Intelligent dosing control system based on multivariate information

Abstract

The invention discloses an intelligent dosing control system based on multivariate information, and relates to the technical field of intelligent dosing, and the system comprises a multivariate fusion module which collects real-time water quality parameters, process operation parameters, external environment parameters and historical operation parameters, fuses the parameters into multivariate information data, carries out the time reference synchronization of the multivariate information data, and generates a dynamic heterogeneous data set; and the execution optimization module is used for executing the control strategy switching instruction to drive a chemical adding execution mechanism, and optimizing a stability quantitative index and an anti-fact deduction decision model according to the change trend of the actual water outlet parameters. According to the method, the main control mechanism and the shadow mechanism which run in parallel are constructed, the water quality sudden change working condition is simulated in the virtual environment, and the regulation and control performance deviation of the two control mechanisms on the water quality parameters is quantitatively compared in real time.

Description

Technical Field

[0001] This invention relates to the field of intelligent dosing technology, and in particular to an intelligent dosing control system based on multi-source information. Background Technology

[0002] In recent years, the field of industrial process control has gradually evolved towards multi-source information fusion and intelligent decision-making. Traditional chemical dosing control mainly relies on closed-loop adjustment of water quality feedback parameters, while modern methods are gradually introducing collaborative analysis of process operating parameters, external environmental factors, and historical operating data. Simultaneously, topological data analysis algorithms such as continuous coherence are applied to feature extraction of process states, while virtual simulation technology supports the pre-validation of control strategies by constructing digital twin environments. These technological advancements provide a theoretical foundation for the intelligent upgrading of water quality control.

[0003] Existing technologies struggle to achieve dynamic optimization and safe migration of control strategies. Particularly in chemical dosing control scenarios, traditional methods lack real-time early warning capabilities for process state phase changes, and the optimized strategies generated through deduction often rely on direct implementation without establishing reliable virtual verification mechanisms or shadow switching paths. This leads to the risk of water quality fluctuations during control strategy switching, making it impossible to achieve incremental strategy optimization while ensuring operational stability. Summary of the Invention

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

[0005] Therefore, this invention provides a smart dosing control system based on multi-source information to solve the operational stability risk problem caused by the lack of a security verification mechanism during the control strategy migration process.

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

[0007] This invention provides a smart dosing control system based on multi-source information, comprising: a multi-source fusion module, which collects real-time water quality parameters, process operation parameters, external environmental parameters, and historical operating parameters, fuses them into multi-source information data, synchronizes the multi-source information data with a time base, and generates a dynamic heterogeneous dataset; a phase change early warning module, which calculates process state characteristics based on the dynamic heterogeneous dataset using a continuous cohomology algorithm, extracts stability quantification indicators from the process state characteristics, and generates a process phase change early warning command when the evolution rate of the stability quantification indicator exceeds a preset rate threshold; a deductive decision module, which loads a counterfactual deductive decision model in a virtual simulation environment, simulates the execution of dosing control actions that have not been actually implemented based on the process phase change early warning command, calculates the potential water quality improvement benefits, and generates deductive results including benefit ratings; and a migration verification module. Based on the simulation results, dosing control actions that achieve a preset benefit threshold are selected, generating a candidate action set. The overall compliance rate and cost controllability of each dosing control action in the candidate action set are verified in a virtual environment. When preset reliability conditions are met, a transferable control strategy data package is generated. The shadow comparison module loads the transferable control strategy data package into a parallel-running shadow control mechanism, comparing the regulation performance deviation between the shadow control mechanism and the main control mechanism in real time under simulated water quality change conditions. When the regulation performance deviation of the shadow control mechanism is consistently better than that of the main control mechanism and reaches a preset performance threshold, a control strategy switching command is triggered. The execution optimization module executes the control strategy switching command to drive the dosing actuator, and simultaneously optimizes the stability quantification indicators and counterfactual inference decision model based on the actual effluent parameter change trends.

[0008] As a preferred embodiment of the intelligent dosing control system based on multi-source information described in this invention, the specific steps for generating the dynamic heterogeneous dataset are as follows:

[0009] Real-time water quality parameters, process operation parameters, external environmental parameters, and historical operating parameters are collected through the edge gateway protocol conversion layer and integrated into multi-dimensional information data;

[0010] By employing a precise clock protocol and seasonal time series decomposition, time reference synchronization is performed on multi-dimensional information data to generate an aligned time series dataset.

[0011] Tensor dimensionality reduction and fusion are performed on the aligned time series dataset to generate a dynamic heterogeneous dataset.

[0012] As a preferred embodiment of the intelligent dosing control system based on multi-source information described in this invention, the specific steps for generating the process phase change early warning command are as follows:

[0013] The lifetime values ​​of homology strips in the dynamic heterogeneous dataset are extracted by the continuous homology algorithm to form a multidimensional topological vector, which serves as a feature of the process state.

[0014] Select principal component dimensions that characterize stability from the process state characteristics, calculate the Euclidean norm of the principal component dimensions, and output them as a stability quantification index.

[0015] The stability quantification index is subjected to time-series difference to obtain the evolution rate. The evolution rate is compared with a preset rate threshold. When the evolution rate does not exceed the preset rate threshold, the current process state snapshot data is collected, the historical steady-state sample library is updated, and the incremental training of the counterfactual inference decision model is triggered.

[0016] When the evolution rate exceeds the preset rate threshold, an early warning command generation operation is triggered, and a process phase change early warning command is output.

[0017] As a preferred embodiment of the intelligent dosing control system based on multi-source information described in this invention, the specific steps for generating the deduction results including benefit rating are as follows:

[0018] The phase change type label of the process phase change early warning command is parsed. When the phase change type label is step type, a first type of strategy with the dosing pump frequency adjustment as the core is generated. When the label is gradual type, a second type of strategy with the reagent concentration adjustment as the core is generated. The strategy candidate set is output.

[0019] The strategy candidate set is input into the virtual simulation environment, and water quality parameter response curves are generated through a pre-trained discrete particle model, while real-time acquisition of actual sensor data is performed.

[0020] Based on the water quality parameter response curve and actual sensor data, the distribution difference between the actual and simulated environments in the operating parameters is calculated, and a confidence correction coefficient is generated.

[0021] By combining the integral area of ​​the water quality parameter response curve with the confidence correction coefficient, a comprehensive benefit value is generated, and the benefit level is divided according to the preset level range to generate a projection result including benefit rating.

[0022] As a preferred embodiment of the intelligent dosing control system based on multi-source information described in this invention, the dosing control actions include adjusting the flocculant dosage ratio and switching the type of coagulant aid.

[0023] As a preferred embodiment of the intelligent dosing control system based on multi-source information described in this invention, the specific steps for generating the candidate action set are as follows:

[0024] The benefit rating in the analysis and deduction results is analyzed. When the benefit rating reaches the preset benefit threshold, high-value actions are screened and the corresponding drug dosing control actions are extracted.

[0025] Verify the technological feasibility and economic constraints of the dosing control actions, and generate a set of candidate actions.

[0026] As a preferred embodiment of the intelligent dosing control system based on multi-source information described in this invention, the specific steps for generating a transferable control strategy data package are as follows:

[0027] Execute candidate actions in a virtual environment, load dynamic disturbance scenarios, record the water quality parameter response curve and cost variation curve for each candidate action, and generate a disturbance test report;

[0028] The overall compliance rate is obtained from the water quality response curve, and the overall volatility is obtained from the cost change curve. When the overall compliance rate reaches the preset compliance threshold and the overall volatility does not reach the preset volatility threshold, a set of reliability label actions is output.

[0029] Extract disturbance data from the reliability label action set, construct migration rules, encapsulate the disturbance test report, reliability label action set, and migration rules, and generate a migrationable policy data package.

[0030] As a preferred embodiment of the intelligent dosing control system based on multi-source information described in this invention, the real-time comparison of the regulation performance deviation between the shadow control mechanism and the main control mechanism under simulated water quality change conditions involves the following specific steps.

[0031] Extract the dosing control parameters from the reliability tag action set and load them into the parallel-running shadow control mechanism and main control mechanism. At the same time, extract the water quality change scenario parameters from the disturbance test report and generate the operating condition sequence.

[0032] Real-time acquisition of control response data between the shadow control mechanism and the main control mechanism under operating condition sequences, and calculation of the regulation performance deviation.

[0033] In a preferred embodiment of the intelligent dosing control system based on multi-source information described in this invention, the specific steps for triggering the control strategy switching command are as follows:

[0034] Real-time monitoring of the regulation performance deviation between the main control mechanism and the shadow control mechanism; if the regulation performance deviation of the shadow control mechanism is consistently better than that of the main control mechanism, a comprehensive advantage index is generated.

[0035] When the overall advantage index reaches the preset performance threshold, a control strategy switching instruction is generated.

[0036] As a preferred embodiment of the intelligent dosing control system based on multi-source information described in this invention, the specific steps for optimizing the stability quantification index and the counterfactual inference decision model are as follows:

[0037] The system analyzes the dosing control parameters in the control strategy switching command, drives the dosing pump to perform actions according to the preset gain mode, collects the effluent parameters in real time, analyzes the direction of change trends, and dynamically adjusts the stability quantification index.

[0038] By combining stability quantification indicators with the trend of changes in effluent parameters, the counterfactual inference decision model is updated collaboratively.

[0039] The beneficial effects of this invention are as follows: By constructing a parallel-running master control mechanism and shadow mechanism, a sudden change in water quality is simulated in a virtual environment, and the performance deviations of the two control mechanisms in regulating water quality parameters are quantitatively compared in real time. Before triggering the actual strategy switch, the regulation deviation of the shadow mechanism is required to be consistently better than that of the master control mechanism and meet the preset performance threshold, thereby ensuring that the new strategy has robustness and adaptability. At the same time, the frequency of action of the equipment actuators is significantly reduced through gradual strategy migration. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a block diagram of an intelligent dosing control system based on multi-source information.

[0042] Figure 2 A flowchart generated for a dynamic heterogeneous dataset.

[0043] Figure 3 A flowchart for generating process phase change early warning commands.

[0044] Figure 4 A flowchart generated based on the deduction results. Detailed Implementation

[0045] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0046] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0047] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0048] Reference Figures 1-4This is one embodiment of the present invention, which provides a smart dosing control system based on multi-source information, including the following steps:

[0049] The multi-source fusion module collects real-time water quality parameters, process operation parameters, external environmental parameters, and historical operating parameters, merges them into multi-source information data, synchronizes the multi-source information data with a time reference, and generates a dynamic heterogeneous dataset.

[0050] Real-time water quality parameters, process operation parameters, external environmental parameters, and historical operating parameters are collected through the edge gateway protocol conversion layer and integrated into multi-dimensional information data;

[0051] Specifically, the edge gateway protocol conversion layer is connected to the data source of real-time water quality parameters to obtain real-time water quality parameters. The real-time water quality parameters come from water quality monitoring sensors and are collected through the communication interface of the edge gateway protocol conversion layer.

[0052] Connect the edge gateway protocol conversion layer to the data source of process operation parameters, obtain process operation parameters, which come from the process control engine and are collected through the protocol conversion function of the edge gateway protocol conversion layer;

[0053] The edge gateway protocol conversion layer connects to the data source of external environmental parameters to obtain external environmental parameters. These external environmental parameters come from environmental monitoring equipment and are collected through the data reception of the edge gateway protocol conversion layer.

[0054] Connect the edge gateway protocol conversion layer to the data source of historical operating parameters, obtain historical operating parameters from the historical database, and collect them through the data query function of the edge gateway protocol conversion layer;

[0055] The collected real-time water quality parameters, process operation parameters, external environmental parameters, and historical operating parameters are standardized in data format and outlier removal is performed, and then integrated into multi-dimensional information data.

[0056] By employing a precise clock protocol and seasonal time series decomposition, time reference synchronization is performed on multi-dimensional information data to generate an aligned time series dataset.

[0057] Specifically, a precise clock protocol is applied to timestamp all data points in the multivariate information data. Seasonal time series decomposition is applied to the timestamp-synchronized multivariate information data. The LOESS method is used to decompose the seasonal and trend components of each parameter. For the time series of each parameter, the local polynomial of each data point is fitted by the LOESS method to decompose the high-frequency seasonal component and the low-frequency trend component. The timestamps of all data points are adjusted according to the trend component to generate an aligned time series dataset.

[0058] Tensor dimensionality reduction and fusion are performed on the aligned time series dataset to generate a dynamic heterogeneous dataset.

[0059] Specifically, the aligned time series dataset is organized into a three-dimensional tensor structure, which includes time points, parameter types, and parameter values. Principal component analysis, a tensor dimensionality reduction and fusion method, is applied to perform dimensionality reduction on the three-dimensional tensor structure to reduce redundant dimensions and generate a dimensionality-reduced three-dimensional tensor structure, which serves as a dynamic heterogeneous dataset.

[0060] The phase transition early warning module calculates process state characteristics based on a dynamic heterogeneous dataset using a continuous homology algorithm, extracts stability quantification indicators from the process state characteristics, and generates a process phase transition early warning command when the evolution rate of the stability quantification indicators exceeds a preset rate threshold.

[0061] The lifetime values ​​of homology strips in the dynamic heterogeneous dataset are extracted by the continuous homology algorithm to form a multidimensional topological vector, which serves as a feature of the process state.

[0062] Specifically, the dynamic heterogeneous dataset is mapped to a point cloud topology, with each point corresponding to multiple parameters in the multivariate information data. A simple complex is constructed through scale changes, topological features at different scales are identified, homology groups are generated, the generation time and extinction time of each topological feature are recorded, the lifetime values ​​of homology groups are classified according to topological dimensions, and the lifetime value sequences of all topological dimensions are merged to form a multidimensional topological vector, which serves as a process state feature.

[0063] Select principal component dimensions that characterize stability from the process state characteristics, calculate the Euclidean norm of the principal component dimensions, and output them as a stability quantification index.

[0064] Specifically, principal component analysis is performed on the process state characteristics to obtain the variance contribution rate of all dimensions in the process state characteristics. The dimension with the highest variance contribution rate is selected as the principal component dimension to characterize stability.

[0065] Calculate the Euclidean norm of the principal component dimension, and output it as a stability quantification index. The expression is: ; ;

[0066] In the formula, Indicates a quantitative indicator of stability. Represents the hyperbolic tangent function. The principal component dimension index represents the stability. This represents the set of principal component dimensions that characterize stability. This indicates that the summation is performed over all dimensions within the principal component dimension set representing stability. Indicates the first The standard values ​​of principal components that characterize stability. Indicates the first The standard deviation of the principal components characterizing stability. This represents the square of the variance ratio. Represents the persistence entropy function, This represents the lifetime distribution probability vector of the homology bars. Indicates a homology bar index. This indicates that the summation is performed by iterating through all homology bars. Indicates the first The probability value of homology lifetime. Represents the base-2 of the first... The probability value of the homology lifetime.

[0067] It should be noted that, Both the input and output are purely numerical (dimensionless). Summing of dimensionless quantities After taking the root, it remains dimensionless. The square of a dimensionless ratio is still dimensionless. It is dimensionless, and the final output is... It is dimensionless, but we maintain dimensional consistency.

[0068] The stability quantification index is subjected to time-series difference to obtain the evolution rate. The evolution rate is compared with a preset rate threshold. When the evolution rate does not exceed the preset rate threshold, the current process state snapshot data is collected, the historical steady-state sample library is updated, and the incremental training of the counterfactual inference decision model is triggered.

[0069] It should be noted that the preset rate threshold is set based on the maximum volatility of the stability quantification index in the historical steady-state sample library, with an example value of 0.1.

[0070] Specifically, the numerical changes of stability quantification indicators between consecutive time points are obtained, and the evolution rate is derived by combining the time interval. The evolution rate is compared with a preset rate threshold. When the evolution rate does not exceed the preset rate threshold, the dynamic heterogeneous dataset and the corresponding stability quantification indicators at the current moment are obtained to form process state snapshot data. The process state snapshot data is stored in the historical steady-state sample library constructed by continuously capturing the operating data under stable conditions. At the same time, the incremental training task of the counterfactual inference decision model is started.

[0071] When the evolution rate exceeds the preset rate threshold, an early warning command generation operation is triggered, and a process phase change early warning command is output.

[0072] Specifically, when the evolution rate exceeds a preset rate threshold, the sudden increase in the evolution rate is analyzed, marked as a process phase transition type label, and the current process state characteristics, stability quantification indicators, and process phase transition type label are encapsulated to generate a process phase transition early warning instruction.

[0073] The deductive decision-making module loads a counterfactual deductive decision-making model in a virtual simulation environment. Based on the process phase change early warning command, it simulates the execution of dosing control actions that have not been actually put into use and calculates the potential water quality improvement benefits, generating deductive results that include benefit ratings.

[0074] Dosing control actions include adjusting the flocculant dosage ratio and switching the type of coagulant aid;

[0075] It should be noted that the dosing control actions include adjusting the flocculant dosage ratio and switching the type of coagulant aid, which refers to changing the percentage of flocculant in the mixed agent and changing the specific chemical type of the coagulant aid.

[0076] The phase change type label of the process phase change early warning command is parsed. When the phase change type label is step type, a first type of strategy with the dosing pump frequency adjustment as the core is generated. When the label is gradual type, a second type of strategy with the reagent concentration adjustment as the core is generated. The strategy candidate set is output.

[0077] Extract the phase change type label from the process phase change early warning instruction. When the phase change type label includes step type and gradual type, the first type of strategy is generated when the phase change type label is step type. The strategy action is based on the adjustment of the dosing pump frequency.

[0078] When the phase transition type label is gradual, a second type of strategy is generated, with the strategy action centered on adjusting the drug concentration.

[0079] All generated policy actions are aggregated into a policy candidate set, and each policy action contains specific control parameters.

[0080] The strategy candidate set is input into the virtual simulation environment, and water quality parameter response curves are generated through a pre-trained discrete particle model, while real-time acquisition of actual sensor data is performed.

[0081] It should be noted that the pre-training process of the discrete particle model is as follows: historical dynamic heterogeneous datasets and corresponding process state features are extracted from the historical steady-state sample library. The historical dynamic heterogeneous datasets are mapped to the initial state of discrete particles, where each particle represents a water quality parameter or environmental parameter. With the goal of minimizing the mean square error between the simulated water quality parameters and the historical actual sensor data, the dynamic parameters of the discrete particle model are iteratively optimized using the gradient descent method until the deviation rate between the simulated data and the historical data converges to below the preset deviation rate threshold, thus completing the pre-training of the discrete particle model.

[0082] It should be noted that the preset deviation rate threshold is set based on the statistical distribution of water quality parameter measurement errors in the historical steady-state sample database, with an example value of 0.05.

[0083] Input the policy candidate set into the virtual simulation environment, and calculate the water quality parameter response curve using a discrete particle model. The expression is: ;

[0084] In the formula, Indicates time step The corresponding instantaneous water quality parameter response values, Indicates the number of horizontal divisions in the fluid mesh. Indicates the number of vertical divisions in the fluid mesh. This indicates that the summation is performed over all discrete velocity directions. This represents the total number of discrete velocity directions. Indicates the discrete velocity direction index. Indicates from the initial time up to the current moment Accumulate points. Indicates the first The probability density of discrete velocity directions. Indicates the first The particle position vector along the discrete velocity direction. Indicates the first The particle velocity vector with discrete velocity directions. Indicates the time step. Represents the time integral variable, Describing the Euclidean norm, Represents the collision operator. This represents a time integral infinitesimal.

[0085] It should be noted that, Dimensionless Dimensionless, integral term The dimension of s -1 , The dimension is s. The dimension is eliminated by time integration, and the final output water quality parameter response curve is dimensionless, thus maintaining dimension uniformity.

[0086] By iterating through the time steps The corresponding instantaneous water quality parameter response values ​​are used to generate a sequence of water quality parameter values, which are then interpolated to form a response curve.

[0087] Based on the water quality parameter response curves and actual sensor data, the difference in the distribution of operating parameters between the actual and simulated environments is calculated, and a reliability correction coefficient is generated. The expression is as follows: ;

[0088] In the formula, This represents the credibility correction factor. This represents the natural exponential function. Represents the sensitivity coefficient. Indicates the index of operating condition parameters. Represents the set of operating parameters. This indicates that the summation is performed by iterating through all operating parameters. The first, representing the actual sensor data The average value of each working condition parameter. The first element representing the simulated environment The average value of each working condition parameter. The first, representing the actual sensor data The standard deviation of each operating condition parameter.

[0089] It should be noted that, , and They have the same dimensions. Dimensionless accumulation remains dimensionless. Square roots retain their dimensionless nature. These are dimensionless empirical coefficients. The input is dimensionless, and the output is also dimensionless. It is dimensionless, but we maintain dimensional consistency.

[0090] The sensitivity coefficient is derived from historical data statistical optimization and is used to adjust the degree to which the confidence level is sensitive to distribution differences. The example value is 0.5.

[0091] It should be noted that the operating parameters include pH value, temperature, and flow rate.

[0092] By combining the integral area of ​​the water quality parameter response curve with the confidence correction coefficient, a comprehensive benefit value is generated, and the benefit level is divided according to the preset level range to generate a projection result including benefit rating.

[0093] Specifically, the water quality parameter response curve is integrated to generate the area of ​​integration of the water quality parameter response curve as the quantitative value of the water quality improvement benefit. The quantitative value of the water quality improvement benefit is combined with the credibility correction coefficient to obtain the comprehensive benefit value. The benefit level is divided according to the preset level range. The strategy candidate set, water quality parameter response curve, comprehensive benefit value and benefit level label are packaged to generate the inference result containing the benefit rating.

[0094] It should be noted that the preset grade range refers to the pre-defined range of comprehensive return values, which is used to map the comprehensive return value to the return rating level. For example, a comprehensive return value of 0-40 corresponds to a low return level, 40-80 corresponds to a medium return level, and 80-100 corresponds to a high return level.

[0095] A better approach is to load a counterfactual inference decision model into a virtual simulation environment, simulate the execution of dosing control actions that have not yet been implemented, calculate the potential water quality improvement benefits, and generate inference results including benefit ratings. This provides a data-driven quantitative assessment before the strategy is actually implemented. Compared with conventional existing technologies such as strategy implementation that directly relies on water quality feedback closed-loop regulation, virtual simulation avoids the risk of water quality fluctuations caused by strategy defects, reduces unnecessary equipment operation frequency and reagent waste, and provides a reliable basis for screening high-value actions through benefit ratings.

[0096] The migration verification module, based on the simulation results, selects dosing control actions that achieve a preset benefit threshold, generates a candidate action set, and verifies the overall compliance rate and cost controllability of each dosing control action in the candidate action set in a virtual environment. When the preset reliability conditions are met, a transferable control strategy data package is generated.

[0097] It should be noted that the preset reliability conditions include two constraints: the overall compliance rate reaches the preset compliance threshold and the overall volatility does not reach the preset volatility threshold. The overall compliance rate ensures the compliance of the results, while the overall volatility ensures the stability of the process.

[0098] The benefit rating in the analysis and deduction results is analyzed. When the benefit rating reaches the preset benefit threshold, high-value actions are screened and the corresponding drug dosing control actions are extracted.

[0099] Specifically, the system inputs the simulation results and analyzes the benefit ratings in the simulation results, including high-benefit, medium-benefit, and low-benefit levels. When the benefit rating is high-benefit and reaches the preset benefit threshold, it performs high-value action screening and extracts high-benefit drug control actions from the strategy candidate set.

[0100] Verify the technological feasibility and economic constraints of the dosing control actions, and generate a set of candidate actions;

[0101] Specifically, the process of verifying the feasibility of the dosing control actions is carried out, checking whether the dosing control actions are within the operating range of the equipment, verifying economic constraints, confirming whether the cost variation of the dosing control actions is lower than the preset cost threshold, and summarizing the verified dosing control actions to generate a candidate action set.

[0102] When the benefit rating does not reach the preset benefit threshold, the incremental training task of the counterfactual inference decision model is triggered, and the current drug dosing control action is stored in the preset low benefit sample library.

[0103] It should be noted that the preset revenue threshold is set based on the revenue level distribution of high-return actions in the historical training data of the counterfactual inference decision model, and the example value is that the revenue rating must reach the high-return level; the preset cost threshold is set based on the upper limit of drug cost fluctuation in the historical operating parameters, and the example value is 5%.

[0104] Execute candidate actions in a virtual environment, load dynamic disturbance scenarios, record the water quality parameter response curve and cost variation curve for each candidate action, and generate a disturbance test report;

[0105] Specifically, a dynamic disturbance scenario is loaded in a virtual simulation environment, and a simulation operation is performed on each dosing control action in the candidate action set. The water quality parameter value sequence is calculated step by step through the discrete particle model, and the cost value sequence is generated by simultaneously superimposing the reagent consumption. The water quality parameter value sequence and the cost value sequence are interpolated over time to form the water quality parameter response curve and the cost variation curve. The water quality parameter response curve and cost variation curve for each action are recorded, and all recorded water quality parameter response curves and cost variation curves are summarized to generate a disturbance test report.

[0106] The overall compliance rate is obtained from the water quality response curve, and the overall volatility is obtained from the cost change curve. When the overall compliance rate reaches the preset compliance threshold and the overall volatility does not reach the preset volatility threshold, a set of reliability label actions is output.

[0107] It should be noted that the preset compliance threshold is set based on the compliance rate distribution of water quality parameters in historical operating data, with an example value of 95%; the preset fluctuation threshold is set based on the upper limit distribution of cost fluctuations in historical operating data, with an example value of 10%; the preset water quality standard refers to the pre-set water quality parameter requirements, specifically including indicators such as turbidity and pH value.

[0108] Specifically, based on the water quality response curve, the proportion of time during which water quality parameters meet the preset water quality standards is counted as the overall compliance rate. The cost value sequence of the cost variation curve is extracted, and the maximum deviation range of the cost value sequence is analyzed. The percentage of the difference between the highest cost and the lowest cost to the average cost is counted as the overall volatility. When the overall compliance rate reaches the preset compliance threshold and the overall volatility does not reach the preset volatility threshold, the corresponding dosing control action in the candidate action set is marked as a reliability-labeled action set and output.

[0109] Extract disturbance data from the reliability label action set, construct migration rules, encapsulate the disturbance test report, reliability label action set, and migration rules, and generate a migrationable policy data package.

[0110] Specifically, the perturbation data in the reliability label action set is extracted, migration rules are constructed based on the perturbation data, a condition-action logical relationship is formed, and the perturbation test report, reliability label action set and migration rules are encapsulated to generate a migrationable policy data package.

[0111] The shadow comparison module loads the transferable control strategy data package into the parallel-running shadow control mechanism and compares the regulation performance deviation between the shadow control mechanism and the main control mechanism in real time under simulated water quality change conditions. When the regulation performance deviation of the shadow control mechanism is consistently better than that of the main control mechanism and reaches a preset performance threshold, a control strategy switching command is triggered.

[0112] Extract the dosing control parameters from the reliability tag action set and load them into the parallel-running shadow control mechanism and main control mechanism. At the same time, extract the water quality change scenario parameters from the disturbance test report and generate the operating condition sequence.

[0113] Specifically, dosing control parameters are extracted from the reliability tag action set and synchronously loaded into the parallel-running shadow control mechanism and main control mechanism. At the same time, water quality mutation scenario parameters are extracted from the disturbance test report. The water quality mutation scenario parameters include mutation type, magnitude and occurrence time. The water quality mutation scenario parameters are encoded into a structured sequence in chronological order to generate the operating condition sequence.

[0114] Real-time acquisition of control response data between the shadow control mechanism and the main control mechanism under operating condition sequences, and calculation of the regulation performance deviation.

[0115] Specifically, the data output interfaces of the shadow control mechanism and the main control mechanism are connected through the edge gateway protocol conversion layer. The adjustment records of the dosing control parameters of the two at each time point in the working condition sequence are read in real time, generating a control response dataset with timestamps. At the same time, the response delay time of each control action is recorded to form control response data.

[0116] Based on the control response data of the shadow control mechanism and the main control mechanism under the operating condition sequence, the regulation performance deviation is calculated, and the expression is: ;

[0117] In the formula, Indicates the deviation index of regulation performance. Indicates the start time of the operating condition sequence. This represents the time averaging coefficient. Indicates the time from the start of the operating condition sequence up to the current moment Accumulate points. Indicates a dynamic weight term. Indicates the operating condition sensitivity factor. express The severity of the operating conditions at any given time express At any given moment, the main control mechanism controls the response vector. express The target value vector at any given time. This represents the main mechanism control deviation vector. The standard deviation of the target value vector. express, express The moment shadow control mechanism controls the response vector. This represents the shadow mechanism control bias vector. This represents the Euclidean norm.

[0118] It should be noted that, and All are dimensionless. Dimensionless Dimensionless The dimension is s. The dimension of s -1 Time dimension eliminated It is dimensionless, but ultimately maintains dimensional uniformity.

[0119] The operating condition sensitivity factor is derived from the statistical optimization of historical operating condition data and is used to adjust the response strength of the control strategy to sudden changes in operating conditions. The example value is 0.8.

[0120] Real-time monitoring of the regulation performance deviation between the main control mechanism and the shadow control mechanism; if the regulation performance deviation of the shadow control mechanism is consistently better than that of the main control mechanism, a comprehensive advantage index is generated.

[0121] Specifically, within a fixed monitoring period, the number of times the shadow control mechanism's regulatory performance deviation is better than that of the main control mechanism is continuously counted. When the cumulative number of times the advantage reaches a preset advantage threshold, the percentage of the number of times the shadow control mechanism has an advantage relative to the total number of monitoring times is counted as the comprehensive advantage index.

[0122] It should be noted that the preset advantage threshold is set based on the reliability requirements of control strategy switching in historical operating data, and the example value is 80%.

[0123] When the overall advantage index reaches the preset performance threshold, a control strategy switching instruction is generated.

[0124] It should be noted that the preset performance threshold is set based on the success rate requirement in the historical control strategy switching verification record, with an example value of 80%.

[0125] Specifically, when the comprehensive advantage index reaches the preset performance threshold, the dosing control parameters and migration rules in the transferable strategy data package are parsed to generate a control strategy switching instruction.

[0126] A superior approach involves loading the transferable control strategy data package into a shadow control mechanism that runs in parallel with the main control mechanism. Under simulated water quality change conditions, the performance deviations of the two mechanisms are compared in real time. The shadow mechanism is required to consistently outperform the main mechanism in terms of performance deviation and meet a preset performance threshold before a switch is triggered. Compared to conventional direct strategy switching techniques, this approach reduces the risk of strategy migration, reduces hardware costs by reusing the virtual verification environment, and verifies the robustness of the strategy through dynamic disturbance scenarios. This approach fundamentally solves the problem of uncontrolled water quality fluctuations caused by the lack of a safe migration path in traditional technologies.

[0127] The optimization module executes control strategy switching commands to drive the dosing actuator, and optimizes stability quantification indicators and counterfactual inference decision models based on the actual effluent parameter change trends.

[0128] The system analyzes the dosing control parameters in the control strategy switching command, drives the dosing pump to perform actions according to the preset gain mode, collects the effluent parameters in real time, analyzes the direction of change trends, and dynamically adjusts the stability quantification index.

[0129] It should be noted that the preset gain mode refers to a gradual control method that adjusts the dosing pump frequency or the amount of pesticide in stages, such as increasing the dosing pump frequency by 15% in three steps (each step is 5 minutes apart).

[0130] Specifically, the dosing control parameters in the control strategy switching command are parsed, and the dosing pump is driven to perform actions according to the preset gain mode. The preset gain mode adjusts the dosing pump frequency or the dosage of the chemical in stages. The effluent parameters are collected in real time through the edge gateway protocol conversion layer, and the changing trend of the effluent parameters is analyzed. The stability quantification index is dynamically adjusted according to the changing trend. If the changing trend is stable improvement, the weight coefficient of the corresponding parameter in the stability quantification index is reduced. If the changing trend is fluctuating deterioration, the weight coefficient is increased.

[0131] By combining stability quantification indicators with the trend of changes in effluent parameters, the counterfactual inference decision model is updated collaboratively.

[0132] Specifically, by combining the dynamically adjusted stability quantification indicators with the trend direction of effluent parameter changes, the counterfactual inference decision model is updated collaboratively. When the trend direction of effluent parameter changes is stable improvement, the current process state characteristics and corresponding stability quantification indicators are added as new samples to the historical steady-state sample library. When the trend direction is fluctuating deterioration, the current dosing control action is stored in the low-yield sample library, triggering the incremental training task of the counterfactual inference decision model. The updated historical steady-state sample library and low-yield sample library are used to re-optimize the parameters of the counterfactual inference decision model.

[0133] In summary, this invention constructs a parallel-operating master control mechanism and a shadow mechanism to simulate sudden changes in water quality in a virtual environment, and quantifies and compares the performance deviations of the two control mechanisms in regulating water quality parameters in real time. Before triggering the actual strategy switch, the shadow mechanism is required to consistently outperform the master control mechanism and meet a preset performance threshold, thereby ensuring the robustness and adaptability of the new strategy. Simultaneously, the gradual strategy migration significantly reduces the frequency of actions of the equipment's actuators.

[0134] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A smart dosing control system based on multi-source information, characterized in that: include, The multi-source fusion module collects real-time water quality parameters, process operation parameters, external environmental parameters, and historical operating parameters, merges them into multi-source information data, synchronizes the multi-source information data with a time reference, and generates a dynamic heterogeneous dataset. The phase transition early warning module calculates process state characteristics based on a dynamic heterogeneous dataset using a continuous homology algorithm, extracts stability quantification indicators from the process state characteristics, and generates a process phase transition early warning command when the evolution rate of the stability quantification indicators exceeds a preset rate threshold. The deductive decision module loads a counterfactual deductive decision model in a virtual simulation environment. Based on the process phase change early warning command, it simulates the execution of dosing control actions that have not been actually put into use and calculates the potential water quality improvement benefits, generating deductive results that include benefit ratings. The migration verification module, based on the simulation results, selects dosing control actions that achieve a preset benefit threshold, generates a candidate action set, and verifies the overall compliance rate and cost controllability of each dosing control action in the candidate action set in a virtual environment. When the preset reliability conditions are met, a transferable control strategy data package is generated. The shadow comparison module loads the transferable control strategy data package into the parallel-running shadow control mechanism and compares the regulation performance deviation between the shadow control mechanism and the main control mechanism in real time under simulated water quality change conditions. When the regulation performance deviation of the shadow control mechanism is consistently better than that of the main control mechanism and reaches a preset performance threshold, a control strategy switching command is triggered. The optimization module executes control strategy switching commands to drive the dosing actuator, and optimizes stability quantification indicators and counterfactual inference decision models based on the actual effluent parameter change trends.

2. The intelligent dosing control system based on multi-source information as described in claim 1, characterized in that: The specific steps for generating a dynamic heterogeneous dataset are as follows. Real-time water quality parameters, process operation parameters, external environmental parameters, and historical operating parameters are collected through the edge gateway protocol conversion layer and integrated into multi-dimensional information data; By employing a precise clock protocol and seasonal time series decomposition, time reference synchronization is performed on multi-dimensional information data to generate an aligned time series dataset. Tensor dimensionality reduction and fusion are performed on the aligned time series dataset to generate a dynamic heterogeneous dataset.

3. The intelligent dosing control system based on multi-source information as described in claim 2, characterized in that: The specific steps for generating the phase transition early warning command are as follows: The lifetime values ​​of homology strips in the dynamic heterogeneous dataset are extracted by the continuous homology algorithm to form a multidimensional topological vector, which serves as a feature of the process state. Select principal component dimensions that characterize stability from the process state characteristics, calculate the Euclidean norm of the principal component dimensions, and output them as a stability quantification index. The stability quantification index is subjected to time-series difference to obtain the evolution rate. The evolution rate is compared with a preset rate threshold. When the evolution rate does not exceed the preset rate threshold, the current process state snapshot data is collected, the historical steady-state sample library is updated, and the incremental training of the counterfactual inference decision model is triggered. When the evolution rate exceeds the preset rate threshold, an early warning command generation operation is triggered, and a process phase change early warning command is output.

4. The intelligent dosing control system based on multi-source information as described in claim 3, characterized in that: The specific steps for generating the projection results including the return rating are as follows: The phase change type label of the process phase change early warning command is parsed. When the phase change type label is step type, a first type of strategy with the dosing pump frequency adjustment as the core is generated. When the label is gradual type, a second type of strategy with the reagent concentration adjustment as the core is generated. The strategy candidate set is output. The strategy candidate set is input into the virtual simulation environment, and water quality parameter response curves are generated through a pre-trained discrete particle model, while real-time acquisition of actual sensor data is performed. Based on the water quality parameter response curve and actual sensor data, the distribution difference between the actual and simulated environments in the operating parameters is calculated, and a confidence correction coefficient is generated. By combining the integral area of ​​the water quality parameter response curve with the confidence correction coefficient, a comprehensive benefit value is generated. The benefit level is then divided according to the preset level range, and a projection result containing the benefit rating is generated.

5. The intelligent dosing control system based on multi-source information as described in claim 4, characterized in that: The dosing control actions include adjusting the flocculant dosage ratio and switching the type of coagulant aid.

6. The intelligent dosing control system based on multi-source information as described in claim 1, characterized in that: The specific steps for generating the candidate action set are as follows: The benefit rating in the analysis and deduction results is analyzed. When the benefit rating reaches the preset benefit threshold, high-value actions are screened and the corresponding drug dosing control actions are extracted. Verify the technological feasibility and economic constraints of the dosing control actions, and generate a set of candidate actions.

7. The intelligent dosing control system based on multi-source information as described in claim 6, characterized in that: The specific steps for generating the portable control policy data package are as follows: Execute candidate actions in a virtual environment, load dynamic disturbance scenarios, record the water quality parameter response curve and cost variation curve for each candidate action, and generate a disturbance test report; The overall compliance rate is obtained from the water quality response curve, and the overall volatility is obtained from the cost change curve. When the overall compliance rate reaches the preset compliance threshold and the overall volatility does not reach the preset volatility threshold, a set of reliability label actions is output. Extract disturbance data from the reliability label action set, construct migration rules, encapsulate the disturbance test report, reliability label action set, and migration rules, and generate a migrationable policy data package.

8. The intelligent dosing control system based on multi-source information as described in claim 7, characterized in that: The real-time comparison of the regulation performance deviation between the shadow control mechanism and the main control mechanism under simulated water quality abrupt changes involves the following specific steps. Extract the dosing control parameters from the reliability tag action set and load them into the parallel-running shadow control mechanism and main control mechanism. At the same time, extract the water quality change scenario parameters from the disturbance test report and generate the operating condition sequence. Real-time acquisition of control response data between the shadow control mechanism and the main control mechanism under operating condition sequences, and calculation of the regulation performance deviation.

9. The intelligent dosing control system based on multi-source information as described in claim 1, characterized in that: The specific steps for triggering the control strategy switching command are as follows: Real-time monitoring of the regulation performance deviation between the main control mechanism and the shadow control mechanism; if the regulation performance deviation of the shadow control mechanism is consistently better than that of the main control mechanism, a comprehensive advantage index is generated. When the overall advantage index reaches the preset performance threshold, a control strategy switching instruction is generated.

10. The intelligent dosing control system based on multi-source information as described in claim 9, characterized in that: The specific steps for optimizing the stability quantification index and the counterfactual inference decision model are as follows. The system analyzes the dosing control parameters in the control strategy switching command, drives the dosing pump to perform actions according to the preset gain mode, collects the effluent parameters in real time, analyzes the direction of change trends, and dynamically adjusts the stability quantification index. By combining stability quantification indicators with the trend of changes in effluent parameters, the counterfactual inference decision model is updated collaboratively.

Images