Water quality stabilization method for filtration equipment based on multi-parameter fusion control

Through the multi-parameter fusion control method, big data and knowledge distillation technology are used to build a mapping model of water quality and control parameters, optimize the control of filtration equipment, and solve the problem of poor adaptability when water quality changes in the existing technology, and achieve stable and efficient purification of water quality .

CN118820724BActive Publication Date: 2024-12-03NANTONG HAIYING MECHANICAL&ELECTRICAL GRP CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411290935.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-14
Publication Date
2024-12-03
Estimated Expiration
2044-09-14

AI Technical Summary

Technical Problem

When existing filtration equipment faces sudden changes in water quality, it is impossible to make accurate adjustments quickly and accurately, resulting in poor water purification effect and poor adaptability.

Method used

Through the multi-parameter fusion control method, water quality indicators are obtained using water quality monitoring equipment, and a mapping model between the effluent water quality and the control parameters of the filtration equipment is constructed based on the principle of big data and knowledge distillation, and the control parameters are optimized to achieve the minimization of the water quality objective function. This allows real-time adjustment and optimization of filtering equipment control.

Benefits of technology

It improves the flexibility and adaptability of filter equipment control, ensures the stability and purification effect of water quality, can respond to changes in water quality in a timely manner, and improves the quality of water quality purification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118820724B_ABST
    Figure CN118820724B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field related to water quality filtration, and specifically includes a method for stabilizing the water quality of filtering equipment by multi-parameter fusion control, including: interactive target scenes, activating monitoring to obtain water quality indicators; building sample sets based on big data; building a water quality and equipment parameter mapping model; optimizing parameters by knowledge distillation, and outputting solutions for fusion control. The technical problem that only filtering of filtering scenes with fixed water quality conditions, a sudden increase in pollution sources or a sharp change in water quality parameters, a decrease in water purification quality, and poor adaptability of filtering control are solved. The water quality indicator set of incoming water is obtained by interactive target scenes and activating water quality monitoring equipment, and the water quality status is accurately monitored in real time. The mapping model established by the sample control data set is used to more accurately reflect the relationship between water quality and filtering equipment control parameters, and multi-parameter fusion control is realized, the flexibility of filtering equipment control is improved, the adaptability of filtering control is enhanced, and the technical effect of ensuring the ability of filtering equipment to stabilize water quality is ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field related to water quality filtration, and in particular to a method for stabilizing the water quality of a filtration device by multi-parameter fusion control. Background Art

[0002] The core goal of filtration equipment is to effectively remove impurities, pollutants and harmful substances in water to ensure the safety and cleanliness of water quality. However, conventional filtration equipment control methods mainly rely on experience and preset control schemes. However, since the water purification mechanism itself is relatively complex and involves multiple physical and chemical processes, it is difficult for traditional methods to fully and accurately adapt to changing water quality conditions. When faced with sudden changes in water quality, such as a sudden increase in pollution sources or a sharp change in water quality parameters, conventional filtration equipment control methods are often unable to make accurate adjustments quickly, resulting in poor water purification effects.

[0003] In summary, the prior art has technical problems such as filtering only for filtering scenarios with fixed water quality conditions, sudden increase in pollution sources or drastic changes in water quality parameters, reduced water purification quality, and poor adaptability of filtering control. Summary of the invention

[0004] The present application provides a method for stabilizing the water quality of filtering equipment with multi-parameter fusion control, aiming to solve the technical problems in the prior art of only filtering scenarios with fixed water quality conditions, sudden increase in pollution sources or drastic changes in water quality parameters, reduced water purification quality, and poor adaptability of filtration control.

[0005] In view of the above problems, the technical solution to implement this application is:

[0006] The present application provides a method for stabilizing water quality of a filtration device by multi-parameter fusion control, wherein the method comprises: interacting with a target scene, activating a water quality monitoring device to collect and obtain a set of water quality indicators of incoming water;

[0007] According to the incoming water quality indicator set, intrinsic control records and homologous control records are obtained based on big data, and output as a sample control data set;

[0008] Based on the sample control data set, establishing an effluent mapping model between effluent water quality and filtration equipment control parameters;

[0009] Based on the knowledge distillation principle, a water quality objective function is constructed according to the effluent mapping model, and the water quality objective function is used as the cost function to make an optimization decision on the control parameters of the filtration equipment;

[0010] The optimization decision result is output as a filter device control scheme, and based on the filter device control scheme, the filter device of the target scene is integrated and controlled.

[0011] In summary, one or more technical solutions provided in this application solve the technical problems of filtering only for filtering scenarios with fixed water quality conditions, sudden increase in pollution sources or drastic changes in water quality parameters, decreased water purification quality, and poor adaptability of filtration control. By interacting with target scenarios and activating water quality monitoring equipment to obtain a set of incoming water quality indicators, real-time and accurate monitoring of water quality conditions is performed, and a mapping model established using a sample control data set is used to more accurately reflect the relationship between water quality and filtration equipment control parameters, thereby realizing multi-parameter fusion control, improving the control flexibility of filtration equipment, enhancing the adaptability of filtration control, and ensuring the technical effect of the filtration equipment's ability to stabilize water quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic flow chart of a method for stabilizing water quality of a filtration device with multi-parameter fusion control is provided for this application;

[0013] Figure 2 A flow chart of an effluent mapping model between effluent water quality and filtration equipment control parameters in a filtration equipment water quality stabilization method with multi-parameter fusion control is provided for this application. DETAILED DESCRIPTION

[0014] The present application is described in detail below with reference to the accompanying drawings. Figure 1 As shown, the present application provides a method for stabilizing water quality of a filtration device by multi-parameter fusion control, wherein the method comprises:

[0015] S1: Interactive target scenario, activate water quality monitoring equipment to collect and obtain the incoming water quality indicator set; S2: According to the incoming water quality indicator set, obtain intrinsic control records and homologous control records based on big data, and output them as a sample control data set.

[0016] A sensor network is established in the target scene, an interactive target scene, wherein the target scene refers to a specific water treatment environment, such as a water plant, a reservoir or a water supply system; real-time communication is carried out through the interactive target scene to trigger or activate various water quality monitoring sensors installed in the target scene, and the various water quality monitoring sensors in the target scene are used to measure and record various water quality indicators, such as temperature, pH value, turbidity, dissolved oxygen, etc.; water quality index data of incoming water is collected through water quality monitoring equipment to form an incoming water quality index set, and the incoming water quality index set is used to describe the current water quality status of the incoming water.

[0017] Using big data technology, control records related to the current incoming water quality indicator set are retrieved from existing databases or data warehouses, where intrinsic control records refer to control records directly related to the current target scenario, such as the past water quality control history of the scenario; homologous control records refer to control records related to other similar scenarios (homologous scenarios), which may be similar to the current target scenario in terms of geography, environment or operating conditions; the retrieved intrinsic control records and homologous control records are integrated into a sample control data set, which will be used for subsequent analysis and modeling to establish the relationship between the effluent water quality and the control parameters of the filtration equipment.

[0018] S3: Based on the sample control data set, an effluent mapping model between effluent water quality and filtration equipment control parameters is established.

[0019] The sample control data set includes water quality indicators, filter equipment control parameters and corresponding effluent water quality data collected from target scenes and homologous scenes; the sample control data set provides actual cases on adjusting filter equipment control parameters to achieve specific effluent water quality; features (i.e., water quality indicators and control parameters) that have a significant impact on effluent water quality are selected from the sample control data set; the model is trained using the selected features and the corresponding effluent water quality data; the performance of the model is evaluated by cross-validation, leave-one-out or other validation techniques, and after training and validation, a mapping model describing the relationship between effluent water quality and filter equipment control parameters is obtained, and the effluent mapping model between effluent water quality and filter equipment control parameters can predict effluent water quality based on given water quality indicators and control parameters. The effluent mapping model between effluent water quality and filter equipment control parameters is established to provide model support for stable control and optimization of water quality.

[0020] S4: Based on the knowledge distillation principle, a water quality objective function is constructed according to the effluent mapping model, and the water quality objective function is used as a cost function to make an optimization decision on the control parameters of the filtration equipment.

[0021] Knowledge distillation allows the knowledge of a complex model (teacher model) to be transferred to a simpler and more efficient model (student model). The effluent mapping model can be regarded as a teacher model, which contains rich information about the relationship between water quality and control parameters. According to the output of the effluent mapping model, a water quality objective function is constructed. The water quality objective function can quantify the gap between the effluent quality and the expected water quality, and reflect the impact of the adjustment of control parameters on the improvement of water quality. The objective function includes multiple items, such as the deviation of water quality indicators, etc. Multiple items in the objective function can be combined by weighted sum or other methods.

[0022] In order to find the optimal control parameters, it is necessary to use an optimization algorithm, combined with the gradient of the water quality objective function, to iteratively adjust the control parameters to minimize the objective function. Commonly used optimization algorithms include gradient descent, genetic algorithms, etc.; use the optimization algorithm to optimize the control parameters of the filtration equipment; perform multiple iterations, and each iteration adjusts the control parameters based on the feedback of the objective function; the goal of the optimization process is to find a set of control parameters that minimizes the water quality objective function, that is, to achieve the best match between the effluent water quality and the expected water quality.

[0023] After finding the optimal control parameters, these parameters need to be verified to ensure that the optimal control parameters can achieve the expected water quality improvement effect in actual application; if necessary, the optimization process can be further adjusted or refined to improve the performance of the water quality objective function and the accuracy of the optimization results. Through the above steps, the water quality objective function is constructed based on the knowledge distillation principle, and it is used as the cost function to optimize the control parameters of the filtration equipment. The interaction between the effluent mapping model, the water quality objective function and the optimization algorithm is deeply understood to achieve the best water quality control effect.

[0024] S5: Outputting the optimization decision result as a filtering device control scheme, and performing integrated control on the filtering device of the target scene based on the filtering device control scheme.

[0025] After being processed by the optimization algorithm, a set of optimal filter equipment control parameters are obtained. The optimal filter equipment control parameters are obtained by minimizing the water quality objective function, which represents the filter equipment control parameters that should be set under the current water quality conditions in order to achieve the desired effluent water quality; the optimization decision results are converted into specific filter equipment control schemes. The filter equipment control scheme should clearly indicate how each control parameter should be set, and how the control parameters work together on the filter equipment to achieve the expected water quality improvement effect; the filter equipment control scheme is applied to the filter equipment in the target scenario; fusion control means that the control parameters do not act on a certain filter equipment in isolation, but work together as an overall scheme to achieve water quality stability and improvement.

[0026] After implementing the control plan, continuously monitor the effluent water quality and the operating status of the filtration equipment; if it is found that the water quality does not meet the expected standards, or the equipment operating status is abnormal, the control plan should be adjusted in time; regularly evaluate the effectiveness of the filtration equipment control plan, including the degree of water quality improvement, equipment operating efficiency, etc.; based on the evaluation results, make necessary optimizations to the control plan to adapt to factors such as water quality changes and equipment aging, and ultimately ensure that the filtration equipment can operate stably and efficiently, thereby providing effluent water quality that meets the expectations.

[0027] Further, based on the sample control data set, an effluent mapping model between effluent water quality and filtration equipment control parameters is established. Before this, the method of the present application includes:

[0028] Preprocessing the sample control data set; calculating the information entropy and Mahalanobis distance of multiple sample control data after the preprocessing, and performing outlier analysis to eliminate outliers.

[0029] Check the missing values, outliers or inconsistent data in the data set and handle them accordingly; standardize or normalize the data to ensure comparability between different features; if the data set contains time series data, timestamp alignment or interpolation is also required. Furthermore, information entropy is an indicator of the amount of information in a data set, which can help identify uncertainty or complexity in the data; Mahalanobis distance is a distance measure in a multidimensional space that takes into account data covariance, which can better reflect the actual distribution of data than Euclidean distance; the information entropy and Mahalanobis distance of multiple sample control data are calculated for subsequent outlier analysis, that is, to identify outliers that are significantly different from most data.

[0030] Outliers are generally caused by measurement errors, data entry errors or extreme events, and there is a certain probability that they will have an adverse impact on the modeling process; by analyzing information entropy and Mahalanobis distance, outliers that do not conform to the overall distribution of the data set can be identified; once outliers are identified, they should be removed from the data set to ensure the accuracy of subsequent modeling.

[0031] Through the above steps, based on the sample control data set, an effluent mapping model between effluent water quality and filtration equipment control parameters is established. Prior to this, it is necessary to preprocess the sample control data set, calculate information entropy and Mahalanobis distance, and perform outlier analysis to eliminate outliers, which is an important basis for ensuring data quality and modeling accuracy.

[0032] Furthermore, if Figure 2 As shown, based on the sample control data set, a water outlet mapping model between the water quality of the outlet and the control parameters of the filtration equipment is established. Before that, the method of the application also includes:

[0033] Acquire a preceding control data set of the sample control data set; compare the sample control data set with the preceding control data set, and perform variable simplicity analysis based on the category of the control variable; based on the result of the variable simplicity analysis, perform data segmentation on the sample control data set to acquire a heterogeneous sample control data set, wherein the heterogeneous sample control data set includes a medication sample control data group, a mechanical sample control data group, and a fusion sample control data group.

[0034] The previous control data set refers to the relevant data collected before the current sample control data set, including water quality indicators, filtration equipment control parameters and effluent water quality data in an earlier time period; the purpose of obtaining the previous control data set is to conduct time series analysis and understand the changing trends of water quality and control parameters over time; the sample control data set is compared with the previous control data set to analyze the differences and similarities between the sample control data set and the previous control data set, which helps to identify any potential data offsets or pattern changes, which will have a certain impact on the subsequent modeling process.

[0035] Control variables refer to the control parameters of the filtration equipment that can affect the effluent water quality, such as dosage, mechanical operating parameters, etc.; variable singularity analysis refers to analyzing the impact of each control variable on the effluent water quality to determine which variables are primary and which are secondary; it helps to simplify the subsequent modeling process and focus only on the main variables to reduce the complexity of the model; based on the results of the variable singularity analysis, the sample control data set is divided into different groups, each group containing data related to a specific control variable, for example, the data set is divided into a dosing sample control data group, a mechanical sample control data group, and a fused sample control data group.

[0036] The sample control data set refers to a collection of data of different categories or sources. In the embodiment of the present application, it specifically refers to a medication sample control data group, a mechanical sample control data group, and a fusion sample control data group. The data in the medication sample control data group, the mechanical sample control data group, and the fusion sample control data group have different characteristics, distributions, and correlations, and therefore need to be processed and analyzed separately.

[0037] Through the above steps, based on the sample control data set, an effluent mapping model between the effluent water quality and the control parameters of the filtration equipment is established. Prior to this, it is also necessary to obtain the previous control data set, perform data set comparison, variable unity analysis, and data segmentation. These steps will help to gain a deeper understanding of the data characteristics and provide strong support for the subsequent modeling process.

[0038] Furthermore, based on the sample control data set, a water outlet mapping model between the water quality of the outlet and the control parameters of the filtration equipment is established. The method of the present application includes:

[0039] Construct multiple first-class weak classifiers and multiple second-class weak classifiers; call the dosing sample control data group to perform supervised training on multiple first-class weak classifiers, and integrate learning to obtain a dosing mapping layer; use the mechanical sample control data group as training data, supervise the training of multiple second-class weak classifiers, and integrate learning to obtain a mechanical mapping layer; based on the dosing mapping layer and the mechanical mapping layer, construct a forward fusion mapping channel and a reverse fusion mapping channel, and call the fusion sample control data group to jointly train the forward fusion mapping channel and the reverse fusion mapping channel to obtain the water outlet mapping model.

[0040] Based on the sample control data set, an effluent mapping model between effluent water quality and filtration equipment control parameters is established. Specifically, multiple first-class weak classifiers and multiple second-class weak classifiers are constructed, wherein the first-class weak classifier focuses on the relationship between dosing control parameters and effluent water quality, aiming to capture the specific impact of dosing on water quality. The second-class weak classifier focuses on the relationship between mechanical control parameters and effluent water quality, aiming to analyze how mechanical operation affects water quality.

[0041] A dosing mapping layer is obtained. Specifically, a drug sample control data group is used as a training set; multiple first-class weak classifiers are supervised and trained to ensure that the multiple first-class weak classifiers can accurately predict the impact of dosing control parameters on effluent water quality; through integrated learning technology, multiple first-class weak classifiers are fused into a dosing mapping layer, which can comprehensively reflect the complex relationship between dosing and water quality.

[0042] A mechanical mapping layer is obtained. Specifically, a mechanical sample control data group is used as training data; multiple two-category weak classifiers are supervised and trained so that they can accurately predict the impact of mechanical control parameters on effluent water quality; and ensemble learning technology is also used to fuse multiple two-category weak classifiers into a mechanical mapping layer, which can comprehensively reflect the intrinsic connection between mechanical operation and water quality.

[0043] Based on the dosing mapping layer and the mechanical mapping layer, a forward fusion mapping channel and a reverse fusion mapping channel are constructed, wherein the forward fusion mapping channel aims to orderly integrate the information of the dosing mapping layer and the mechanical mapping layer to simulate the sequential effects of dosing and mechanical operations in the actual filtering process; the reverse fusion mapping channel attempts to integrate the two layers of information in reverse order to explore various nonlinear or interactive effects.

[0044] Using the fusion sample control data set as the joint training set, the forward fusion mapping channel and the reverse fusion mapping channel are jointly trained to ensure that the forward fusion mapping channel and the reverse fusion mapping channel can work together and accurately predict the effluent water quality; the effluent mapping model is obtained through training, and the effluent mapping model can fully and accurately reflect the complex mapping relationship between the control parameters of the filtration equipment and the effluent water quality. Weak classifiers are constructed, mapping layers are obtained, fusion mapping channels are constructed and jointly trained to obtain the effluent mapping model, which provides strong support for subsequent filtration equipment control.

[0045] Furthermore, based on the dosing mapping layer and the mechanical mapping layer, a forward fusion mapping channel and a reverse fusion mapping channel are constructed, and the fusion sample control data group is called to jointly train the forward fusion mapping channel and the reverse fusion mapping channel to obtain the water outlet mapping model. The method of the present application includes:

[0046] Call the dosing mapping layer and the mechanical mapping layer, and connect the output end of the dosing mapping layer with the sequence input end of the mechanical mapping layer, connect the input end of the dosing mapping layer with the span input end of the mechanical mapping layer, and construct the forward fusion mapping channel; connect the output end of the mechanical mapping layer with the sequence input end of the dosing mapping layer, connect the input end of the mechanical mapping layer with the span input end of the dosing mapping layer, and construct the reverse fusion mapping channel; connect the forward fusion mapping channel and the reverse fusion mapping channel in parallel to the mean output layer, and construct a joint training function to perform joint training.

[0047] Furthermore, to construct a joint training function, the present application method includes:

[0048] The joint training function is expressed as:

[0049] ; Wherein, S represents the sample output in the heterogeneous sample control data set; f represents the mapping output of the forward fusion mapping channel; b represents the mapping output of the reverse fusion mapping channel; Characterize the channel mapping deviation of the forward fusion mapping; Characterize the channel mapping deviation of the reverse fusion mapping; Characterizes the joint output deviation of the forward fusion mapping channel and the backward fusion mapping channel.

[0050] Based on the dosing mapping layer and the mechanical mapping layer, a forward fusion mapping channel and a reverse fusion mapping channel are constructed. Specifically, a forward fusion mapping channel is constructed: the first connection method is that the output end of the dosing mapping layer is connected to the sequence input end of the mechanical mapping layer, and further, the trained dosing mapping layer and the mechanical mapping layer are called; the output end of the dosing mapping layer is connected to the sequence input end of the mechanical mapping layer, so that the output of the dosing mapping layer can be used as the input of the mechanical mapping layer, simulating the order of first dosing and then mechanical processing in the actual filtering process; at the same time, under the first connection method, the mechanical mapping layer also considers the original input of the dosing mapping layer to obtain more comprehensive information.

[0051] Constructing a reverse fusion mapping channel: The first connection method is to connect the output of the mechanical mapping layer to the sequence input of the dosing mapping layer, and further, call the trained dosing mapping layer and the mechanical mapping layer; similarly, connect the output of the mechanical mapping layer to the sequence input of the dosing mapping layer to simulate the order of mechanical processing first and then dosing. Thus, the output of the mechanical mapping layer can be used as the input of the dosing mapping layer to simulate the order of mechanical processing first and then dosing in the actual filtering process. At the same time, in the second connection method, the dosing mapping layer also takes into account the original input of the mechanical mapping layer to obtain more comprehensive information.

[0052] The forward fusion mapping channel and the reverse fusion mapping channel are connected in parallel to a mean output layer. Furthermore, the forward fusion mapping channel and the reverse fusion mapping channel are connected in parallel to a mean output layer. The function of the mean output layer is to calculate the average value of the outputs of the forward fusion mapping channel and the reverse fusion mapping channel to obtain a more stable and accurate prediction result. A joint training function is constructed. The joint training function takes the fusion sample control data group as input, considers the outputs of the forward and reverse fusion mapping channels at the same time, and adjusts the model parameters through the optimization algorithm to minimize the difference between the predicted effluent water quality and the actual effluent water quality. The joint training function is trained using the fusion sample control data set, and the model parameters are continuously adjusted through the iterative optimization algorithm until the predetermined training target or convergence condition is reached; after the joint training, a water outlet mapping model including a dosing mapping layer, a mechanical mapping layer, a forward fusion mapping channel, a reverse fusion mapping channel, and a mean output layer is obtained, and the water outlet mapping model can comprehensively consider the influence of dosing and mechanical control parameters on the effluent water quality, and capture the complex relationship between dosing control parameters and mechanical control parameters through the forward and reverse fusion mapping channels, thereby providing accurate water quality prediction and control strategies. Constructing the forward and reverse fusion mapping channels and performing joint training to obtain the water outlet mapping model provides strong support for the optimization control of the filtration equipment.

[0053] A joint training function is constructed. On the one hand, the parameters of the forward and backward fusion mapping channels are optimized by minimizing the deviation. On the other hand, the forward fusion mapping channel and the backward fusion mapping channel verify each other to improve the consistency between the forward fusion mapping channel and the backward fusion mapping channel. Further, the parameters of the initialized forward fusion mapping channel and the initialized backward fusion mapping channel, including weights and deviations; forward propagation is performed through the forward fusion mapping channel and the backward fusion mapping channel to obtain the predicted output of the forward fusion mapping channel and the backward fusion mapping channel; the joint training function is used to calculate the total loss, including the prediction deviation and the consistency deviation; the gradient is calculated according to the total loss, and the parameters of the forward and backward fusion mapping channels are updated through back propagation; the above process is repeated continuously to improve the accuracy and stability of the output water mapping model.

[0054] Furthermore, the present application method also includes:

[0055] According to the incoming water quality indicator set, a poor solution sample control data set is obtained based on big data; the poor solution sample control data set is parsed, key poor solution features are extracted, and a key poor solution feature set is obtained; and the penalty factor of the water quality objective function is configured with the key poor solution feature set.

[0056] Based on the water quality index set, with the help of big data, samples that do not match the water quality index set or cause water quality deterioration are screened out from massive historical data to form a deteriorated sample control data set. The deteriorated sample control data set reflects the actual situation of water quality deterioration under specific control parameters. The deteriorated sample control data set is deeply analyzed to identify the control parameters or conditions closely related to water quality deterioration. The control parameters or conditions closely related to water quality deterioration are defined as key deteriorated solution features. Key deteriorated solution features are essential for understanding the mechanism of water quality deterioration and formulating effective control strategies; based on the extracted key deteriorated solution feature set, the water quality objective function is finely configured. Preferably, a penalty factor is set for the control parameters related to the key deteriorated solution feature. The purpose is that during the optimization process, when the key deteriorated solution feature appears, the outlet water mapping model can give a greater penalty, thereby guiding the search algorithm in the direction of avoiding water quality deterioration. The deteriorated sample control data set is obtained by means of big data, and the key deteriorated solution features are extracted from it, and the penalty factor of the water quality objective function is accurately configured, thereby improving the accuracy of water quality prediction and control, and more efficiently handling changes in the water quality of the incoming water.

[0057] Furthermore, the present application method also includes:

[0058] According to the sample control data set, a homologous device set is extracted; based on the homologous device set, multiple homologous devices are connected to share data, wherein the data sharing includes model parameter sharing and control record sharing; and through the data sharing results, the water outlet mapping model is enhanced and trained.

[0059] Based on the sample control data set, devices with the same or similar characteristics are identified. Devices with the same or similar characteristics constitute a homologous device set, which includes devices with the same manufacturer, model, treatment process or operating conditions; multiple homologous devices in the homologous device set are connected to achieve data sharing, which includes model parameter sharing and control record sharing. Model parameter sharing means that the parameters of the effluent mapping model can be shared between homologous devices so as to improve the model by using collective wisdom; control record sharing involves sharing the operating records and control parameters of the equipment to obtain a more comprehensive understanding of the equipment performance and water quality response.

[0060] The data sharing results of the homologous device set can be used to enhance the training of the water outlet mapping model. In short, the shared data is integrated into the model training process to more accurately capture the complex relationship between device performance and water quality; enhanced training involves using more training data, more complex model structures, or more advanced optimization algorithms to improve the prediction accuracy and generalization ability of the model. By extracting the homologous device set, connecting multiple homologous devices for data sharing, and enhancing the training of the water outlet mapping model, the performance of the water outlet mapping model is further improved, and the challenges in actual water quality control are more efficiently addressed, providing a more powerful tool for water quality prediction and control.

[0061] In summary, the beneficial effects of the embodiments of the present application are:

[0062] 1. Through real-time monitoring of water quality indicators, timely respond to water quality changes and automatically adjust control strategies.

[0063] 2. The machine learning-based mapping model, combined with big data analysis and intelligent algorithms, can quickly derive effective control parameters and more accurately predict and regulate effluent water quality.

[0064] 3. Due to the use of joint training function, it is expressed as:

[0065] ; Wherein, S represents the sample output in the heterogeneous sample control data set; f represents the mapping output of the forward fusion mapping channel; b represents the mapping output of the reverse fusion mapping channel; Characterize the channel mapping deviation of the forward fusion mapping; Characterize the channel mapping deviation of the reverse fusion mapping; Characterize the joint output deviation of the forward fusion mapping channel and the reverse fusion mapping channel. Construct a joint training function. On the one hand, optimize the parameters of the forward and reverse fusion mapping channels by minimizing the deviation. On the other hand, the forward fusion mapping channel and the reverse fusion mapping channel verify each other, improve the consistency between the forward fusion mapping channel and the reverse fusion mapping channel, and thus improve the accuracy and stability of the output water mapping model.

[0066] In summary, any step can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor, and no unnecessary restrictions are made here.

[0067] Furthermore, the above-mentioned technical scheme only reflects the preferred technical scheme of the technical scheme of the embodiment of the present application. Some changes that may be made to some parts thereof by technicians in this technical field all reflect the novel principles of the embodiment of the present application. Obviously, technicians in this field can make various changes and modifications to the present application without departing from the scope of the present application.

Claims

1. A method for stabilizing water quality of a filtration device using multi-parameter fusion control, characterized in that: The method comprises: Interactive target scene, activate water quality monitoring equipment to collect and obtain the incoming water quality indicator set; According to the incoming water quality indicator set, intrinsic control records and homologous control records are obtained based on big data, and output as a sample control data set; Based on the sample control data set, establishing an effluent mapping model between effluent water quality and filtration equipment control parameters; Based on the knowledge distillation principle, a water quality objective function is constructed according to the effluent mapping model, and the water quality objective function is used as the cost function to make an optimization decision on the control parameters of the filtration equipment, wherein knowledge distillation allows the knowledge of the teacher model to be transferred to the student model, and the effluent mapping model is regarded as the teacher model, which contains rich information about the relationship between water quality and control parameters. According to the output of the effluent mapping model, a water quality objective function is constructed; Outputting the optimization decision result as a filter device control scheme, and performing integrated control on the filter device of the target scene based on the filter device control scheme; Wherein, based on the sample control data set, an effluent mapping model between effluent water quality and filtration equipment control parameters is established, before which, it also includes: Acquire a preceding control data set of the sample control data set; Comparing the sample control data set with the previous control data set, and performing variable homogeneity analysis based on the categories of control variables; Based on the result of the variable singularity analysis, the sample control data set is segmented to obtain a heterogeneous sample control data set, wherein the heterogeneous sample control data set includes a medication sample control data group, a mechanical sample control data group, and a fusion sample control data group; Wherein, based on the sample control data set, an effluent mapping model between effluent water quality and filtration equipment control parameters is established, including: Construct multiple first-class weak classifiers and multiple second-class weak classifiers; Calling the medication sample control data group, performing supervised training of a plurality of the first-class weak classifiers, and integrating learning to obtain a medication mapping layer; Using the mechanical sample control data group as training data, supervised training of a plurality of the two-class weak classifiers, and integrated learning to obtain a mechanical mapping layer; Based on the dosing mapping layer and the mechanical mapping layer, a forward fusion mapping channel and a reverse fusion mapping channel are constructed, and the fusion sample control data group is called to jointly train the forward fusion mapping channel and the reverse fusion mapping channel to obtain the water outlet mapping model.

2. The method for stabilizing water quality of filtering equipment by multi-parameter fusion control as claimed in claim 1, characterized in that: Based on the sample control data set, an effluent mapping model between effluent water quality and filtration equipment control parameters is established, including: Preprocessing the sample control data set; For the preprocessed sample control data set, the information entropy and Mahalanobis distance of multiple sample control data are calculated and outlier analysis is performed to eliminate outliers.

3. The method for stabilizing water quality of filtering equipment by multi-parameter fusion control as claimed in claim 1, characterized in that: Based on the dosing mapping layer and the mechanical mapping layer, a forward fusion mapping channel and a reverse fusion mapping channel are constructed, and the fusion sample control data group is called to jointly train the forward fusion mapping channel and the reverse fusion mapping channel to obtain the water outlet mapping model, including: Calling the drug administration mapping layer and the mechanical mapping layer, connecting the output end of the drug administration mapping layer with the sequence input end of the mechanical mapping layer, connecting the input end of the drug administration mapping layer with the span input end of the mechanical mapping layer, and constructing the forward fusion mapping channel; Connecting the output end of the mechanical mapping layer with the sequence input end of the drug administration mapping layer, connecting the input end of the mechanical mapping layer with the span input end of the drug administration mapping layer, and constructing the reverse fusion mapping channel; The forward fusion mapping channel and the reverse fusion mapping channel are connected in parallel to the mean output layer, and a joint training function is constructed to perform joint training.

4. The method for stabilizing water quality of filtering equipment by multi-parameter fusion control as claimed in claim 3, characterized in that: Construct a joint training function, including: The joint training function is expressed as: ; Wherein, S represents the sample output in the heterogeneous sample control data set; f represents the mapping output of the forward fusion mapping channel; b represents the mapping output of the reverse fusion mapping channel; Characterize the channel mapping deviation of the forward fusion mapping; Characterize the channel mapping deviation of the reverse fusion mapping; Characterizes the joint output deviation of the forward fusion mapping channel and the backward fusion mapping channel.

5. The method for stabilizing water quality of filtering equipment by multi-parameter fusion control as claimed in claim 1, characterized in that: The method further comprises: According to the incoming water quality indicator set, a control data set of inferior solution samples is obtained based on big data; Parsing the inferior solution sample control data set, extracting key inferior solution features, and obtaining a key inferior solution feature set; The penalty factor of the water quality objective function is configured based on the key inferior solution feature set.

6. The method for stabilizing water quality of filtering equipment by multi-parameter fusion control as claimed in claim 1, characterized in that: The method further comprises: Extracting a homologous device set according to the sample control data set; Based on the homologous device set, multiple homologous devices are connected to share data, wherein the data sharing includes model parameter sharing and control record sharing; The water outlet mapping model is enhanced by training the data sharing results.

Citation Information

Patent Citations

  • Water quality monitoring management method and device, server and readable storage medium

    CN111539477A

  • Fish pond water quality evaluation and prediction system and method based on fuzzy evaluation and improved support vector machine

    CN114357877A

  • Water quality state fusion perception and prediction traceability method

    CN118070234A

  • Method and device for constructing water quality and fish behavior bidirectional mapping model

    CN118379788A

  • GIS risk management and control system and method for pollutant migration in mining area basin

    WO2024148683A1