Water quality prediction method of deep learning model based on physical law and process driving

Abstract

The invention provides a water quality prediction method of a deep learning model based on a physical law and process driving. The method comprises the following steps: modifying a loss function of the deep learning model according to the physical law; using the water quality model to generate simulation time sequence data of the water quality indexes; training the deep learning model by using the simulation data to obtain a pre-trained model; adjusting and optimizing the pre-training model by using historical measured data of the water quality indexes to obtain a physical constraint and process driven deep learning model PRPGDL; and finally, predicting future water quality index data based on the PRPGDL model. Compared with a water quality model, the method needs fewer boundary conditions and parameters, and has higher prediction accuracy, speed and flexibility; compared with a deep learning model, the method has higher accuracy and universality, and needs less actual measurement data; the water quality prediction method is higher in accuracy, higher in generalization ability and applicability and less in actual measurement data demand.

Description

technical field [0001] The invention relates to the field of water quality prediction, in particular to a water quality prediction method based on a physical law and a process-driven deep learning model. Background technique [0002] Water quality prediction is an effective tool for flexible water quality management and water pollution prevention and control. At present, there are two types of water quality prediction methods: mechanism model method and non-mechanism model method. Process-based mechanism models (such as EFDC, SWAT, etc.) contain the understanding of processes and mechanisms developed based on decades of observations and experiments, have strong generalization ability and applicability, and are currently the preferred method for water quality prediction . For example, patent CN107091911B discloses a water quality prediction method based on the EFDC model. However, the process-based water quality model realizes only a part of the ecosystem process, which ma...

AI Technical Summary

Problems solved by technology

However, the deep learning method has no assumptions about the process mechanism hidden in the data, and ignores known laws or theories (such as the law of conservation of matter, law of conservation of energy, etc.), which will bring unreal and inaccurate predictions, leading to The generalization ability, applicability and generalization of the model are poor, especially when its application exceeds the scope of the training data of the deep learning model

In addition, deep learning models are difficult to apply in areas where historical data is scarce due to the need for a large amount of historical data to learn the process mechanism of complex systems

Images