the weight of the model with a larger output in the long-term model and short-term model of coagulant dosage at the corresponding moment is too large, so the weight of the model with a larger output in the long-term model and short-term model at the current moment is reduced, and the weight of the other model at the current moment is increased; Step 4, according to the adjusted weights, weighted calculation of the long-term model and short-term model outputs to obtain the coagulant dosage under the current raw water quality and sedimentation tank inlet flow conditions.

2. The intelligent modeling method for adding coagulant in a water treatment plant based on big data analysis as claimed in claim 1, It is characterized in that In step 1, the following steps are specifically included: Step 11, extracting, cleaning and analyzing the historical big data of raw water quality indicators, sedimentation tank inlet flow, coagulant dosage and sedimentation tank effluent turbidity in the past three years and above in the production and operation database of the water treatment plant, screening out the raw water quality indicators, sedimentation tank inlet flow, and coagulant dosage historical time big data sample sets that meet the sedimentation tank effluent turbidity requirements at the corresponding time, and dividing them into 12 sub-sample sets by month; the raw water quality indicators include pH, water temperature, dissolved oxygen, oxygen consumption, and turbidity, and the corresponding time is before the large lag time of the coagulation / sedimentation process; In step 12, the raw water quality index and the sedimentation tank inlet flow rate are used as input, and the coagulant dosage is used as output. The random forest algorithm is used to train 12 long-term models of coagulant dosage according to the 12 sub-sample sets obtained in step 11.

3. The intelligent modeling method for adding coagulant in a water treatment plant based on big data analysis as claimed in claim 1, It is characterized in that In step 2, the following steps are specifically included: Step 21, extracting, cleaning and analyzing the historical big data of raw water quality indicators, sedimentation tank inlet flow, coagulant dosage and sedimentation tank effluent turbidity in the past 15 days in the production and operation database of the water treatment plant, and screening out the historical big data sample set of raw water quality indicators, sedimentation tank inlet flow and coagulant dosage in the past 15 days at the corresponding time that meets the sedimentation tank effluent turbidity requirements; the raw water quality indicators include pH, water temperature, dissolved oxygen, oxygen consumption and turbidity, and the corresponding time is before the large lag time of the coagulation / sedimentation process; Step 22, taking the raw water quality index and the sedimentation tank inlet flow rate as input and the coagulant dosage as output, a random forest algorithm is used to train a short-term model of coagulant dosage based on the big data sample set obtained in step 21.

4. The intelligent modeling method for adding coagulant in a water treatment plant based on big data analysis as claimed in claim 2 or 3, Its characteristics are: The maximum lag time of the coagulation / sedimentation process is inversely proportional to the inflow rate of the sedimentation tank. When the inflow rate of the sedimentation tank is 3000 cubic meters per hour, the corresponding maximum lag time of the coagulation / sedimentation process is 2.1 hours. Therefore, when the inflow rate of the sedimentation tank is Q, the maximum lag time of the coagulation / sedimentation process is 5. The intelligent modeling method for adding coagulant in a water treatment plant based on big data analysis as claimed in claim 2 or 3, It is characterized in that The random forest algorithm specifically includes the following steps: Step (1), randomly extracting m groups of data samples with replacement from n groups of large data sample sets as training subsets, and the remaining nm groups of data samples as test subsets; Step (2), each training subset is grown into a regression decision tree without leaf pruning. At each node of the decision tree, k features are randomly selected from the K features, k ≤ K. At each node, the optimal feature is selected from the k features for branch growth. This decision tree is fully grown without pruning, so that the impurity of each node is minimized. Step (3), according to the nm group data samples of the test subset, respectively establish nm decision tree models, and take the average value of the outputs of the nm decision tree models as the output of the random forest model; Step (4), calculate the deviation between the random forest model output of the test subset and the sample output. If the deviation is greater than the threshold, repeat steps 1 to 4; otherwise, the training ends, and the random forest model obtained from the last training is the long / short-term model of coagulant dosage under the big data sample set.

Images