CN-121974416-A - Dosing control method, system, equipment and medium for water treatment process

CN121974416ACN 121974416 ACN121974416 ACN 121974416ACN-121974416-A

Abstract

The invention provides a dosing control method, a system, equipment and a medium in a water treatment process, wherein the dosing control method comprises the steps of performing discrete wavelet multi-scale decomposition on a standardized input sequence to obtain a low-frequency trend component and a high-frequency detail component; the method comprises the steps of inputting a standardized input sequence into a large language submodel of a pre-trained prediction model to obtain a high-dimensional time sequence feature, inputting the low-frequency trend component, the high-frequency detail component and the high-dimensional time sequence feature into the prediction model to generate a prediction coefficient of the fusion feature, carrying out inverse discrete wavelet inverse transformation recombination processing on the prediction coefficient of the fusion feature to obtain an expected output sequence of water quality parameter data at a future moment, extracting instantaneous dosing flow in the expected output sequence, and carrying out dosing control on a water treatment process according to the extracted instantaneous dosing flow. The invention can promote the expected accurate control of the dosing treatment at the future moment.

Inventors

CHENG HONGBING
ZHOU CHONGDONG
FANG JIAYAO
WANG QIANGBIAO
WANG LEI
LU QIANG
YU YE
Jian Sangui
LUO XUEYU

Assignees

歙县自来水有限公司

Dates

Publication Date: 20260505
Application Date: 20260204

Claims (10)

1. A dosing control method for a water treatment process, comprising: acquiring historical water quality parameter data in the water treatment process, and preprocessing the historical water quality parameter data to obtain a standardized input sequence, wherein the water quality parameter data comprises instantaneous dosing flow; Performing discrete wavelet multi-scale decomposition on the standardized input sequence to obtain a low-frequency trend component reflecting a long-term change rule and a high-frequency detail component representing multi-scale short-term fluctuation; inputting the standardized input sequence into a large language submodel of a pre-trained prediction model, obtaining a prompt word set, and carrying out global feature extraction on the prompt word set to obtain high-dimensional time sequence features; Inputting the low-frequency trend component, the high-frequency detail component and the high-dimensional time sequence feature into a cross attention mechanism of the prediction model to perform feature fusion, so as to obtain fusion features; Nonlinear mapping is carried out on the fusion characteristics through a multi-layer perceptron layer of the prediction model, so as to generate prediction coefficients of the fusion characteristics, wherein the prediction coefficients are expressed as the change trend of the fusion characteristics at future time; and carrying out inverse discrete wavelet transform recombination processing on the prediction coefficient of the fusion characteristic to obtain an expected output sequence of water quality parameter data at a future moment, extracting the instantaneous dosing flow in the expected output sequence, and carrying out dosing control on the water treatment process according to the extracted instantaneous dosing flow.
2. The method for controlling dosing in a water treatment process according to claim 1, wherein said performing a discrete wavelet multi-scale decomposition on said normalized input sequence results in a low frequency trend component reflecting a long-term variation law and a high frequency detail component characterizing multi-scale short-term fluctuations, The low frequency trend component and the high frequency detail component satisfy: ; Wherein, the Representing a one-dimensional discrete wavelet decomposition map, Expressed as final decomposition scale The obtained low-frequency trend component; denoted as the first The high frequency detail components corresponding to the individual decomposition scales, The total resolution scale corresponding to the high-frequency detail component is represented as a constant which is non-zero; Represented as a normalized input sequence.
3. The method for controlling chemical dosing in a water treatment process according to claim 1, wherein the prediction coefficients of the fusion features are subjected to inverse discrete wavelet transform reconstruction processing to obtain an expected output sequence of water quality parameter data at a future time, instantaneous chemical dosing flow in the expected output sequence is extracted, and during chemical dosing control of the water treatment process according to the extracted instantaneous chemical dosing flow, The expected output sequence The method comprises the following steps: , ; Wherein, the Represented as a one-dimensional inverse discrete wavelet decomposition transform, The characteristic fusion is carried out on the characteristic corresponding to the low-frequency trend component and the high-dimensional time sequence characteristic, and the prediction coefficient corresponding to the fusion characteristic is obtained; represented by the first And carrying out feature fusion on the features corresponding to the high-frequency detail components and the high-dimensional time sequence features to obtain prediction coefficients corresponding to the fusion features.
4. The method for controlling dosing in a water treatment process according to claim 1, wherein the inputting the standardized input sequence into a large language submodel of a pre-trained predictive model, obtaining a set of cue words, and performing global feature extraction on the set of cue words to obtain high-dimensional time sequence features, comprises: inputting the standardized input sequence into a large language sub-model of a pre-trained prediction model, and calculating a corresponding trend value by the large language sub-model according to the sequence value of the standardized input sequence; Constructing a prompt word set by the large language submodel according to the sequence value of the standardized input sequence and the corresponding trend value, and carrying out global feature extraction on the prompt word set to generate high-dimensional time sequence features; Wherein the trend value The method comprises the following steps: ; 、 represented as normalized input sequence number 、 The sequence value of the individual time nodes, A total number of time nodes expressed as a normalized input sequence; high-dimensional timing features The method comprises the following steps: , wherein, Denoted as the first The feature vectors of the individual time nodes, Represented as a mapping function of the large language submodel, Represented as Is a parameter set of (2); represented as a set of alert words, Represented as To the point of Is used for the whole sequence of the prompt words, Represented as the total length of the set of hint words.
5. The method of dosing control of a water treatment process according to claim 1, wherein the inputting the low frequency trend component, the high frequency detail component, and the high dimensional time series feature into the cross attention mechanism of the prediction model for feature fusion, to obtain a fusion feature, comprises: Inputting the low-frequency trend component and the high-frequency detail component into a cross attention mechanism of the prediction model, respectively dividing the low-frequency trend component and the high-frequency detail component into data blocks to obtain corresponding low-frequency trend component data blocks and high-frequency detail component data blocks, and extracting features of the low-frequency trend component data blocks and the high-frequency detail component data blocks to obtain low-frequency trend component features and high-frequency detail component features; Feature fusion is carried out on the high-dimensional time sequence features and the low-frequency trend component features through the cross attention mechanism to obtain global trend fusion features, and feature fusion is carried out on the high-dimensional time sequence features and the high-frequency detail component features to obtain local detail fusion features; the global trend fusion feature and the local detail fusion feature meet the following conditions: ; , ; Represented as a high-dimensional timing feature, Represented as low-frequency trend component features, Denoted as the first A high frequency detail component feature; expressed as the total number of high frequency detail component feature correspondences; Expressed as low frequency trend component features The corresponding global trend is fused with the features, Denoted as the first Features of high-frequency detail components The corresponding local detail is fused to the feature, Represented as a mapping function of the cross-attention mechanism.
6. The method of controlling dosing of a water treatment process according to claim 5, wherein the generating the prediction coefficients of the fusion features by nonlinear mapping of the fusion features by the multi-layer perceptron layer of the prediction model comprises: the global trend fusion feature is subjected to nonlinear mapping through a multi-layer perceptron layer of the prediction model, a prediction coefficient of the global trend fusion feature is generated, the local detail fusion feature is subjected to nonlinear mapping, and a prediction coefficient of the local detail fusion feature is generated; The prediction coefficients of the global trend fusion features and the prediction coefficients of the local detail fusion features meet the following conditions: , , ; expressed as global trend fusion features The corresponding prediction coefficient is used to determine the prediction coefficient, Denoted as the first Individual local detail fusion features Corresponding prediction coefficients; expressed as global trend fusion features The corresponding mapping function is used to map the data to the data, Denoted as the first Individual local detail fusion features A corresponding mapping function.
7. The dosing control method of a water treatment process of claim 6, wherein the global trend fusion feature Corresponding mapping function First, the Individual local detail fusion features Corresponding mapping function The method comprises the following steps: , ; Wherein, the And Fusing features for global trends A corresponding matrix of mapping parameters is provided, And Denoted as the first Individual local detail fusion features A corresponding matrix of mapping parameters is provided, Represented as a nonlinear operator.
8. A dosing control system for a water treatment process, comprising: the pretreatment unit is used for obtaining historical water quality parameter data in the water treatment process, and carrying out pretreatment on the historical water quality parameter data to obtain a standardized input sequence, wherein the water quality parameter data comprises instantaneous dosing flow; The decomposition unit is used for carrying out discrete wavelet multi-scale decomposition on the standardized input sequence to obtain a low-frequency trend component reflecting a long-term change rule and a high-frequency detail component representing multi-scale short-term fluctuation; The global feature acquisition unit is used for inputting the standardized input sequence into a large language submodel of a pre-trained prediction model, acquiring a prompt word set, and extracting global features of the prompt word set to obtain high-dimensional time sequence features; the fusion feature acquisition unit is used for inputting the low-frequency trend component, the high-frequency detail component and the high-dimensional time sequence feature into a cross attention mechanism of the prediction model to perform feature fusion so as to obtain fusion features; the prediction coefficient acquisition unit is used for carrying out nonlinear mapping on the fusion characteristics through a multi-layer perceptron layer of the prediction model to generate prediction coefficients of the fusion characteristics, wherein the prediction coefficients are expressed as the change trend of the fusion characteristics at future time; and the recombination unit is used for carrying out inverse discrete wavelet transform recombination processing on the prediction coefficient of the fusion characteristic to obtain an expected output sequence of the water quality parameter data at the future moment, extracting the instantaneous dosing flow in the expected output sequence, and carrying out dosing control on the water treatment process according to the extracted instantaneous dosing flow.
9. An electronic device, the electronic device comprising: One or more processors; Storage means for storing one or more programs which, when executed by the one or more processors, cause the electronic device to implement the dosing control method of a water treatment process according to any one of claims 1 to 7.
10. A computer-readable storage medium, having stored thereon a computer program which, when executed by a processor of a computer, causes the computer to perform the dosing control method of the water treatment process according to any one of claims 1 to 7.

Description

Dosing control method, system, equipment and medium for water treatment process Technical Field The invention relates to the technical field of sewage treatment, in particular to a dosing control method, a dosing control system, dosing control equipment and dosing control medium for a water treatment process. Background In the water treatment process of a water works, coagulation is one of key processes, and colloid, suspended matters and part of soluble pollutants in water can be aggregated and settled by adding coagulant such as polyaluminium chloride (PAC, polyaluminum Chloride), so that the subsequent filtering and disinfection effects are improved. The reasonable control of the PAC dosing amount has important influence on water quality stability, running cost and equipment load. However, existing waterworks still face many challenges in PAC dosing control. The traditional control method mainly depends on the experience of operators or simple rules based on conventional indexes of water quality, is difficult to adapt to rapid fluctuation of raw water quality, and is easy to cause insufficient or excessive dosing, influence the stability of a coagulation process and increase the consumption of chemical agents. In order to improve prediction accuracy, some waterworks try to introduce statistical models or traditional machine learning methods, such as linear regression, support vector machines or shallow neural networks based on feature engineering, and although these methods can predict by using historical data to a certain extent, the nonlinear dynamic feature capturing capability of the time series data is limited, and multi-scale fluctuation is difficult to process, so that the prediction accuracy is still not ideal. Disclosure of Invention The invention provides a dosing control method, a dosing control system, dosing control equipment and a dosing control medium for a water treatment process, and aims to solve the technical problem that the nonlinear dynamic characteristic capturing capability of the water treatment process on time sequence data is limited in the prior art. The invention provides a dosing control method in a water treatment process, which comprises the following steps: acquiring historical water quality parameter data in the water treatment process, and preprocessing the historical water quality parameter data to obtain a standardized input sequence, wherein the water quality parameter data comprises instantaneous dosing flow; Performing discrete wavelet multi-scale decomposition on the standardized input sequence to obtain a low-frequency trend component reflecting a long-term change rule and a high-frequency detail component representing multi-scale short-term fluctuation; inputting the standardized input sequence into a large language submodel of a pre-trained prediction model, obtaining a prompt word set, and carrying out global feature extraction on the prompt word set to obtain high-dimensional time sequence features; Inputting the low-frequency trend component, the high-frequency detail component and the high-dimensional time sequence feature into a cross attention mechanism of the prediction model to perform feature fusion, so as to obtain fusion features; Nonlinear mapping is carried out on the fusion characteristics through a multi-layer perceptron layer of the prediction model, so as to generate prediction coefficients of the fusion characteristics, wherein the prediction coefficients are expressed as the change trend of the fusion characteristics at future time; and carrying out inverse discrete wavelet transform recombination processing on the prediction coefficient of the fusion characteristic to obtain an expected output sequence of water quality parameter data at a future moment, extracting the instantaneous dosing flow in the expected output sequence, and carrying out dosing control on the water treatment process according to the extracted instantaneous dosing flow. In one embodiment of the invention, the normalized input sequence is subjected to discrete wavelet multi-scale decomposition to obtain a low-frequency trend component reflecting a long-term variation rule and a high-frequency detail component representing multi-scale short-term fluctuation, The low frequency trend component and the high frequency detail component satisfy: ; Wherein, the Representing a one-dimensional discrete wavelet decomposition map,Expressed as final decomposition scaleThe obtained low-frequency trend component; denoted as the first The high frequency detail components corresponding to the individual decomposition scales,The total resolution scale corresponding to the high-frequency detail component is represented as a constant which is non-zero; Represented as a normalized input sequence. In one embodiment of the invention, the predicted coefficient of the fusion characteristic is subjected to inverse discrete wavelet transform recombination processing to obtain an expected output se