Method for optimizing reservoir operation for multiple objectives based on graph convolutional neural network and NSGA-II algorithm

Patent metadata

Field	Value
Publication number	US-11748628-B2
Application number	US-202117909032-A
Country	US
Kind code	B2
Filing date	Oct 21, 2021
Priority date	Mar 15, 2021
Publication date	Sep 5, 2023
Grant date	Sep 5, 2023

Abstract

A method for optimizing a reservoir operation for multiple objectives based on a GCN and a NSGA-II algorithm. The method includes collecting relevant data for reservoir flood-control operation and establishing a multi-objective optimization model for the flood control. An initial population is obtained. Grouping individuals by an encoding operation and the grouped classifications are nodes of the GCN, and mapping parent-child relationships obtained by crossover and mutation operations as edges between the nodes in the GCN. A preliminary Pareto frontier is obtained, abscissas of the preliminary Pareto frontier are grouped and labeled, and a GCN model is trained by using the grouping labels and the graphic structure obtained in Step 2. The nodes in the graphic structure are classified by using the trained GCN model, and a uniformity of the Pareto frontier is adjusted. A set of non-inferior schemes of the multi-objective optimization problem for the reservoir operation is output.

First claim

What is claimed is:

1. A method for optimizing a reservoir operation for multiple objectives based on a graph convolutional neural network (GCN) and a Non-Dominated Sorting Genetic Algorithm (NSGA-II algorithm), wherein the method comprises following steps:
Step 1, collecting relevant data for reservoir flood-control operation and establishing a multi-objective optimization model for the reservoir operation;
Step 2, obtaining an initial population of a multi-objective optimization problem for the reservoir operation by using the NSGA-II algorithm, grouping individuals in the population by an encoding operation and marking the individuals with classifications, each of which is taken as a node in a GCN graphic structure, and mapping parent-child relationships obtained by crossover and mutation operations as edges between nodes in the GCN graphic structure, to obtain the GCN graphic structure and a preliminary Pareto frontier;
Step 3, grouping and labeling abscissas of the preliminary Pareto frontier obtained in Step 2, and then training a GCN model by using the grouping labels and the GCN graphic structure obtained in Step 2;
Step 4, classifying the nodes in the GCN graphic structure by using the trained GCN model, and then adjusting a uniformity of the Pareto frontier by using the NSGA-II algorithm until an algorithm iteration ends, to obtain a more uniform Pareto frontier; and
Step 5, outputting, according to the more uniform Pareto frontier obtained in Step 4, a set of non-inferior schemes of the multi-objective optimization problem for the reservoir operation.

2. The method for optimizing the reservoir operation for the multiple objectives based on the graph convolutional neural network and the NSGA-II algorithm according to claim 1 , wherein the objectives considered in the multi-objective optimization model for the reservoir operation in Step 1 comprise:
firstly, minimizing an upstream water level and ensuring that the reservoir maintains a low water level during a flood season, thereby ensuring safety of a dam; and
secondly, minimizing a maximum discharge volume in the reservoir, and storing as much floodwater as possible by the reservoir to ensure safety in a downstream to minimize an inundation loss.

3. The method for optimizing the reservoir operation for the multiple objectives based on the graph convolutional neural network and the NSGA-II algorithm according to claim 1 , wherein Step 2 specifically comprises:
Step 2.1, randomly initializing the population of the multi-objective optimization problem for the reservoir operation, encoding the individuals in the population according to a following encoding method, and taking encoded individual classifications as the nodes in the GCN graphic structure; making, through the encoding method, a plurality of individuals in the population belong to a same classification represented by a same node, to avoid a redundancy in the graphic structure caused by too many individuals in the population;
the encoding method being that: a definition domain of the maximum discharge flow Qmax is divided at an equal distance, each distance is ┌Q max /N┐, where Qmax is the maximum discharge volume, and N is a number of classifications, that is, a number of nodes in the GCN graphic structure, and the individuals are encoded by using a method ┌Q/┌Q max /N┐┐; the Pareto frontier is divided at an equal distance by the abscissas, intervals of which are defined as 0, 1, 2 . . . , wherein a node 0 is defined as an elimination node, and a length of a gene sequence in the population is determined by a number of variables; and
Step 2.2, performing the crossover and mutation operations on the individuals in the population in Step 2.1 to determine whether the constraint conditions are satisfied, and pointing codes in the gene sequence of eliminated individuals that satisfy the constraint conditions to the node 0; recording the individuals that satisfy the conditions, and defining the parent-child relationships generated by the crossover and mutation operations as the edges of the graphic structure.

4. The method for optimizing the reservoir operation for the multiple objectives based on the graph convolutional neural network and the NSGA-II algorithm according to claim 1 , wherein Step 3 specifically comprises:
Step 3.1, grouping and labeling the preliminary Pareto frontier obtained by the NSGA-II algorithm by the abscissas, that is, dividing a value domain of an objective function at an equal distance, to define digital labels;
Step 3.2, setting input parameters, representing relationships between the individuals in the population by an adjacency matrix A∈R N×N in the graph convolutional neural network, and initializing a characteristic matrix X∈R N×D ;
Step 3.3, taking an input layer with N individuals as a first portion, that is, constituting an N-layered input layer with N nodes in the graph, where the characteristic matrix X and the adjacency matrix A serve as inputs;
Step 3.4, taking a convolutional layer composed of two layers of graph convolutions as a second portion, and transmitting characteristics of a base layer to a next layer by trigging information transmission of the edges between the two layers; and
Step 3.5, taking an output layer as a third portion, and outputting, for the characteristic matrix X obtained by calculating and transmitting of the two layers of convolutional layers, in turn a classification probability of each of the nodes through an activation function.

5. The method for optimizing the reservoir operation for the multiple objectives based on the graph convolutional neural network and the NSGA-II algorithm according to claim 1 , wherein Step 4 specifically comprises:
Step 4.1, obtaining the preliminary Pareto frontier by the NSGA-II algorithm through selection, crossover and mutation operations, generating, during the process, relationships of a parent-child tree-shaped structure between the individuals in the population of the multi-objective optimization problem for the reservoir operation, and converting the tree-shaped structure into the GCN graphic structure;
Step 4.2, determining whether the population iteration threshold is satisfied, and if satisfied, starting training the GCN, learning node characteristics in a population evolution relationship graphic network, and making a node classification result correspond to a classification of the Pareto frontier; and
Step 4.3, classifying the nodes by using the GCN, and then adjusting the individuals in the Pareto frontier through a loop iteration by using the NSGA-II algorithm, which specifically comprises following steps:
Step 4.3.1, classifying the nodes by using the GCN, and supplementing and improving the preliminary Pareto frontier;
Step 4.3.2, traversing the nodes with different classifications in the Pareto frontier, deleting redundant nodes with a same classification, and increasing population of classifications with a fewer number, to increase differences;
Step 4.3.3, performing the crossover and mutation operations on the nodes with a good performance obtained through the GCN classification by using the NSGA-II algorithm to obtain new nodes; and
Step 4.3.4, determining whether the nodes satisfy the constraint conditions, and retaining, if satisfied, the nodes.

Assignees

• Univ Hohai

Inventors

• Hu Hexuan
• Hu Qiang
• Zhang Ye
• Hu Zhenyun

Classifications

• G06N3/126

  Evolutionary algorithms, e.g. genetic algorithms or genetic programming · CPC title

• G06N3/08

  Learning methods · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/086 Primary

  using evolutionary algorithms, e.g. genetic algorithms or genetic programming · CPC title

• G06Q50/06 Primary

  Energy or water supply · CPC title