Optimal dispatching method of water inflow in water plant based on hybrid model
By building a hydrodynamic model and data model in a tap water plant, using the XGBoost algorithm to predict and optimize water inlet volume, the problem of insufficient water inlet scheduling dependence on experience and prediction in the existing technology is solved, and more efficient water inlet volume is achieved Management and cost reduction.
Patent Information
- Application Number
- CN202411186279.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-28
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-08-28
AI Technical Summary
The existing water inlet scheduling method of tap water plants depends on experience, lacks a complete prediction model, and cannot fully describe the hydraulic conditions of the entire process section, especially in special scenarios, which leads to insufficient safety operation level of the pipeline network.
The water inlet optimization scheduling method based on the mixed model is adopted to construct the water plant hydrodynamic model by extracting design data and historical monitoring data, generate multiple data samples, use the XGBoost regression algorithm to predict and optimize and process it, and output step prediction. The water volume meets the water supply demand, achieves smooth operation of each process section and high liquid level operation of the clean water pool, reduces the suction range of the water pump and reduces electricity consumption.
It improves the accuracy and adaptability of water inlet forecasting, reduces production costs, enhances adaptability to special scenarios, and ensures the safe and stable operation of the tap water plant.
Smart Images

Figure CN118710019B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of water inlet optimization scheduling of waterworks, and in particular to a method for optimizing water inlet scheduling of waterworks based on a hybrid model. Background Art
[0002] The control of water inflow in a waterworks is a key link in ensuring the stable operation of the water supply system, but water inflow scheduling is a complex process that takes into account multiple factors, and requires comprehensive analysis and judgment based on the water quantity and quality of the water source, water supply demand, and water plant processing capacity. With the continuous advancement of science and technology, waterworks have adopted more advanced technologies and methods in water inflow control, such as: through the application of automated control systems, real-time monitoring and precise control of water inflow can be achieved; through the establishment of a water inflow database, historical data can be collected, analyzed and mined, the changing laws of water inflow can be mastered, and decision support can be provided for water inflow control. Although waterworks have made certain progress in water inflow control, there is still a lack of a complete water inflow prediction model. At present, water inflow scheduling mostly relies on scheduling personnel to schedule based on experience according to water supply demand, clear water tank level, pipeline pressure, and water use plan.
[0003] At present, the following methods are mainly used in the prediction research of water inflow: BP neural network and vector machine methods based on big data and machine learning, based on time series analysis, based on considering multiple factors and grey theory to establish a water plant water inflow prediction model. Researchers have constructed various mathematical models and algorithms to try to mine the changing laws of water supply from massive data to improve the accuracy and reliability of the prediction. However, most methods rely on real-time monitoring data, have high requirements for data quality and completeness, cannot fully describe the hydraulic conditions of the entire process section, and lack predictions for special scenarios.
[0004] Therefore, the existing technology still has obvious limitations in the water inlet scheduling management of water plants. There is an urgent need for a more intelligent and comprehensive method to support the water inlet scheduling management and improve the level of safe operation of the pipeline network.
[0005] The applicant has conducted as detailed a search of the prior art as possible and has not found any technical solution in the prior art related to the water intake scheduling of the waterworks related to the present application.
[0006] Therefore, it is necessary to provide a new technical solution to solve the above technical problems. Summary of the invention
[0007] The present invention aims to propose a method for optimizing the water intake of a water plant based on a hybrid model, comprising the following steps:
[0008] S1: Extract the design data of the water plant and the historical monitoring data of the water plant. The extracted historical monitoring data include water inflow, water supply, flow rate of each process section, liquid level of each water tank, and operation data of each water pump;
[0009] S2: Construct a waterworks hydrodynamic model based on the design data and historical monitoring data extracted in S1;
[0010] S3: Synchronously adjust the water inlet and water supply in S1 to generate multiple data samples;
[0011] S4: Run the waterworks hydrodynamic model based on the data samples generated in S3 to generate the clear water tank level, and form water inflow, water supply and clear water tank level data samples;
[0012] S5: Based on the data samples generated in S4, train the data samples and predict the water inflow;
[0013] S6, performing optimization processing based on the water inflow predicted in S5, and outputting the optimized predicted water inflow.
[0014] As a preferred solution, the water tank includes a sedimentation tank, a front sand tank, a carbon filter tank, a rear sand tank, a recycled water tank, and a clear water tank.
[0015] As a preferred solution, the waterworks design data includes process flow, dimensions of each building, dimensions of each pipe section, and parameters of each water pump.
[0016] As a preferred solution, the historical monitoring data extracted in S1 is preprocessed, and a waterworks hydrodynamic model is constructed in S2 based on the design data and the preprocessed historical monitoring data.
[0017] As a preferred solution, the extracted historical monitoring data preprocessing method includes but is not limited to data cleaning, removal of outliers, smoothing and interpolation.
[0018] As a preferred solution, SWMM is used in S2 to construct a waterworks hydrodynamic model.
[0019] As a preferred solution, the water power model of the waterworks is further provided between S2 and S3.
[0020] After construction, the uncertainty parameters are calibrated. After the model is calibrated, the water inlet and water supply data in S1 are used for simulation to generate the clear water tank level. The historically monitored clear water tank level is analyzed with the simulated clear water tank level data to determine whether the water plant hydrodynamic model meets the standards. If so, S3 is executed, otherwise the water plant hydrodynamic model is readjusted.
[0021] As a preferred solution, when the absolute error between the historically monitored clear water tank level and the simulated clear water tank level is less than 0.3 m, the waterworks hydrodynamic model meets the standards.
[0022] As a preferred solution, S3 specifically includes: synchronously adjusting the water intake and water supply at the same time according to random parameters to generate multiple data samples.
[0023] As a preferred solution, the water supply, the water inflow at the previous moment and the liquid level of the clean water tank in the data sample generated by S4 in S5 are used as feature columns, and the water inflow is used as the target variable.
[0024] As a preferred solution, the data samples generated by S4 in S5 include a training set and a test set, wherein the training set accounts for 70% of the total data samples and the test set accounts for 30% of the total data samples.
[0025] As a preferred solution, S6 comprises the following steps:
[0026] S61: Check whether the water inflow value predicted by S5 exceeds the highest threshold or the lowest threshold, and perform a backward summation process on the part exceeding the threshold or the part below the threshold. After the summation process is completed, the check and summation process are repeated until there is no situation where the water inflow value exceeds the threshold or is below the threshold;
[0027] S62: Calculate the average value of the predicted water inflow after the addition process in step S61, and mark the predicted water inflow values in S5 differently based on the average value of the water inflow, mark the data below the average value, and mark the data above the average value differently;
[0028] S63: Merge points with the same mark, and when the continuous time of data is less than the set time, merge the mark to the mark of the previous data segment or the next data segment;
[0029] S64: Based on the marks in S63, the average value of the same segment and the same mark is calculated;
[0030] S65: Output the predicted water inflow after optimization processing.
[0031] This application aims to optimize scheduling and reduce production costs by establishing a hybrid model that couples the water plant hydrodynamic model and the data model. The hybrid model is based on monitoring the water inflow and water supply. By adjusting different water inflows and water supplies, the water plant hydrodynamic model is used to simulate the operation of multiple water plant operation events under different operating conditions, and the clear water tank level heights under different water inflows and water supplies are generated. The operation scenarios fully cover the water plant operation conditions including extreme conditions, and then the data of multiple operating condition operation events are trained and simulated to generate a data model. The water supply and clear water tank level can be input into the data model to predict the water inflow, and The prediction results are optimized to generate a step-wise predicted water inflow; the model is established with the goal of optimizing scheduling and reducing production costs. On the basis of meeting the water supply demand, the smooth operation of each process section of the water plant and the high liquid level operation of the clean water are achieved, the clean water tank maintains a high liquid level, reduces the suction of the pump, reduces the consumption of electricity, reduces production costs, and improves economic benefits; in the data processing stage, this application not only focuses on real-time monitoring data, but also processes the design data of the water plant, comprehensively considers information from more dimensions, rather than relying solely on real-time monitoring data, uses SWMM to construct a hydrodynamic model, and combines it with the XGBoost regression algorithm, which is different in the comprehensiveness and adaptability of the model, and better describes the hydraulic conditions of the entire process section; and this application includes steps for optimizing the prediction results, not only for prediction, but also for optimization and adjustment of the results to meet the actual scheduling needs, which to a certain extent makes up for the problem that common methods are insufficient in predicting special scenarios, and improves adaptability to special situations through more comprehensive data processing and comprehensive model construction. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 It is a schematic diagram of the optimization scheduling method of the present application;
[0033] Figure 2 The output diagram of the predicted water inflow, the marking points, and the optimized predicted water inflow of the second embodiment of the present application; DETAILED DESCRIPTION
[0034] The specific implementation of the present invention is described in detail below. It should be noted that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention. Embodiment 1
[0035] This embodiment proposes a method for optimizing the water intake of a water plant based on a hybrid model, which specifically includes the following steps:
[0036] S1: Extracting the design data of the water plant from the CAD design drawing of the water plant, the extracted design data includes the process flow, the size of each building (including the length, width, height and bottom elevation of the building), the size of each pipe section (including the diameter of each pipe section and the elevation of each pipe section), and the parameters of each water pump; extracting the historical monitoring data of the water plant, the extracted historical monitoring data includes the water inflow, water supply, flow of each process section, liquid level of each water tank, and the operating status of each water pump, wherein the water tank includes a sedimentation tank, a front sand tank, a carbon filter tank, a rear sand tank, a recycled water tank, and a clear water tank; the operating status of the water pump includes the switch status and frequency; preferably, the historical monitoring data is preprocessed, and the preprocessing method includes but is not limited to data cleaning, removing outliers, smoothing and interpolation;
[0037] S2: constructing a water power model of the water plant using SWMM based on the design data and historical monitoring data in S1; preferably, constructing a water power model of the water plant using SWMM based on the design data and pre-processed historical monitoring data;
[0038] S21, in order to improve the accuracy of the waterworks hydrodynamic model, after the waterworks hydrodynamic model is constructed, the uncertainty parameters are calibrated, and the uncertainty parameters include the orifice coefficient, etc.; after the model is calibrated, the water inlet and water supply data in S1 are used for simulation to generate the clear water tank level, and the historically monitored clear water tank level data is analyzed with the simulated clear water tank level data to determine whether the waterworks hydrodynamic model meets the standards. If so, S3 is executed, otherwise the waterworks hydrodynamic model is readjusted.
[0039] Specifically: when the absolute error value between the historically monitored clear water tank level and the simulated clear water tank level is less than 0.3m, the water plant hydrodynamic model meets the standards.
[0040] S3: Adjust the water inflow and water supply in S1 to generate multiple data samples, including the water inflow and water supply. For example, select 2 days of data in S1, process them into 15-minute intervals as a data sample, and synchronously adjust the water inflow and water supply at the same time according to random parameters, such as multiplying the water inflow and water supply at each moment of a piece of data by 0.9 times, or by 1.1 times, etc., to generate a total of no less than 1,000 data;
[0041] S4: Run the waterworks hydrodynamic model based on the data samples generated in S3 to generate the clear water tank level, and form water inflow, water supply and clear water tank level data samples;
[0042] S5: Based on the data samples generated in S4, the XGBoost regression algorithm is used to train the data samples and predict the water inflow; 70% of the data samples are randomly selected as the training set, and 30% of the data are selected as the test set. The water supply, the water inflow at the previous moment, and the liquid level of the clean water tank in the data samples are used as feature columns, and the water inflow is used as the target variable. The test set is predicted and the results are output; for example, if the water inflow at 12:00 is predicted, the water supply at 12:00, the liquid level of the clean water tank, and the water inflow at 11:45 are used as feature columns and input into the model;
[0043] S6, performs optimization processing based on the water inflow predicted in S5, and outputs the optimized predicted water inflow; specifically, checks whether the data exceeds the set water inflow threshold, optimizes and adjusts the frequently changing water inflow to step data, and outputs the optimized water inflow time series data. Embodiment 2
[0044] This embodiment specifically describes S6. Specifically, S6 includes the following steps:
[0045] S61: Check whether the water inflow value predicted in S5 exceeds the highest threshold or the lowest threshold, and perform a backward summation process on the part exceeding the threshold or the part below the threshold. After the summation process is completed, perform a check and summation process in a loop until there is no situation where the value exceeds the threshold or is below the threshold. It should be noted that if the last value still exceeds the threshold or is below the threshold, it is necessary to increase the data of the predicted water inflow;
[0046] S62: Calculate the average value of the predicted water inflow after the addition process in step S61, and mark the predicted water inflow values in S5 differently based on the average value of the water inflow, mark the data below the average value, and mark the data above the average value differently;
[0047] S63: Merge points with the same mark, and when the continuous time of data is less than the set time, merge the mark to the mark of the previous data segment or the next data segment;
[0048] S64: Based on the marks in S63, the average value of the same segment and the same mark is calculated;
[0049] S65: Output the predicted water inflow after optimization processing. Embodiment 3
[0050] This embodiment provides a specific application scenario:
[0051] S1, extracting the design data of the water plant from the CAD design drawing of the water plant, the extracted design data includes the process flow, the size of each building (including the length, width, height and the elevation of the bottom of the building), the size of each pipe section (including the diameter of each pipe section and the elevation of each pipe section), and the parameters of each water pump; in this embodiment, the water inflow, water supply, flow rate of each process section, liquid level of each water tank, switch status and frequency data of each water pump monitored for 3 days in the water plant are collected, and the data interval is 1 minute; the historical monitoring data is preprocessed, and after preprocessing, the data of a fixed time period of 1 minute is taken as a sample;
[0052] S2, using SWMM to build a waterworks hydrodynamic model based on the design data in S1 and the preprocessed historical monitoring data;
[0053] S21, after the waterworks hydrodynamic model is constructed, the uncertainty parameters are calibrated; in this embodiment, a fast non-dominated sorting genetic algorithm (NSGA-3) is used to calibrate the uncertainty parameters, and two water balance data are used to calibrate the orifice coefficient and orifice height of different structures through NSGA-3. After the model is calibrated, the water inflow and water supply data in S1 are used for simulation to generate the clear water tank level, and the monitoring data and simulation data are analyzed and statistically simulated. In the analysis results, 81.7% of the absolute values of the errors are ≤0.2m, and 100% of the absolute values of the errors are ≤0.3m. The model simulation results and the measured results meet the requirements;
[0054] S3, adjusting the water inflow and water supply extracted in S1, generating multiple data samples. In this embodiment, 2 days of data are selected and processed into 15-minute intervals as one data sample; the water inflow and water supply at the same time are synchronously adjusted according to random parameters, generating 50 sets of 6550 data in total, including the water inflow and water supply;
[0055] S4, running the waterworks hydrodynamic model based on the 50 sets of data generated in S3, generating the clear water tank level in each set of data, and forming a sample including water inflow, water supply and clear water tank level data;
[0056] S5, based on the data samples generated in S4, uses the XGBoost regression algorithm to train and test the data samples; randomly selects 70% of the sample data as the training set and 30% of the data as the test set; the water supply, the water inflow at the last moment and the clear water tank level in the sample are used as feature columns, and the water inflow is used as the target variable; predicts the test set and outputs the results; forms the following Figure 2 The “wavy line” shown by line segment 1 in ;
[0057] S6, based on the prediction result output in S5, the frequently changing water inflow is optimized and adjusted to step data, and the data is checked to see if it exceeds the set water inflow threshold, and the optimized water inflow time series data is output to form a data structure such as Figure 2 The "staircase line" shown by line segment 2 in ; specifically includes the following steps:
[0058] S61: Check whether the water inflow value predicted in S5 exceeds the highest threshold or the lowest threshold, and perform a backward summation process on the part exceeding the threshold or the part below the threshold. After the summation process is completed, perform a check and summation process in a loop until there is no situation where the value exceeds the threshold or is below the threshold. It should be noted that if the last value still exceeds the threshold or is below the threshold, it is necessary to increase the data of the predicted water inflow. In this embodiment, the highest threshold is set to 4500 and the lowest threshold is set to 3000.
[0059] This embodiment takes 5 simple data as an example to illustrate the principle. The amount of data in the actual working process is much larger than this value. For example, when a 15-minute interval is selected as a data sample, 14 data are required for the time period from 01:00 to 04:30, which will not be described in detail here.
[0060] The predicted water inflow values from 01:00 to 04:30 are 3500, 2900, 2800, 3000, and 3100;
[0061] The predicted water inflow values from 04:30 to 16:00 are 3750, 3800, 4500, 4700, and 4000;
[0062] The predicted water inflow values from 16:00 to 02:30 are 2800, 2900, 3000, 3200, and 3600;
[0063] The predicted water inflow values from 02:30 to 23:00 are 3800, 3750, 3700, 3650, and 3600;
[0064] First, check the predicted water inflow value from 01:00 to 04:30. 3500 is neither above the highest threshold nor below the lowest threshold, so no action is required.
[0065] 2900 is lower than the minimum threshold, so we take the minimum threshold of 3000, add the -100 below to the value 2800, and the value 2800 becomes 2700.
[0066] 2700 is lower than the lowest threshold, so we take the lowest threshold 3000, add the -300 below to the value 3000, and the value 3000 becomes 2700;
[0067] 2700 is lower than the minimum threshold, so we take the minimum threshold 3000, add the -300 below to the value 3100, and the value 3100 becomes 2800.
[0068] 2800 is lower than the minimum threshold, so we take the minimum threshold of 3000, add the -200 below to the value of 3750 between 04:30 and 16:00, and the value of 3750 becomes 3550;
[0069] At this time, the predicted water inflow values from 01:00 to 04:30 are 3500, 3000, 3000, 3000, 3000;
[0070] Check the predicted water inflow values from 04:30 to 16:00. 3550, 3800, and 4500 are neither above the highest threshold nor below the lowest threshold, so no action is required.
[0071] 4700 is higher than the highest threshold, so we take the highest threshold of 4500 and add the excess 200 to the value of 4000, so the value of 4000 becomes 4200.
[0072] 4200 is neither higher than the upper threshold nor lower than the lower threshold, so no action is required.
[0073] At this time, the predicted water inflow values from 04:30 to 16:00 are 3550, 3800, 4500, 4500, and 4200;
[0074] Check the predicted water inflow value from 16:00 to 02:30. 2800 is lower than the minimum threshold. Take the minimum threshold of 3000, add the -200 below to the value 2900, and the value 2900 is 2700.
[0075] 2700 is lower than the minimum threshold, so take the minimum threshold 3000, add the -300 below to the value 3000, and the value 3000 is 2700;
[0076] 2700 is lower than the minimum threshold, so take the minimum threshold 3000, add the -300 below to the value 3200, and the value 3200 is 2900;
[0077] 2900 is lower than the minimum threshold, so take the minimum threshold 3000, add the -100 below to the value 3600, and the value 3600 is 3500;
[0078] 3500 is neither higher than the upper threshold nor lower than the lower threshold, so no action is required;
[0079] At this time, the predicted water inflow values from 16:00 to 02:30 are 3000, 3000, 3000, 3000, 3500;
[0080] Check the predicted water inflow values from 02:30 to 23:00. 3800, 3750, 3700, 3650, and 3600 are neither above the highest threshold nor below the lowest threshold, so no action is required.
[0081] At this time, the predicted water inflow values from 02:30 to 23:00 are 3800, 3750, 3700, 3650, and 3600.
[0082] S62: Calculate the average value of the predicted water inflow after the addition process in step S61, and mark the predicted water inflow values in S5 differently based on the average value of the water inflow, mark the data below the average value, and mark the data above the average value differently;
[0083] According to the above data, the average value of the predicted water inflow of S5 is calculated to be: 3502.5; the above data are marked differently with 3502.5 as the boundary, and the data below the average value is marked as 0, and the data above the average value is marked as 1; Figure 2 As shown by line segment 3 in;
[0084] S63: Merge points with the same mark, and when the continuous time of data is less than the set time, merge the mark to the mark of the previous data segment or the next data segment;
[0085] In this case, the mark of the data with a continuous time of less than 3 hours is changed to the mark of the previous data segment or the next data segment; Figure 2 In the time period from 16:00 to 02:30 shown, the value from 16:00 to 00:30 is less than 3502.5 and is marked as 0, but the value from 00:30 to 02:30 is greater than 3502.5 and should be marked as 1, but the continuous time of this period is less than 3 hours, so it is merged into the mark of the previous period of data, that is, merged into the period from 16:00 to 00:30 and uniformly marked as 0;
[0086] S63: Based on the marks in S63, the average value is calculated for the same segment and the same mark;
[0087] In this embodiment: the water inflow values marked as 0 in the section from 01:00 to 04:30 are 3500, 3000, 3000, 3000, 3000, and the average value of the same mark in the same section is 3100;
[0088] The water inflow values for the segment marked as 1 from 04:30 to 16:00 are 3550, 3800, 4500, 4500, and 4200, and the average value for the same segment with the same mark is 4110;
[0089] The water inflow values for the section marked as 0 from 16:00 to 02:30 are 3000, 3000, 3000, 3000, 3500; the average value of the same mark in the same section is 3100;
[0090] The water inflow values of the section marked as 1 from 02:30 to 23:00 are 3800, 3750, 3700, 3650, and 3600; the average value of the same mark in the same section is 3700;
[0091] S65: Based on the average value of S64, the predicted water inflow time series after optimization is output according to the step distribution (such as Figure 2 The “step line” shown by line segment 2).
[0092] This application aims to optimize scheduling and reduce production costs by establishing a hybrid model that couples the water plant hydrodynamic model and the data model. The hybrid model is based on monitoring the water inflow and water supply. By adjusting different water inflows and water supplies, the water plant hydrodynamic model is used to simulate the operation of multiple water plant operation events under different operating conditions, and the clear water tank level heights under different water inflows and water supplies are generated. The operation scenarios fully cover the water plant operation conditions including extreme conditions, and then the data of multiple operating condition operation events are trained and simulated to generate a data model. The water supply and clear water tank level can be input into the data model to predict the water inflow, and The prediction results are optimized to generate a step-wise predicted water inflow; the model is established with the goal of optimizing scheduling and reducing production costs. On the basis of meeting the water supply demand, the smooth operation of each process section of the water plant and the high liquid level operation of the clean water are achieved, the clean water tank maintains a high liquid level, reduces the suction of the pump, reduces the consumption of electricity, reduces production costs, and improves economic benefits; in the data processing stage, this application not only focuses on real-time monitoring data, but also processes the design data of the water plant, comprehensively considers information from more dimensions, rather than relying solely on real-time monitoring data, uses SWMM to construct a hydrodynamic model, and combines it with the XGBoost regression algorithm, which is different in the comprehensiveness and adaptability of the model, and better describes the hydraulic conditions of the entire process section; and this application includes steps for optimizing the prediction results, not only for prediction, but also for optimization and adjustment of the results to meet the actual scheduling needs, which to a certain extent makes up for the problem that common methods are insufficient in predicting special scenarios, and improves adaptability to special situations through more comprehensive data processing and comprehensive model construction.
[0093] The various implementation modes of the present application may be arbitrarily combined, and as long as they do not violate the concept of the present application, they should also be regarded as the contents disclosed in the present application.
[0094] A person skilled in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the disclosure of the embodiments of the present invention (including the claims) is limited to these examples; under the concept of the embodiments of the present invention, the technical features in the above embodiments or different embodiments can also be combined, and there are many other changes in different aspects of the embodiments of the present invention as described above, which are not provided in detail for the sake of simplicity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the protection scope of the embodiments of the present invention.
Claims
1. A method for optimizing the water inflow of a waterworks based on a hybrid model, characterized in that: The steps include: S1: Extract the design data of the water plant and the historical monitoring data of the water plant. The extracted historical monitoring data include water inflow, water supply, flow rate of each process section, liquid level of each water tank, and operation data of each water pump; S2: Construct a waterworks hydrodynamic model based on the design data and historical monitoring data extracted in S1; S3: Synchronously adjust the water inlet and water supply in S1 to generate multiple data samples; S4: Run the waterworks hydrodynamic model based on the data samples generated in S3 to generate the clear water tank level, and form water inflow, water supply and clear water tank level data samples; S5: Based on the data samples generated in S4, training the data samples and predicting the water inflow, wherein the water inflow is time series data; S61: Check whether the water inflow value predicted in S5 exceeds the highest threshold or the lowest threshold, add the part exceeding the threshold or the part below the threshold to the following value for backward summing, and after the summing is completed, repeat the checking and summing process in a loop until there is no situation that the water inflow exceeds the threshold or is below the threshold; S62: Calculate the average value of the predicted water inflow after the addition process in step S61, and mark the predicted water inflow values in S5 differently based on the average value of the water inflow, mark the data below the average value, and mark the data above the average value differently; S63: Merge points with the same mark, and merge the marks of data whose continuous time is less than the set time into the mark of the previous period of data or the next period of data; S64: Based on the mark of S63, the average value is calculated according to the water inflow data with the same mark in the same period; S65: Output the predicted water inflow after optimization based on the average value of S64.
2. The method for optimizing the water intake of a waterworks based on a hybrid model according to claim 1, characterized in that: The extracted waterworks design data include process flow, dimensions of each building, dimensions of each pipe section, and parameters of each water pump.
3. The method for optimizing the water intake of a waterworks based on a hybrid model according to claim 1 is characterized in that: The historical monitoring data extracted in S1 are preprocessed, and a water plant hydrodynamic model is constructed in S2 based on the design data and the preprocessed historical monitoring data.
4. The method for optimizing the water intake of a waterworks based on a hybrid model according to claim 1 is characterized in that: In S2, SWMM is used to construct a water power model of the water plant.
5. The method for optimizing the water intake of a waterworks based on a hybrid model according to claim 1 is characterized in that: S21 is also included between S2 and S3. After the hydrodynamic model of the water plant is constructed, the uncertainty parameters are calibrated. After the model is calibrated, the water inlet and water supply data in S1 are used for simulation to generate the clear water tank level. The historically monitored clear water tank level is analyzed with the simulated clear water tank level data to determine whether the hydrodynamic model of the water plant meets the standards. If so, S3 is executed, otherwise the hydrodynamic model of the water plant is readjusted.
6. The method for optimizing the water intake of a waterworks based on a hybrid model according to claim 5 is characterized in that: When the absolute error between the historically monitored clear water tank level and the simulated clear water tank level is less than 0.3 m, the water plant hydrodynamic model meets the standards.
7. The method for optimizing the water intake of a waterworks based on a hybrid model according to claim 1, characterized in that: In S3: the water inflow and water supply at the same time are synchronously adjusted according to the random parameters to generate multiple data samples.
8. The method for optimizing the water intake of a waterworks based on a hybrid model according to claim 1 is characterized in that: In the data sample generated by S4 in S5, the water supply, the water inflow at the previous moment and the liquid level of the clean water tank are used as feature columns, and the water inflow is used as the target variable.
9. The method for optimizing the water intake of a waterworks based on a hybrid model according to claim 1, characterized in that: The data samples generated in S4 in S5 include a training set and a test set, wherein the training set accounts for 70% of the total data samples and the test set accounts for 30% of the total data samples.
