CN113723704B - Water quality rapid prediction method based on continuous and graded mixed data - Google Patents

Abstract

The present invention provides a rapid water quality prediction method based on continuous and graded mixed data. The method utilizes the classification levels of parameters such as watershed land use area, GDP, and population to predict changes in water quality in different areas in the watershed and build a deep learning neural network. , quickly calculate the surface water quality change and its spatial distribution under different levels of social and economic development in the basin, and use it as the basis for water environment governance in the basin. Compared with the prior art, the beneficial effect of the present invention is: the water quality prediction method of the present invention avoids the problem of lack of socio-economic statistical data when using socio-economic data for water quality prediction, does not need accurate socio-economic data, only uses social Economic classification and grade data can be used for forecasting, which is more practical. Moreover, the water quality prediction is performed through the deep learning neural network. Compared with the traditional numerical simulation method, the calculation speed is faster and the scope of application is wider.

Description

A rapid water quality prediction method based on continuous and hierarchical mixed data

Technical Field

The invention relates to the technical field of water quality prediction, and in particular to a method for rapid water quality prediction based on continuous and graded mixed data.

Background Art

With the development of social economy and the increase of human pollution discharge activities, water pollution has become an important factor restricting the sustainable development of our society and the pursuit of quality life by the people. However, with the continuous strengthening of social and economic ties between regions and the impact of climate change, the evolution of my country's water environment presents complex changes with cross-regional and multi-factor coupling effects. The water quality prediction and water environment management of a single water sample or river section can no longer meet the requirements, and there is an urgent need for comprehensive water quality prediction with multiple influencing factors in a large spatial range.

Traditional water quality prediction methods mainly include numerical simulation methods and mathematical statistics methods. However, due to the complexity of the watershed hydrodynamic process and the hydrochemical process of pollutants, the migration and transformation mechanism of water pollutants is not fully understood, resulting in low accuracy of numerical simulation. In addition, the numerical simulation method has a large amount of calculation and has certain difficulties in large-scale application. In terms of mathematical statistics methods, although the numerical simulation method has a low amount of calculation, it has strict requirements on the time series, continuity and normal distribution of data, and is relatively difficult to model and has low applicability.

At the same time, existing water quality prediction methods based on deep learning are based on recurrent neural networks, such as the long short-term memory model (LSTM). They generally predict future water quality changes based on historical water quality data, and often only consider the changing trend of a single water quality indicator or the impact between water quality indicators. They fail to consider the impact of external factors such as social economy on water quality changes. As a result, such prediction methods are insufficient in supporting decision makers in water environment governance decisions and economic and social development plans, and are not very practical.

In addition, due to the complexity of the human socio-economic system, there are many socio-economic indicators related to water quality, including GDP, population, food production, cultivated land area, sewage discharge, sewage treatment level, etc., and the statistics of basin socio-economic data often face missing data or incomplete statistics for a certain period or region. The present invention uses socio-economic classification and grade data for water quality prediction, avoiding the problem of missing socio-economic statistical data, and does not require accurate socio-economic indicator values. Water quality prediction can be performed using only socio-economic indicator grade data, which is more practical.

Summary of the invention

In order to solve the shortcomings of the water quality prediction methods in the background technology, such as the singleness, poor applicability, and low practicality of using only a single indicator for prediction, the present invention provides a water quality rapid prediction method based on continuous and graded mixed data. The method integrates the socio-economic classification grade data and has the advantages of fast calculation speed and high operation efficiency.

To achieve the above purpose, the technical solution of the method for rapid prediction of water quality in a river basin based on deep learning of classified data of the present invention is:

A method for rapid water quality prediction based on continuous and hierarchical mixed data, comprising the following steps:

S1. Determine the scope of the watershed and its internal divisions, wherein the internal divisions include sub-watershed divisions and administrative district divisions;

S2, determining the meteorological parameters of the basin within the forecast period, and then converting the meteorological parameters into sub-basin meteorological data; the meteorological parameters include rainfall, evaporation and temperature data observed at meteorological point i at time t;

S3, determining the socio-economic parameters of the sub-basin within the forecast period, and generating a set of socio-economic parameters of the sub-basin;

S4. Determine sub-basin water quality parameters;

S5. Classify the data in the sub-basin socio-economic parameter set;

S6, hot-coding the socioeconomic parameters classified by grade to form a hot-coding matrix of the socioeconomic parameters;

S7. Standardize meteorological parameters and water quality parameters;

S8, divide the training set and the test set into the hot coding matrix of socioeconomic parameters and the standardized sub-basin meteorological data and water quality parameters;

S9. Build a fully connected deep learning neural network and define the loss function and iterative optimization algorithm;

S10, inputting the training set into the deep learning neural network to obtain a trained deep learning neural network, and then inputting the test set into the trained deep learning neural network to obtain predicted water quality parameters; comparing the predicted water quality parameters with the measured water quality parameters, adjusting the model parameters, and finally storing the deep learning neural network and its parameter data that meet the accuracy requirements;

S11. Select meteorological and socio-economic data of a certain period of time in the basin, input them into the deep learning neural network stored in S10, and predict the water quality parameters of all sub-basins in that period.

Furthermore, step S2 is more specifically:

S21, the meteorological station obtains rainfall, evaporation and temperature data within the forecast period;

S22, creating a meteorological station spatial distribution map based on the latitude and longitude coordinates of the meteorological stations, and importing the rainfall, evaporation and temperature data of each station and each time period into the attribute table of the station distribution map;

S23, taking the entire watershed as the boundary, using the Kriging interpolation method to perform spatial interpolation on the meteorological data of each station, generating raster data of rainfall, evaporation and temperature in the watershed for each period, and storing them in the geospatial database;

S24. Taking the sub-basin as the range, calculate the average values of the rainfall, evaporation and temperature data of all grids within each sub-basin to obtain the meteorological parameters of the sub-basin.

Furthermore, the socio-economic parameters include land use area, permanent population, GDP, GDP per capita, population density, food production, and urban sewage treatment rate.

Furthermore, step S3 is more specifically:

S31. Obtain socioeconomic parameters such as land use area, permanent population, GDP, GDP per capita, population density, food production, and urban sewage treatment rate during the forecast period;

S32, converting the socio-economic parameters in S31 into relevant parameters of the sub-basin;

S33. Generate a set of socio-economic parameters for the sub-basin.

Furthermore, the water quality parameters include the concentration values of dissolved oxygen, COD manganese, ammonia nitrogen, total phosphorus and total nitrogen in the surface water; step S4 is more specifically:

S41, determining the longitude and latitude coordinates of the water quality monitoring stations in all the sub-basins to produce a spatial distribution map of the water quality observation stations, and then importing the water quality index data of each water quality observation station in each time period into the attribute table of the station distribution map;

S42, calculate the water quality distribution in the river channel of the sub-basin;

S43. Taking the sub-basin as the scope, calculate the average value of the river water quality within each sub-basin as the water quality parameter of the sub-basin.

Further, step S42 is more specifically:

S421. Collect river network data of rivers above level 5 in the entire basin;

S422. Taking the river network in the basin as the boundary, the water quality data of each time period at the site is spatially interpolated using the inverse distance weighted method to generate water quality raster data for each time period along the river and store them in the geospatial database.

Furthermore, the land use area includes the utilization area of cultivated land, forest land, grassland, water area, urban and residential land, and unused land; step S5 is more specifically: for the level thresholds of the socioeconomic parameters of the areas of cultivated land, forest land, grassland, water area, urban and residential land, and unused land, as well as the permanent population, GDP, grain output, per capita GDP, population density and urban sewage treatment rate, calculate the level values of each socioeconomic variable to form a socioeconomic level variable.

Furthermore, step S6 is more specifically:

S61, convert the socioeconomic grade variable of each sub-basin and each time period into a 0-1 row vector; the length of the row vector is the total number of grades N of the socioeconomic grade variable, that is, the row vector is 1 row and N columns; in the row vector, the nth column corresponding to the grade value n of the socioeconomic grade variable has a value of 1; the remaining elements in the row vector have a value of 0;

S62. Construct a hot-coded 0-1 matrix of the socioeconomic class variable.

Furthermore, step S8 is more specific as follows: merging the hot-coded 0-1 matrices of the sub-basin meteorological data and socioeconomic data into a digital matrix as the input digital matrix of the deep learning neural network; converting the sub-basin water quality parameters into a digital matrix as the output label matrix of the deep learning neural network; and then using 70% of the rows in the input digital matrix and the output label matrix as a training set and 30% of the rows as a test set.

Furthermore, the parameters of the deep learning neural network include the number of hidden layers k, the number of nodes n in each hidden layer, the activation function, the loss function, the accuracy calculation function, the learning rate, the batch size, the maximum number of iterations and the node random discard ratio; the deep learning neural network adopts four hidden layers; the activation function of each layer adopts the Selu function; the fourth hidden layer adopts a node random discard mechanism;

The loss function and iterative optimization algorithm are defined as follows: the loss function uses the mean square error MSE, and the accuracy uses the mean square error MAE. The specific calculation formula is:

Among them, _zi is the actual value and _yi is the predicted value.

Compared with the prior art, the advantages and beneficial effects of the present invention are as follows: the present invention uses land use area, GDP, population and other river basin socio-economic classification grade data to predict water quality over a large range of river basin scales, overcoming the shortcomings of small calculation range and slow calculation speed caused by relying on numerical simulation in traditional water quality prediction, while avoiding the strict requirements of traditional statistical methods and currently commonly used recurrent neural networks for water quality prediction on the continuity of input data, and can more conveniently calculate the quantitative relationship between discrete socio-economic classification grade data and continuously changing water quality data, thereby improving the practicality of prediction. In addition, the present invention avoids the problem of missing socio-economic statistical data when using socio-economic data for water quality prediction, and does not require accurate socio-economic data, and can only use socio-economic classification grade data for prediction, with clear theoretical significance, simple and easy operation, and easy to apply in actual water quality management.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a method for rapid prediction of water quality in a river basin according to the present invention;

Figure 2 is a scatter plot of the predicted and measured values of CODMn;

Figure 3 is the spatial distribution of the predicted value of CODMn on a certain day.

DETAILED DESCRIPTION

The accompanying drawings are only used for illustrative purposes and should not be construed as limiting the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

The technical solution of the present invention is further described below in conjunction with FIGS. 1 to 3 and embodiments.

A rapid water quality prediction method based on continuous and hierarchical mixed data, as shown in FIG1 , comprises the following steps:

S1. Determine the scope of the watershed and its internal divisions, wherein the internal divisions include sub-watershed divisions and administrative district divisions;

According to the Chinese water resources division of the National Water Resources Comprehensive Plan, the Pearl River area was selected as the basin of the embodiment. The digital elevation data of the basin of the embodiment was extracted and the catchment area was calculated using NASA's SRTM 90-meter digital elevation data, and 138 sub-basins in the Pearl River area were obtained. The national prefecture-level city administrative district distribution map (SHP format) of the National Basic Geographic Information Center (http://www.ngcc.cn/ngcc/) was used to extract 75 prefecture-level city administrative districts in the Pearl River area; and the prefecture-level city distribution map of the Pearl River area was drawn.

表示；接着将所述气象参数转换为子流域气象数据。具体包括以下步骤：S2, determine the meteorological parameters of the basin during the forecast period, the meteorological parameters include the rainfall, evaporation and temperature data observed at the meteorological point i at time t, respectively

Then, the meteorological parameters are converted into sub-basin meteorological data. Specifically, the following steps are included:

S21. Use the daily value dataset of China's ground climate data (V3.0) provided by the China Meteorological Data Network to extract the daily average rainfall, evaporation and temperature data of 69 meteorological stations in the basin of the embodiment from January 1, 2020 to April 30, 2021;

S22. Using ARCGIS software to create a spatial distribution map of meteorological stations based on the longitude and latitude coordinates of the meteorological stations, and importing the rainfall, evaporation and temperature data of each station and each time period into the attribute table of the station distribution map;

S23. Using the Pearl River area as the boundary, spatially interpolate the daily meteorological data of the station using the Kriging interpolation method; create a batch program for the Kriging interpolation method to generate raster data of rainfall, evaporation, and temperature in the Pearl River area for each period (day), and store them in the geospatial database. Among them, there are 486 raster maps for rainfall, evaporation, and temperature data, one map per day.

S24, making a raster data processing program, taking the sub-basin as the boundary, extracting the rainfall, evaporation and temperature data of each sub-basin in the Pearl River area for each time period (day), and storing them in an EXCEL document. There is one EXCEL document for each sub-basin, a total of 138 EXCEL documents, each of which contains data of three variables (rainfall, evaporation, temperature) for 486 time periods (486 rows, 3 columns).

S3, determine the socio-economic parameters of the sub-basin during the forecast period, and generate a set of socio-economic parameters of the sub-basin; the above socio-economic parameters include land use area, permanent population, GDP, per capita GDP, population density, food production, and urban sewage treatment rate. The land use area includes cultivated land, forest land, grassland, water area, urban and residential land, and unused land. Step S3 specifically includes the following steps:

S31. Obtain socioeconomic parameters such as land use area, permanent population, GDP, per capita GDP, population density, food production, and urban sewage treatment rate during the forecast period. More specifically, the above parameters can be obtained in the following ways: using satellite remote sensing data to obtain annual land use data for the entire river basin during the forecast period; obtaining the permanent population, GDP, and food production of each administrative region for each year during the forecast period through regional statistical yearbooks; obtaining per capita GDP, population density, and urban sewage treatment rate of each administrative region for each year during the forecast period through regional statistical yearbooks.

S32, converting the socio-economic parameters in S31 into relevant parameters of the sub-basin;

(1) Determine the land use parameters of each sub-basin and each time period. Specifically, reclassify the basin land use data according to cultivated land, forest land, grassland, water area, urban and residential land, and unused land to generate basin land use raster data. Then, calculate the area of the six land use types in each sub-basin according to the sub-basin division, as shown below.

表示第y年、子流域j内、属于d类别土地的栅格数目；

分别表示第y年和第t时刻子流域j内d类别土地的面积。其中，t时刻在第y年内。Wherein, j is the sub-basin number; d is the land use type number, including cultivated land (d = 1), forest land (d = 2), grassland (d = 3), water area (d = 4), urban and residential land (d = 5), and unused land (d = 6); t is the time period (day), y is the year; AreaS is the area of a grid in sub-basin j;

represents the number of grids belonging to category d in sub-basin j in year y;

They represent the area of land of category d in sub-basin j in year y and time t, respectively. Time t is in year y.

(2) Determine the permanent population, GDP and grain output of the basin during the forecast period by converting the socioeconomic data of the administrative area into relevant parameters of the sub-basin using the area-weighted average method. The specific formula is as follows.

为第y年、第k行政区的常住人口、GDP和粮食产量；

和

分别为第y年和第t时刻、第j子流域的常住人口、GDP和粮食产量；其中，t时刻在第y年内。

表示第k行政区处于第j子流域内部的面积占第k行政区总面积的比值；

表示第k行政区处于第j子流域内部的面积。Among them, k represents the administrative district number in the basin, N is the total number of administrative districts in the basin; j is the sub-basin number; e represents the number of the socio-economic parameter, e = 1 represents the permanent population, e = 2 represents GDP, and e = 3 represents grain output;

is the permanent population, GDP and food output of the kth administrative district in year y;

and

are the permanent population, GDP and grain output of the j-th sub-basin in the y-th year and the t-th moment respectively; among which, the t-th moment is in the y-th year.

It represents the ratio of the area of the kth administrative district inside the jth sub-basin to the total area of the kth administrative district;

Represents the area of the kth administrative district within the jth sub-basin.

(3) Determine the GDP per capita, population density and urban sewage treatment rate of the basin during the forecast period. Specifically, GDP per capita, population density and urban sewage treatment rate are proportional data, so the area-weighted average method cannot be used to calculate the parameters of the sub-basin. By constructing the topological relationship between administrative regions and sub-basins, the GDP per capita, population density and urban sewage treatment rate of the sub-basin are calculated using the arithmetic average method. The details are shown below.

为第y年、第k行政区的人均GDP，人口密度和城市污水处理率。

和

分别为第y年和第t时刻、第j子流域的人均GDP，人口密度和城市污水处理率；其中，t时刻在第y年内。Among them, k represents the administrative district number in the basin, N is the total number of administrative districts in the basin; j is the sub-basin number; a represents the number of the socio-economic parameter, a = 1 represents GDP per capita, a = 2 represents population density, and a = 3 represents urban sewage treatment rate;

is the GDP per capita, population density and urban sewage treatment rate of the kth administrative district in the yth year.

and

are the per capita GDP, population density and urban sewage treatment rate of the j-th sub-basin in the y-th year and the t-th moment respectively; among which, the t-th moment is in the y-th year.

According to the above (steps S31 and S32), steps S31 and S32 are more specific as follows: using the 1km resolution annual land use data provided by the Landsat 8 satellite, extract the land use data of the Pearl River area in 2020 and 2021, totaling 2 raster data maps. Prepare a land use data processing program, taking the 138 sub-basins in the Pearl River area as the scope, extract the area of cultivated land, forest land, grassland, water area, urban and residential land, and unused land in each sub-basin and each year; convert the annual land use data of the sub-basin into daily data, and the area of various types of land on each day is equal to the corresponding land use data of the year in which the day is located. Store the five types of land area data of the sub-basin every day in an EXCEL document. One EXCEL document for each sub-basin, a total of 138 EXCEL documents, each of which contains data (486 rows and 6 columns) of 6 variables (six types of land area data) for 486 time periods.

The CNKI yearbook database was used to collect the permanent population, GDP, GDP per capita, population density, grain output, and urban sewage treatment rate of 75 prefecture-level cities in the Pearl River area in 2020 and 2021. A socioeconomic data processing program was compiled. According to the area-weighted average method and arithmetic average method, the socioeconomic data of 75 prefecture-level cities were converted into socioeconomic data of 138 sub-basins based on the spatial topological relationship between the prefecture-level city administrative districts and the sub-basins. The annual socioeconomic data of the sub-basin was converted into daily data, and the daily socioeconomic data was equal to the corresponding data of the year in which the day was located. The data was stored in an EXCEL document. There was one EXCEL document for each sub-basin, a total of 138 EXCEL documents, each of which contained data of 6 variables (6 socioeconomic parameters) for 486 time periods (486 rows and 6 columns).

S33. Generate a set of socio-economic parameters for the sub-basin.

S4, determine the water quality parameters of the sub-basin; the water quality parameters include the concentration values of dissolved oxygen, COD manganese, ammonia nitrogen, total phosphorus, and total nitrogen in surface water; use the national water quality monitoring data provided by the National Surface Water Quality Automatic Monitoring Real-time Data Release System of the China National Environmental Monitoring Center to extract the daily average water quality data of 204 water quality stations in the Pearl River area from January 1, 2020 to April 30, 2021. Step S4 specifically includes the following steps:

S41. Determine the longitude and latitude coordinates of the river water quality monitoring stations in the entire basin (all sub-basins), use ARCGIS software to create a spatial distribution map of the water quality observation stations, and then import the water quality index data of each river water quality station and each time period into the attribute table of the station distribution map.

S42, calculate the water quality distribution in the river channel of the sub-basin;

Since the surface water quality is a parameter of the river water body, it has a certain spatial continuity in the river water body, but does not have spatial correlation on the basin surface. Therefore, the spatial interpolation method of the basin surface is not used to calculate the water quality of the sub-basin. The river line interpolation method is used to calculate the water quality distribution in the sub-basin river. The specific steps include:

(1) Collect river network data (SHP data) of rivers above level 5 in the entire basin;

(2) Taking the river network in the basin as the boundary, the water quality data of each time period at the station are spatially interpolated using the inverse distance weighted method to generate water quality raster data for each time period along the river and store them in the geospatial database.

S43. Taking the sub-basin as the scope, calculate the average value of the river water quality within each sub-basin as the water quality parameter of the sub-basin. The calculation formula is as follows.

表示子流域j范围内河道位置处栅格s的水质数据。

表示子流域j范围内河道中的栅格总数，

表示t时刻子流域j的水质数据。Wherein, j represents the sub-basin number, t represents the time period, s represents the grid number of the river channel position in sub-basin j; w represents the number of the water quality parameter, w=1 represents dissolved oxygen, w=2 represents COD manganese, w=3 represents ammonia nitrogen, w=4 represents total phosphorus, and w=5 represents total nitrogen;

Represents the water quality data of grid s at the river location within sub-basin j.

represents the total number of grids in the river within the scope of sub-basin j,

Represents the water quality data of sub-basin j at time t.

There is one EXCEL document for each sub-basin, for a total of 138 EXCEL documents, each of which contains data (486 rows and 5 columns) of five variables (five water quality indicators) for 486 time periods.

S5. Classify the data in the sub-basin socio-economic parameter set;

The data in the socioeconomic parameter set of each sub-basin include: cultivated land, forest land, grassland, water area, urban and residential land, unused land area, permanent population, GDP and grain output, GDP per capita, population density and urban sewage treatment rate, a total of 12 variables. A level threshold is set for each type of socioeconomic variable, and the level value of each socioeconomic variable is calculated as shown in the following formula.

示第t时刻子流域j内d类别土地的面积；

表示d类别土地面积的分级阈值，N表示分级数目；

表示第t时刻子流域j内d类别土地面积的等级数值。

表示第t时刻子流域j内e类别社会经济变量的数值；

表示e类别社会经济变量的分级阈值，L表示分级数目；

表示第t时刻子流域j内e类别社会经济变量的等级数值。

表示第t时刻子流域j内a类别社会经济变量的数值；

表示a类别社会经济变量的分级阈值，V表示分级数目；

表示第t时刻子流域j内a类别社会经济变量的等级数值。Among them, j is the sub-basin number, t is the time period; d is the land use type number, d=1 is cultivated land, d=2 is forest land, d=3 is grassland, d=4 is water area, d=5 is urban and residential land, and d=6 is unused land; e is the number of the socio-economic parameter, e=1 represents the permanent population, e=2 represents GDP, and e=3 represents grain output; a is the number of the socio-economic parameter, a=1 represents per capita GDP, a=2 represents population density, and a=3 represents urban sewage treatment rate.

represents the area of land of category d in sub-basin j at time t;

represents the classification threshold of the land area of category d, and N represents the number of classifications;

It represents the numerical grade of the land area of category d in sub-basin j at time t.

represents the value of the socioeconomic variable of category e in sub-basin j at time t;

represents the classification threshold of the socioeconomic variable of category e, and L represents the number of classifications;

It represents the numerical value of the socioeconomic variable of category e in sub-basin j at time t.

represents the value of the socioeconomic variable of category a in sub-basin j at time t;

represents the classification threshold of the socioeconomic variable of category a, and V represents the number of classifications;

It represents the level value of the socioeconomic variable of category a in sub-basin j at time t.

The classification is based on the maximum and minimum values of each sub-basin and each category of socioeconomic variables. The classification is as follows: cultivated land is divided into 4 levels according to area, forest land is divided into 5 levels, grassland is divided into 5 levels, water area is divided into 3 levels, urban and residential land is divided into 3 levels, the area of unused land is divided into 3 levels, the permanent population is divided into 4 levels, GDP and grain output are divided into 4 levels, and per capita GDP, population density and urban sewage treatment rate are divided into 3 levels.

S6, hot-coding the socioeconomic parameters of the classification levels to form a hot-coding matrix of the socioeconomic parameters;

转换为0-1行向量；行向量中，对应该社会经济等级变量等级数值n的第n列，取值为1；行向量中其余元素取值为0；具体如下式所示：S61. The socioeconomic class variables for each sub-basin and each time period are

Converted into a 0-1 row vector; in the row vector, the nth column corresponding to the value n of the socioeconomic class variable takes the value 1; the remaining elements in the row vector take the value 0; as shown in the following formula:

ca _{1, n} = 1 (12)

表示第t时刻子流域j内d类别土地面积的等级数值，等于n；

为d类别土地面积的等级数值所对应的0-1行向量，ca_1，n为行向量中第n列的数值，取值为1。N表示d类别土地面积的分级总数。采用上述方法，计算其他社会经济等级变量

和

的0-1行向量。Wherein, j represents the sub-basin number, t represents the time period, and d represents the land use type number;

It represents the numerical value of the land area of category d in sub-basin j at time t, which is equal to n;

is a 0-1 row vector corresponding to the grade value of the land area in category d, ca _{1, n} is the value of the nth column in the row vector, which is 1. N represents the total number of grades of land area in category d. Using the above method, calculate other socioeconomic grade variables

and

A 0-1 row vector of .

S62. Construct a hot-coded 0-1 matrix of the socioeconomic class variable.

人均GDP、人口密度、城市污水处理率的等级数值所对应的0-1行向量

合并在一起构成矩阵

具体如下式所示。0-1 row vector corresponding to the level values of permanent population, GDP, and grain output

0-1 row vector corresponding to the level values of GDP per capita, population density, and urban sewage treatment rate

Merge together to form a matrix

The details are shown in the following formula.

和水质参数

进行标准化处理；具体如下式所示：S7. Meteorological parameters

and water quality parameters

Standardization is performed; the specific formula is as follows:

代表第t时刻子流域j的气象和水质参数，包括

和

和

分别为各指标在所有时刻t和所有子流域j中的最大值和最小值；

为标准化后的数据。in,

represents the meteorological and water quality parameters of sub-basin j at time t, including

and

and

are the maximum and minimum values of each index at all times t and in all sub-basins j, respectively;

is the standardized data.

s8, divide the training set and test set in the hot coding matrix of socio-economic parameters and the standardized sub-basin meteorological data and water quality parameters;

The hot-coded 0-1 matrix of continuous sub-basin meteorological data and socioeconomic data is merged into a data matrix of t×j rows and 3+N+L+V columns as the input digital matrix of deep learning, represented by TableX; the continuous water quality data is converted into a data matrix of t×j rows as the output label matrix of deep learning, represented by TableY; then 70% of the rows in the input digital matrix and the output label matrix are used as the training set, and 30% of the rows are used as the test set, as shown in the following formula:

More specifically, the meteorological parameters of the sub-basins and the socio-economic 0-1 matrix parameters are integrated into a data set. The data is stored in an EXCEL table, one EXCEL document for each sub-basin, for a total of 138 EXCEL documents. Each document contains 44 columns of data for 486 time periods. The variables include: rainfall, evaporation, temperature continuity variables, and 0-1 row vectors of the level data of cultivated land, forest land, grassland, water area, urban and residential land, unused land, permanent population, GDP, per capita GDP, population density, food production, and urban sewage treatment rate. The 138 EXCEL document data are merged into one data matrix. This matrix is recorded as TableX as the input data matrix for deep learning.

The water quality parameters of the sub-basins are integrated into a data set. The water quality data of the 138 sub-basins are combined into a data matrix. The columns of the matrix are the five water quality variables. The rows of the matrix are 138 sub-basins and 486 days, totaling 67068 rows. This matrix is recorded as TableY as the output label matrix of deep learning.

TableX and TableY are standardized using the maximum and minimum standardization. Each data in the matrix is subtracted from the minimum value of the column where the data is located and then divided by the difference between the maximum and minimum values of the column. The matrices TableX and TableY are divided into two parts: training set and test set. 70% of the rows are used as training sets to form TrainX and TrainY; 30% of the rows are used as test sets to form TestX and TestY.

S9. Build a fully connected deep learning neural network (DNN) and define the loss function and iterative optimization algorithm;

The deep learning neural network parameters include the number of hidden layers k, the number of nodes n in each hidden layer, the activation function, the loss function (Ioss), the accuracy calculation function (metrics), the learning rate (learning rate), the batch size (batch size), the maximum number of iterations (epochs), and the node random discard ratio (dropout). The network uses four hidden layers, with 128 nodes in the first layer, 64 nodes in the second layer, 32 nodes in the third layer, and 16 nodes in the fourth layer; the activation function of each layer uses the Selu function (scaled exponentiallinear units). The fourth hidden layer uses a node random discard mechanism. In this embodiment, the discard rate is 20%, the learning rate (learning rate) is 0.001; the batch size (batchsize) is 64, and the maximum number of iterations (epochs) is 1000.

The loss function and iterative optimization algorithm are specifically defined as follows: the loss function uses mean square error (MSE), and the accuracy uses mean absolute error (MAE). The calculation formula is shown below. The training uses the Adam algorithm to optimize the training process.

Among them, z _i is the actual value and y _i is the predicted value. The smaller the MSE and MAE values are, the closer the predicted value of the model is to the true value, the higher the accuracy of the model is, and the better the performance is.

S10, inputting the training set into the deep learning neural network to obtain a trained deep learning neural network, and then inputting the test set into the trained deep learning neural network to obtain predicted water quality parameters; comparing the predicted water quality parameters with the measured water quality parameters, adjusting the model parameters, and finally storing the deep learning neural network and its parameter data that meet the accuracy requirements;

More specifically: TrainX is used as the input of the DNN network, and TrainY is used as the output label to train the DNN network. The weights of the DNN network are iteratively updated based on the training data until the model converges. The trained DNN network is used to predict water quality.

Input TestX into the DNN network and predict the corresponding water quality parameters. Denormalize the predicted values to get the actual values, and denormalize TestY to get the measured water quality values. Then, draw a scatter plot of the predicted actual values and the measured water quality values, and calculate the square of the correlation coefficient r (r ² ). The larger the r ² , the more accurate the prediction result. Continuously compare the predicted surface water quality values with the measured values, calculate the corresponding MSE and r ² , and adjust the model parameters according to the values of MSE and r ² , and retrain to ensure accuracy. Store the DNN network and its weights and other parameter data that meet the accuracy requirements to provide decision support for managers. The denormalization formula is as follows:

y′ _i =y _min +y _i ·(y _max -y _min ) (20)

Among them, _y′i is the predicted value after restoration; _yi is the predicted value output by the DNN network.

S11, select the meteorological and socio-economic data of a certain period of time in the basin, input it into the deep learning neural network stored in S10, and predict the water quality parameter values of all sub-basins in the period, including the concentration values of dissolved oxygen, COD manganese, ammonia nitrogen, total phosphorus, and total nitrogen. For each water quality index, draw a spatial distribution map of the water quality of the sub-basin, and convert the distribution map into a raster data format. Taking the river network of the basin as the boundary, extract the water quality index data of the grid at the river location to obtain the water quality distribution map of the river network. In this embodiment, the predicted value and measured value of CODMn, and the spatial distribution of the predicted value of CODMn on a certain day are shown in Figures 2 and 3, respectively.

Compared with the prior art, the beneficial effects of this embodiment are as follows: this embodiment takes the Pearl River Basin as an example, and adopts the socio-economic classification grade data of land use area, permanent population, GDP, per capita GDP, population density, grain output, and urban sewage treatment rate to predict water quality in a large range of the basin scale, which overcomes the shortcomings of small calculation range and slow calculation speed caused by relying on numerical simulation in traditional water quality prediction, and at the same time avoids the strict requirements of traditional statistical methods and currently commonly used recurrent neural networks for water quality prediction on the continuity of input data, and can more conveniently calculate the quantitative relationship between discrete socio-economic classification grade data and continuously changing water quality data, thereby improving the practicality of prediction. In addition, this invention avoids the problem of missing socio-economic statistical data when using socio-economic data for water quality prediction, and does not require accurate socio-economic data, and only socio-economic classification grade data can be used for prediction, which has clear theoretical significance, is simple and easy to operate, and is easy to apply in actual water quality management.

It is worth noting that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should be included in the scope of the claims of the present invention.

Claims (10)

1. A method for rapid water quality prediction based on continuous and hierarchical mixed data, characterized in that it comprises the following steps: S1. Determine the scope of the watershed and its internal divisions, wherein the internal divisions include sub-watershed divisions and administrative district divisions; S2, determining the meteorological parameters of the basin within the forecast period, and then converting the meteorological parameters into sub-basin meteorological data; the meteorological parameters include rainfall, evaporation and temperature data observed at meteorological point i at time t; S3, determining the socio-economic parameters of the sub-basin within the forecast period, and generating a set of socio-economic parameters of the sub-basin; S4. Determine sub-basin water quality parameters; S5. Classify the data in the sub-basin socio-economic parameter set; calculate the levels of various socio-economic variables according to the level thresholds of the socio-economic parameters, and obtain the sub-basin socio-economic development level of high, medium, and low levels; S6, hot-coding the socioeconomic parameters classified by grade to form a hot-coding matrix of the socioeconomic parameters; The level of the socioeconomic parameter is encoded in the form of a 0-1 row vector. The length of the row vector is the total number of levels N of the socioeconomic variable, that is, the row vector is 1 row and N columns. In the row vector, the nth column corresponding to the level value n of the socioeconomic level variable is 1; the rest of the elements in the row vector are 0. The level hot encoding row vectors of different socioeconomic variables in a sub-basin are spliced left and right according to the columns to obtain the level hot encoding row vectors of all socioeconomic variables in a sub-basin. The hot coding row vectors of the socioeconomic variables of all sub-basins are stacked up and down according to the rows to obtain the hot coding matrix of the socioeconomic variables of all basins; S7. Standardize meteorological parameters and water quality parameters; S8. Divide the training set and the test set in the hot coding matrix of the socio-economic parameters and the standardized sub-basin meteorological data and water quality parameters; splice the hot coding 0-1 matrix of the continuous sub-basin meteorological data and the discrete socio-economic grade data from left to right by column to obtain a unified digital matrix as the input digital matrix of the deep learning neural network, so as to realize the unified processing of continuous data and discrete data; S9. Build a fully connected deep learning neural network and define the loss function and iterative optimization algorithm; S10, inputting the training set into the deep learning neural network to obtain a trained deep learning neural network, and then inputting the test set into the trained deep learning neural network to obtain predicted water quality parameters; comparing the predicted water quality parameters with the measured water quality parameters, adjusting the model parameters, and finally storing the deep learning neural network and its parameter data that meet the accuracy requirements; S11. Select meteorological and socio-economic data of a certain period of time in the basin, input them into the deep learning neural network stored in step S10, and predict the water quality parameters of all sub-basins in the period. 2. The method according to claim 1, characterized in that: step S2 is more specifically: S21, the meteorological station obtains rainfall, evaporation and temperature data within the forecast period; S22, creating a meteorological station spatial distribution map based on the latitude and longitude coordinates of the meteorological stations, and importing the rainfall, evaporation and temperature data of each station and each time period into the attribute table of the station distribution map; S23, taking the entire watershed as the boundary, using the Kriging interpolation method to perform spatial interpolation on the meteorological data of each station, generating raster data of rainfall, evaporation and temperature in the watershed for each period, and storing them in the geospatial database; S24. Taking the sub-basin as the range, calculate the average values of the rainfall, evaporation and temperature data of all grids within each sub-basin to obtain the meteorological parameters of the sub-basin. 3. The method according to claim 1 is characterized in that the socio-economic parameters include land use area, permanent population, GDP, per capita GDP, population density, food production, and urban sewage treatment rate. 4. The method according to claim 3, characterized in that: step S3 is more specifically: S31. Obtain socioeconomic parameters such as land use area, permanent population, GDP, GDP per capita, population density, food production, and urban sewage treatment rate during the forecast period; S32, converting the socio-economic parameters in step S31 into relevant parameters of the sub-basin; S33. Generate a set of socio-economic parameters for the sub-basin. 5. The method according to claim 1, characterized in that: the water quality parameters include the concentration values of dissolved oxygen, COD manganese, ammonia nitrogen, total phosphorus and total nitrogen in surface water; step S4 is more specifically: S41, determining the longitude and latitude coordinates of the water quality monitoring stations in all the sub-basins to produce a spatial distribution map of the water quality observation stations, and then importing the water quality index data of each water quality observation station in each time period into the attribute table of the station distribution map; S42, calculate the water quality distribution in the river channel of the sub-basin; S43. Taking the sub-basin as the scope, calculate the average value of the river water quality within each sub-basin as the water quality parameter of the sub-basin. 6. The method according to claim 5, characterized in that: step S42 is more specifically: S421. Collect river network data of rivers above level 5 in the entire basin; S422. Taking the river network in the basin as the boundary, the water quality data of each time period at the site is spatially interpolated using the inverse distance weighted method to generate water quality raster data for each time period along the river and store them in the geospatial database. 7. The method according to claim 4 is characterized in that: the land use area includes the utilization area of cultivated land, forest land, grassland, water area, urban and residential land, and unused land; step S5 is more specifically: for the level thresholds of the socioeconomic parameters of the area of cultivated land, forest land, grassland, water area, urban and residential land, and unused land, as well as the permanent population, GDP, grain output, per capita GDP, population density and urban sewage treatment rate, the level value of each socioeconomic variable is calculated to form a socioeconomic level variable. 8. The method according to claim 7, characterized in that: step S6 is more specifically: S61, convert the socioeconomic grade variable of each sub-basin and each time period into a 0-1 row vector; the length of the row vector is the total number of grades N of the socioeconomic grade variable, that is, the row vector is 1 row and N columns; in the row vector, the nth column corresponding to the grade value n of the socioeconomic grade variable is 1; the rest of the elements in the row vector are 0; S62. Construct a hot-coded 0-1 matrix of the socioeconomic class variable. 9. The method according to claim 8 is characterized in that: step S8 is more specifically: merging the hot-coded 0-1 matrix of the sub-basin meteorological data and the socio-economic data into a digital matrix as the input digital matrix of the deep learning neural network; converting the sub-basin water quality parameters into a digital matrix as the output label matrix of the deep learning neural network; then using 70% of the rows in the input digital matrix and the output label matrix as a training set, and 30% of the rows as a test set. 10. The method according to claim 1 is characterized in that: the parameters of the deep learning neural network include the number of hidden layers k, the number of nodes n in each hidden layer, the activation function, the loss function, the accuracy calculation function, the learning rate, the batch size, the maximum number of iterations and the node random discard ratio; the deep learning neural network adopts four hidden layers; the activation function of each layer adopts the Selu function; the fourth hidden layer adopts the node random discard mechanism; The loss function and iterative optimization algorithm are defined as follows: the loss function uses the mean square error MSE, and the accuracy uses the mean square error MAE. The specific calculation formula is:

Among them, _zi is the actual value and _yi is the predicted value.

Images