CN-121978970-A - Intelligent dosing control method and system based on multi-source data fusion and time sequence prediction

CN121978970ACN 121978970 ACN121978970 ACN 121978970ACN-121978970-A

Abstract

The invention relates to the field of intelligent dosing control of water treatment and provides an intelligent dosing control method and system based on multi-source data fusion and time sequence prediction, wherein the method comprises the steps of extracting raw water working conditions, colloid electrical balance and floc form attribute sets, and determining basic dosing proportion; the method comprises the steps of obtaining a dynamic delay constant according to the volume of a sedimentation tank and an instantaneous flow value, generating a causal mapping sample set by combining three attribute sets and a sedimentation tank effluent turbidity signal, generating a dynamic weight factor sequence according to the causal mapping sample set, generating a process working condition sensitive characteristic set by carrying out point-by-point inner product on the causal mapping sample set, generating a predicted effluent turbidity value based on the process working condition sensitive characteristic set, carrying out reverse game optimizing through an effluent target turbidity, generating a medicament supplementing correction increment, and carrying out amplitude limiting interception to generate a medicament adding control instruction by combining a basic medicament adding proportion. The invention combines the multisource heterogeneous working condition sensing characteristics and the space-time causal alignment logic to construct an intelligent dosing decision and feedback compensation closed-loop mechanism.

Inventors

LIU HAIZHI
LI LI
YANG YANG
FANG NING

Assignees

上海熊猫机械(集团)有限公司

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. The intelligent dosing control method based on multi-source data fusion and time sequence prediction is characterized by comprising the following steps: collecting raw water intake sensor signals, online flowing current instrument electric signals and real-time underwater image streams, respectively extracting into raw water working condition attribute sets, colloid electric property balance attribute sets and floc form attribute sets, and synchronously determining basic dosing proportion; Generating a causal mapping sample set based on the dynamic delay constant, a comprehensive working condition state vector and an acquired turbidity signal of the sedimentation tank, wherein the comprehensive working condition state vector comprises a raw water working condition attribute set, a colloid electric balance attribute set and a floc form attribute set; Generating a dynamic weight factor sequence according to the causal mapping sample set, and performing point-by-point inner product operation on the causal mapping sample set by utilizing the dynamic weight factor sequence to generate a process condition sensitive characteristic set; And generating a predicted effluent turbidity value based on the process condition sensitive characteristic set, performing reverse game optimizing on the water quality deviation between the predicted effluent turbidity value and a preset effluent target turbidity to generate a reagent supplementing and correcting increment, and performing amplitude limiting interception in a physical safety boundary by combining a basic dosing and proportioning quantity to generate a dosing control instruction.
2. The intelligent dosing control method based on multi-source data fusion and time sequence prediction according to claim 1 is characterized in that the raw water working condition attribute set extracting method comprises the steps of extracting raw water working condition attribute sets comprising a raw water turbidity value, a water temperature value and an instantaneous flow value based on collected raw water intake sensor signals; the colloid electric property balance attribute set extracting method includes extracting colloid electric property balance attribute set containing flow current value and hydrogen ion concentration balance index according to the on-line flow current instrument electric signal at the rear end of the mixing tank; generating a binarization characteristic diagram according to an acquired real-time underwater image stream, and generating a floc morphology attribute set comprising a floc fractal dimension, a floc equivalent average diameter and a floc sedimentation rate per unit volume by analyzing the geometric distribution state of a pixel connected domain; the method for confirming the basic dosing proportioning quantity comprises the step of confirming the basic dosing proportioning quantity according to the mapping relation between the raw water turbidity value and the water temperature value in a pre-stored expert experience library.
3. The intelligent dosing control method based on multi-source data fusion and time sequence prediction according to claim 2, wherein the analysis method of the geometric distribution state of the pixel connected domain comprises the following steps: counting the total number of pixel points of each pixel connected domain and the covered window step length in the binarization feature map to respectively obtain a floc projection area and an observation scale feature length, and executing a least square method to determine a fitting residual constant based on the natural logarithmic value of the floc projection area and the natural logarithmic value of the observation scale feature length; carrying out averaging treatment on the equivalent geometric diameters of all pixel connected domains in the binarization feature map to generate a floc equivalent average diameter; Calculating the geometric center coordinates of the same pixel connected domain in two adjacent frames of binarization feature images, defining the geometric center coordinates as specific floc mass centers, and carrying out ratio operation on the basis of dynamic displacement vectors of the specific floc mass centers between the two adjacent frames of binarization feature images and the inter-frame sampling period to obtain the floc sedimentation rate of unit volume.
4. The intelligent dosing control method based on multi-source data fusion and time sequence prediction according to claim 1, wherein the method for constructing the causal mapping sample set comprises the following steps: According to the hydrodynamic balance relation, carrying out time delay calculation on the rated physical volume and the instantaneous flow value of the sedimentation tank to obtain a dynamic delay constant, wherein the dynamic delay constant is the product of the algebraic quotient of the rated physical volume and the instantaneous flow value and a preset hydraulic efficiency coefficient; based on the time of occurrence of the dosing action according to the dynamic delay constant Is a comprehensive working condition state vector sum And obtaining the minimum accumulated regular cost by the sedimentation tank effluent turbidity signals acquired at the moment so as to generate a causal mapping sample set.
5. The intelligent dosing control method based on multi-source data fusion and timing prediction according to claim 4, wherein the method for obtaining the minimum cumulative regular cost comprises: determining the width of a time domain search window according to the turbulence intensity of water flow in the sedimentation tank and the hydraulic retention time variance generated in the mud discharge period, and locking the time domain search window by taking a dynamic delay constant as a central offset; Calculating the deviation degree between the comprehensive working condition state vector and the effluent turbidity signal of the sedimentation tank in the time domain search window by utilizing a local distance measure operator to obtain a deviation degree sequence; Performing accumulated summation on the deviation degree sequence by utilizing a parameter optimizing operator, and performing extremum retrieval to lock a corresponding time point when the accumulated summation is minimum as an optimal matching site; And extracting the cumulative summation at the best matching site, and resolving to generate the minimum cumulative regular cost.
6. The intelligent dosing control method based on multi-source data fusion and time sequence prediction according to claim 1, wherein the generating method of the process condition sensitive feature set comprises the following steps: calculating the contribution measure score of each characteristic component in the causal mapping sample set, and generating a dynamic weight factor sequence based on the analysis of the distribution proportion of the contribution measure score in the causal mapping sample set, wherein the characteristic components comprise raw water turbidity values, flowing current values and floc fractal dimensions; And carrying out point-to-point inner product operation on the causal mapping sample set by utilizing a Hadamard product operator according to the dynamic weight factor sequence, and analyzing and generating a process condition sensitive characteristic set.
7. The intelligent dosing control method based on multi-source data fusion and timing prediction of claim 6, wherein the calculation method of the contribution measure score comprises: Constructing a multidimensional working condition characteristic interaction space, and projecting a raw water turbidity value, a flowing current value and a floc fractal dimension in a causal mapping sample set into the multidimensional working condition characteristic interaction space; Measuring geometrical projection coincidence degree between each characteristic dimension in the multidimensional working condition characteristic interaction space and a fluctuation deviation sequence of a sedimentation tank effluent turbidity signal relative to a preset ideal control target by utilizing dot product similarity, and generating a causal coupling strength score; And executing the contribution index quantification processing on the causal coupling strength scores by using the causal gain mapping function, and analyzing to obtain the contribution measure scores of the feature dimensions.
8. The intelligent dosing control method based on multi-source data fusion and timing prediction of claim 1, wherein the method for performing reverse game optimization comprises: executing strategy gradient search optimization based on generalized advantage estimation, and iteratively adjusting the medicament adding action control component and calculating a corresponding reward function until a convergence solution enabling the reward function to obtain a maximum value is searched; The rewarding function is obtained by summing three constraint items and taking a negative value, wherein the three constraint items respectively comprise a first item which is the product of a water quality deviation square item and a preset water quality deviation punishment weight coefficient, a second item which is the product of a medicament addition cost function and a preset medicament consumption cost punishment weight coefficient, and a third item which is the product of a medicament addition correction increment absolute value, a preset stability constraint coefficient and a dynamic weight factor mean component; The medicament addition cost function takes medicament addition correction increment as an independent variable, and the dynamic weight factor mean value is obtained by performing arithmetic average operation on a dynamic weight factor sequence; And locking the optimal addition amount according to the convergence solution of the maximum value, and analyzing to generate a medicament addition correction increment.
9. The intelligent dosing control method based on multi-source data fusion and time sequence prediction according to claim 1, wherein the generating method of the dosing control instruction comprises the following steps: Inputting the process working condition sensitive feature set into a long-short-term memory network, carrying out phase calibration by combining a dynamic delay constant, and analyzing to generate a predicted water turbidity value; and taking the water quality deviation between the predicted effluent turbidity value and the preset effluent target turbidity as control excitation, carrying out reverse game optimizing by combining dynamic weight factors, and analyzing to generate a reagent supplementing and correcting increment.
10. An intelligent dosing control system based on multi-source data fusion and time sequence prediction, which is used for realizing the intelligent dosing control method based on multi-source data fusion and time sequence prediction as claimed in any one of claims 1 to 9, and is characterized in that the system comprises a multi-source sensing module, a space-time causal alignment module, a sensitive characteristic sensing module and an intelligent decision control module: the multi-source sensing module is used for collecting raw water intake sensor signals, online flowing current meter electric signals and real-time underwater image streams, respectively extracting raw water working condition attribute sets, colloid electric property balance attribute sets and floc form attribute sets, and synchronously determining basic dosing proportion; The space-time causal alignment module is used for generating a causal mapping sample set according to the rated physical volume of the sedimentation tank and the instantaneous flow value in the raw water working condition attribute set and based on the dynamic delay constant, the comprehensive working condition state vector and the collected sedimentation tank effluent turbidity signal, wherein the comprehensive working condition state vector comprises a raw water working condition attribute set, a colloid electric property balance attribute set and a floc morphology attribute set; the sensitive characteristic sensing module is used for generating a dynamic weight factor sequence according to the causal mapping sample set, and performing point-by-point inner product operation on the causal mapping sample set by utilizing the dynamic weight factor sequence to generate a process working condition sensitive characteristic set; The intelligent decision control module generates a predicted water turbidity value based on a process condition sensitive characteristic set, generates a medicament addition correction increment by performing reverse game optimizing on a water quality deviation between the predicted water turbidity value and a preset water outlet target turbidity, and generates a dosing control instruction by performing amplitude limiting interception in a physical safety boundary in combination with a basic dosing proportion.

Description

Intelligent dosing control method and system based on multi-source data fusion and time sequence prediction Technical Field The invention relates to the field of intelligent dosing control of water treatment, in particular to an intelligent dosing control method and system based on multi-source data fusion and time sequence prediction. Background Along with the deep promotion of intelligent water purification plant construction and the continuous promotion of water supply safety standards, the control of the coagulation and dosing process has become a key technology for guaranteeing the terminal water outlet compliance and the water plant operation energy efficiency. In a multivariable environment with severe fluctuation of raw water physical and chemical properties, real-time dynamic optimizing of the dosing amount of the medicament is realized through a sensing microscopic reaction mechanism, so that the problems of insufficient recognition precision, delayed physical response and the like caused by the fact that a traditional system depends on a single macroscopic index are effectively solved, and the technical problem to be solved in the process of realizing intelligent control of the water treatment dosing process is solved. The Chinese patent application with the publication number of CN120736654B provides a dosing control system in the water treatment process, which comprises a programmable logic controller, a camera, an edge gateway and a programmable logic controller, wherein the programmable logic controller is used for transmitting current characteristic values of water inlet parameter characteristics and water outlet parameter characteristics acquired by a sensor to the edge gateway and driving the camera to shoot a floc image, the camera is used for transmitting the shot floc image to the edge gateway, the edge gateway is used for obtaining a dosing predicted value based on a preset dosing predicted model according to the water inlet parameter characteristics, the water outlet parameter characteristics and the floc image and feeding the obtained dosing predicted value back to the programmable logic controller, and the programmable logic controller is also used for controlling a dosing pump to execute dosing operation in the water treatment process according to the dosing predicted value. However, the current technology still faces many challenges. Under the low-temperature and low-turbidity running condition, the raw water turbidity index is easy to represent a characterization distortion phenomenon, and even if the turbidity value is relatively stable, the colloid electrical state and the chemical dosage of the raw water turbidity index can still be obviously changed. The traditional control mode mainly depends on the linear relation between the turbidity and the flow of raw water, so that the microscopic instability state at the initial stage of the coagulation reaction is difficult to reflect in time, and the observability of the dosing process is lacking. Because of the physical lag of about 2-4 hours between the dosing and treatment effects, when the turbidity of the effluent is monitored to be out of standard, a large amount of water treatment is often not up to standard, and an effective regulation window is missed. Under the condition, if the addition amount is increased to compensate, the addition excess is easy to cause, so that the concentration of aluminum and iron ions in the effluent is increased, the running cost is increased, and the risk is also formed for the safety of water supply. Disclosure of Invention In order to achieve the above purpose, the invention provides an intelligent dosing control method based on multi-source data fusion and time sequence prediction, which comprises the following specific technical scheme: collecting raw water intake sensor signals, online flowing current instrument electric signals and real-time underwater image streams, respectively extracting into raw water working condition attribute sets, colloid electric property balance attribute sets and floc form attribute sets, and synchronously determining basic dosing proportion; Generating a causal mapping sample set based on the dynamic delay constant, a comprehensive working condition state vector and an acquired turbidity signal of the sedimentation tank, wherein the comprehensive working condition state vector comprises a raw water working condition attribute set, a colloid electric balance attribute set and a floc form attribute set; Generating a dynamic weight factor sequence according to the causal mapping sample set, and performing point-by-point inner product operation on the causal mapping sample set by utilizing the dynamic weight factor sequence to generate a process condition sensitive characteristic set; And generating a predicted effluent turbidity value based on the process condition sensitive characteristic set, performing reverse game optimizing on the water quality deviation betw