CN-122022557-A - Water quality extremum prediction method based on gating residual error network and empirical mode decomposition

CN122022557ACN 122022557 ACN122022557 ACN 122022557ACN-122022557-A

Abstract

The application belongs to the technical field of water quality prediction, and particularly discloses a water quality extremum prediction method based on a gate control residual error network and empirical mode decomposition. According to the method, an original water quality data sequence is generated according to the historical water quality data, after backbone network parameters of a water quality prediction base model are frozen, other network parts are trained according to a residual network and a gating mechanism of a full-connection network block, and extremum prediction is carried out on water quality data of sudden water pollution based on a target water quality extremum prediction model. By the method, a residual network and a gating mechanism of the full-connection network block are respectively constructed, the extremum is predicted by the correction model, other network parts except for freezing in the water quality prediction base model are trained, so that model training efficiency is improved, at the moment, the target water quality extremum prediction model captures the water quality fluctuation law of sudden water pollution, the extremum of the water quality data of the sudden water pollution can be predicted, and therefore efficiency and accuracy of predicting the water quality extremum can be effectively improved.

Inventors

CHEN JIAHE
Shu Yaorong
LIU HUI
MAO JUAN
WU XIAOHUI

Assignees

华中科技大学

Dates

Publication Date: 20260512
Application Date: 20260107

Claims (10)

1. A water quality extremum prediction method based on a gating residual network and empirical mode decomposition is characterized by comprising the following steps: Generating an original water quality data sequence according to the historical water quality data, and respectively constructing a water quality prediction base model, a residual error network and a gating mechanism of a full-connection network block according to the original water quality data sequence; After backbone network parameters of the water quality prediction base model are frozen, training other network parts of the water quality prediction base model according to the residual network and a gating mechanism of the fully-connected network block to obtain a target water quality extremum prediction model; And carrying out extremum prediction on the water quality data of sudden water pollution based on the target water quality extremum prediction model.
2. The method of claim 1, wherein the step of generating the raw water quality data sequence from the historical water quality data comprises: performing missing detection on the historical water quality data, and performing primary linear interpolation on the detected missing segments; Decomposing the history water quality data after primary interpolation based on a seasonal trend decomposition algorithm to obtain a current trend item and a current residual error item; Performing secondary linear interpolation on the historical water quality data, and decomposing the historical water quality data after secondary interpolation based on the seasonal trend decomposition algorithm to obtain a target trend item and a target residual error item; replacing the current trend item according to the target trend item, and replacing the current residual item according to the target residual item; And reorganizing the replaced trend item, the replaced residual item and the reserved season item to generate an original water quality data sequence.
3. The method of claim 1, wherein the step of constructing a water quality prediction basis model, a residual network, and a gating mechanism for fully connected network blocks from the original water quality data sequence, respectively, comprises: Slicing the original water quality data sequence according to a preset input window and a preset step length, and determining target sequence characteristics according to a sequence slicing result; constructing a water quality prediction base model by taking the target sequence characteristics as input and the concentration at the next moment as output; inputting an original water quality data sequence into the water quality prediction base model, and obtaining a predicted value output by the water quality prediction base model; and respectively constructing a residual network and a gating mechanism of the full-connection network block according to the predicted value and the original water quality data sequence.
4. A method according to claim 3, wherein the step of constructing a residual network and a gating mechanism for a fully connected network block from the predicted value and the original water quality data sequence, respectively, comprises: Calculating a difference between the predicted value and the true value; decomposing the original water quality data sequence based on an empirical mode decomposition algorithm to obtain a plurality of eigenvalue function components and trend remainder; constructing a residual error network by taking the intrinsic mode function components, the trend remainder and the original water quality data sequence as inputs and taking the difference as output; And calculating the forward difference of the original water quality data sequence based on a target difference algorithm, and constructing a gating mechanism of the fully-connected network block by taking the forward difference as input and a target gating value as output.
5. The method according to any one of claims 1 to 4, wherein after the backbone network parameters of the water quality prediction base model are frozen, training other network parts of the water quality prediction base model according to the residual network and the gating mechanism of the fully-connected network block to obtain a target water quality extremum prediction model, and the step of training the other network parts of the water quality prediction base model comprises the following steps: After the backbone network parameters of the water quality prediction base model are frozen, respectively obtaining output data of the residual error network, output data of a gating mechanism of the fully-connected network block and output data of the backbone network parameters of the water quality prediction base model; Performing product calculation on the output data of the residual error network and the output data of the gating mechanism of the fully-connected network block; adding the product result of the output data with the output data of the backbone network parameters of the water quality prediction base model to obtain a target predicted value; And fully training other network parts of the water quality prediction base model according to the target predicted value by taking the real value as a label to obtain a target water quality extremum prediction model.
6. The method of claim 5, wherein the step of using the real value as a label to fully train other network parts of the water quality prediction base model according to the target predicted value to obtain a target water quality extremum prediction model further comprises: Testing the target water quality extreme value prediction model through a target test sample set to obtain a comprehensive prediction value; Calculating a comprehensive prediction mean value according to the comprehensive prediction value and the sample number of the target test sample set; Based on a multi-index algorithm, respectively calculating a plurality of general evaluation indexes according to the comprehensive predicted value, the comprehensive predicted average value, the sample number of the target test sample set and the comprehensive true value; Acquiring a special sample marked as an extremum in the target test sample set, and testing the target water quality extremum prediction model through the special sample to obtain a special predicted value; calculating a special prediction mean value according to the special prediction value and the sample number of the special samples; Calculating a plurality of extreme value prediction performance evaluation indexes according to the special predicted value, the special predicted mean value, the sample number of the special samples and the special true value based on the multi-index algorithm; and when the plurality of general evaluation indexes and the plurality of extremum predicting performance evaluation indexes meet preset requirements, continuing to execute the step of extremum predicting the water quality data of the sudden water pollution based on the target water quality extremum predicting model.
7. A water quality extremum prediction device based on a gate residual network and empirical mode decomposition is characterized by comprising: the construction module is used for generating an original water quality data sequence according to the historical water quality data and respectively constructing a water quality prediction base model, a residual error network and a gating mechanism of a full-connection network block according to the original water quality data sequence; The training module is used for training other network parts of the water quality prediction base model according to the residual error network and the gating mechanism of the fully-connected network block after the backbone network parameters of the water quality prediction base model are frozen, so as to obtain a target water quality extremum prediction model; And the prediction module is used for carrying out extremum prediction on the water quality data of sudden water pollution based on the target water quality extremum prediction model.
8. An electronic device, comprising: At least one memory for storing a computer program; At least one processor for executing the memory-stored program, which processor is adapted to perform the method according to any of claims 1-6 when the memory-stored program is executed.
9. A computer readable storage medium storing a computer program, characterized in that the computer program, when run on a processor, causes the processor to perform the method according to any one of claims 1-6.
10. A computer program product, characterized in that the computer program product, when run on a processor, causes the processor to perform the method according to any of claims 1-6.

Description

Water quality extremum prediction method based on gating residual error network and empirical mode decomposition Technical Field The application belongs to the technical field of water quality prediction, and particularly relates to a water quality extremum prediction method based on a gate control residual error network and empirical mode decomposition. Background In the water quality time sequence prediction of a sewage treatment plant, the extremum prediction is an extremely important index, the extremum of water quality data is usually closely related to sudden environmental changes, such as the concentration of suddenly increased pollutants, sudden weather changes and the like, and the accuracy of the extremum prediction directly determines the sensitivity of an early warning system. Therefore, the accurate prediction of the water quality extreme value has important significance for the intelligent and fine management and risk avoidance of the sewage treatment plant. At present, a common mode for water quality extremum prediction relies on a mixed model based on multi-mode and integrated deep learning to predict fluctuating inflow load in a wastewater treatment plant, and an empirical mode decomposition algorithm is fused to capture nonlinear and non-stationary characteristics in data, but the model is insufficient to capture the change rule of the model for predicting the sudden change of inflow pollutant concentration caused by sudden water pollution, so that the accuracy of extremum prediction is low. And all network parameters are trained by the model, so that the model training speed is low, and the efficiency of predicting the water quality extremum is low. Disclosure of Invention Aiming at the defects of the prior art, the application aims to provide a water quality extremum prediction method based on a gate control residual error network and empirical mode decomposition, and aims to solve the problems of low efficiency and accuracy of predicting water quality extremum caused by insufficient water quality fluctuation law of sudden water pollution captured by a mixed model of multi-mode and integrated deep learning in the prior art. In order to achieve the above object, in a first aspect, the present application provides a water quality extremum prediction method based on a gated residual network and empirical mode decomposition, including: Generating an original water quality data sequence according to the historical water quality data, and respectively constructing a water quality prediction base model, a residual error network and a gating mechanism of a full-connection network block according to the original water quality data sequence; After backbone network parameters of the water quality prediction base model are frozen, training other network parts of the water quality prediction base model according to the residual network and a gating mechanism of the fully-connected network block to obtain a target water quality extremum prediction model; And carrying out extremum prediction on the water quality data of sudden water pollution based on the target water quality extremum prediction model. In one embodiment, the step of generating the raw water quality data sequence from the historical water quality data comprises: performing missing detection on the historical water quality data, and performing primary linear interpolation on the detected missing segments; Decomposing the history water quality data after primary interpolation based on a seasonal trend decomposition algorithm to obtain a current trend item and a current residual error item; Performing secondary linear interpolation on the historical water quality data, and decomposing the historical water quality data after secondary interpolation based on the seasonal trend decomposition algorithm to obtain a target trend item and a target residual error item; replacing the current trend item according to the target trend item, and replacing the current residual item according to the target residual item; And reorganizing the replaced trend item, the replaced residual item and the reserved season item to generate an original water quality data sequence. In an embodiment, the step of constructing a water quality prediction base model, a residual network and a gating mechanism of a fully connected network block according to the original water quality data sequence respectively includes: Slicing the original water quality data sequence according to a preset input window and a preset step length, and determining target sequence characteristics according to a sequence slicing result; constructing a water quality prediction base model by taking the target sequence characteristics as input and the concentration at the next moment as output; inputting an original water quality data sequence into the water quality prediction base model, and obtaining a predicted value output by the water quality prediction base model; and respectively constructing