A River Water Quality Spatiotemporal Prediction Method and System Based on Multi-Source Data

Abstract

The object of the present invention is to propose a spatio-temporal prediction method and system for the water quality of river channels based on multi-source data. The spatio-temporal prediction method for the water quality of river channels based on multi-source data includes the following steps: Step S1: The located type of river channel is a river channel dominated by domestic sewage; Step S2: Select the existing fixed monitoring station of the river channel as the central monitoring station; Step S3: Establish a spatial coordinate system for the river channel; Step S4: Select inspection equipment for cruise monitoring to obtain the water quality distribution; Step S5: Using temperature, wind speed, sunshine, air pressure, humidity, and rainfall meteorological factors as characteristic indicators, construct a water quality time series prediction model for the central monitoring station through deep learning methods; Step S6: According to the cruise monitoring data of the inspection equipment, use the LSTM model to construct a pollution mapping relationship model between the central monitoring station and any spatial coordinate point at different monitoring time periods; Step S7: Combine the water quality time series prediction model and the pollution mapping relationship model to perform spatio-temporal prediction on the water quality of any monitoring point of the river channel.

Description

Technical Field

[0001] The present invention relates to the technical field of water quality monitoring, in particular to a spatio-temporal prediction method and system for river water quality based on multi-source data. Background Technique

[0002] In recent years, with the increasing intensity of water pollution prevention and control, the water quality of surface water in China has been significantly improved. According to the data released by the Ministry of Ecology and Environment, the proportion of sections with excellent surface water quality (Class I-III) reached 83.4% in 2020, an increase of 15.6 percentage points compared with 2016. The 14th Five-Year Plan requires that the proportion of surface water reaching or better than Class III water bodies should reach more than 85%, and the water resource governance goal has been further improved. However, with the acceleration of the urbanization process, the pollution of urban inland rivers is becoming increasingly severe. In order to more efficiently monitor the water quality of rivers and the discharge of pollutants into rivers and improve the prevention and control efficiency of river water pollution, environmental management departments need to understand the spatio-temporal variation characteristics of river water quality more timely, accurately and comprehensively, and take measures in advance for the main discharge outlets to improve the excellent ratio of river water quality.

[0003] Generally, the water quality of rivers is affected by external pollution such as upstream incoming water, surface runoff, emissions along the way, rainfall and dustfall, as well as meteorological factors such as temperature, flow velocity, sunshine, air pressure, and humidity, resulting in differences in time and space distribution. In the existing river water quality prediction methods, it is often a time series prediction for a single monitoring point, which cannot comprehensively reflect the overall spatial distribution characteristics of water quality. Therefore, it is very important to conduct spatio-temporal prediction of river water quality.

[0004] Currently, in the spatial distribution prediction methods of water pollution concentration, the following mainly include:

[0005] One is the MIKE11AD advection-diffusion model, which is mainly used to study the spatio-temporal distribution law and diffusion attenuation process of river water quality. The MIKE11AD model calculates the diffusion and attenuation of pollutants based on the assumption that the river has a certain flow velocity and needs to meet the conditions such as uniform mixing of pollutants, linear attenuation of pollutant concentration changes, and diffusion conforming to Fick's diffusion law. However, most urban inland rivers are not directly connected to the outer river, but the water volume of the river is regulated through sluices. Rivers often have the characteristics of slow flow velocity, long residence time of pollutants, non-uniform mixing of pollutants, and non-linear variation of pollutant concentration with space; in addition, the diffusion coefficient, attenuation coefficient, etc. are affected by different river hydrological characteristics and have great uncertainties, and there are large errors in the estimation of parameters. The assumption conditions of the MIKE11AD model are no longer satisfied in the diffusion model.

[0006] Second, there is the spatial interpolation method, which is often used to convert the measurement data of discrete points into a continuous data surface for comparison with the distribution patterns of other spatial phenomena. It includes two algorithms: spatial interpolation and extrapolation. When applying the spatial interpolation method to predict the spatial distribution of river water quality, first, multiple monitoring stations need to be set up in the river. Then, after averaging the monitoring values of each station within each time period, the Kriging spatial interpolation method commonly used in statistics is used to draw the spatial distribution map of each pollution factor along the entire river to understand its spatial distribution changes. The spatial interpolation method requires multiple monitoring stations to be set up in the river, which means higher instrument monitoring costs and operation and maintenance costs, not in line with the actual situation.

[0007] Third, there is mobile monitoring. Mobile monitoring mainly includes the following three types: First, there is the mobile monitoring vehicle. The equipment of the fixed monitoring station is installed on the vehicle, and the monitoring equipment is moved to the target monitoring point by the vehicle, and the monitoring results are transmitted back to the control center. The mobile monitoring vehicle can monitor the water quality at different longitudes and latitudes of the river to understand the overall spatial distribution characteristics of the water quality. However, it cannot monitor the spatial distribution of water quality at the same moment and can only move to the next monitoring point after monitoring one point. In addition, the mobile monitoring vehicle faces problems such as large volume, troublesome deployment, and large capital and manpower investment, and also has great limitations in practical applications. Second, there is the mobile monitoring boat. The equipment of the fixed monitoring station is installed on the boat, and the boat is manually operated for mobile monitoring. Different from the vehicle-mounted method, the mobile monitoring boat can monitor the water samples in the middle of the river. Similar to the mobile monitoring vehicle, it also faces problems such as deployment, capital and manpower investment, greatly limiting the application scenarios of the mobile monitoring boat. Third, there is the small unmanned boat cruise monitoring. Water quality monitoring facilities are carried on the unmanned boat to monitor the spatial distribution characteristics of the river water quality in real time. However, in order to ensure the mobile monitoring accuracy, the monitoring equipment carried on the small unmanned boat is often expensive; and generally, the small unmanned boat needs to be remotely controlled for cruising on-site and faces problems such as anti-theft and cruise power supply. All these make the unmanned boat still unable to achieve 24-hour continuous cruising, and the application scenarios are relatively single. Currently, it is mainly used for water area inspection, sampling of target water areas, and emergency monitoring of sudden pollution, but it cannot perform long-term spatial distribution prediction of water quality.

[0008] With the development of the deep learning method, neural networks with characteristics such as non-linearity and self-organizing learning have been widely used in the prediction of water quality spatial distribution. Therefore, there is an urgent need to develop a prediction method that combines the deep learning method with the spatial distribution characteristics and time variation characteristics of water quality to more comprehensively reflect the spatio-temporal changes of water quality. Summary of the Invention

[0009] In view of the above defects, the purpose of the present invention is to propose a spatiotemporal prediction method and system for river water quality based on multi-source data. First, the monitoring station of the river is selected as the central monitoring station, and the spatial coordinate system of the river is established through water depth, longitude and latitude; then, an unmanned boat is used for cruise monitoring to obtain the water quality distribution of each spatial coordinate point; then, meteorological factors are used as characteristic indicators, and a water quality time series prediction model of the central monitoring station is constructed through a deep learning method; finally, an LSTM model is used to construct a mapping relationship model between the central monitoring station and any spatial coordinate point to realize the spatial prediction of river water quality.

[0010] To achieve this purpose, the present invention adopts the following technical scheme: A spatiotemporal prediction method for river water quality based on multi-source data comprises the following steps:

[0011] Step S1: The located river type is a river dominated by domestic sewage;

[0012] Step S2: Select an existing fixed monitoring station of a river or creek as a central monitoring station;

[0013] Step S3: Establishing a spatial coordinate system of the river, and establishing a spatial coordinate system through longitude, latitude, and depth to reflect any monitoring point of the river;

[0014] Step S4: Select patrol monitoring of patrol equipment to obtain the water quality distribution of any monitoring point in each spatial coordinate system;

[0015] Step S5: Using temperature, wind speed, sunshine, air pressure, humidity and rainfall meteorological factors as characteristic indicators, a water quality time series prediction model of the central monitoring station is constructed through a deep learning method;

[0016] Step S6: Based on the patrol monitoring data of the inspection equipment, a pollution mapping relationship model between the central monitoring station in different monitoring periods and any spatial coordinate point is constructed using the LSTM model;

[0017] Step S7: Combining the water quality time series prediction model with the pollution mapping relationship model, the water quality of any monitoring point of the river is predicted in time and space.

[0018] Preferably, the details of step S5 are as follows:

[0019] Step S51: Obtain the data of the central monitoring station and the meteorological factor data, check the data integrity, complete the missing time period of the message, and mark the monitoring value of the completed time period as a NULL value, and eliminate the missing value / NULL value, unchanged value, negative value, over-range value, minimum value below 0.3% quantile, and maximum value exceeding 99.7% quantile, which are significant abnormal values ​​that do not conform to the on-site monitoring situation;

[0020] Step S52: Exclude the data during the failure period, maintenance period, out-of-date calibration period, out-of-control period, planned maintenance period, and abnormal monitoring periods of calibration verification of the monitoring instrument, where the monitoring instrument is a tool for obtaining the data of the central monitoring station;

[0021] Step S53: Use the mean imputation method to supplement the excluded data and missing data;

[0022] Step S54: Construct a water quality time series prediction model, take the data of the central monitoring station at the previous monitoring time of the central monitoring station and meteorological factors as the characteristic indicators of the current monitoring time, and input them into the water quality time series prediction model. Through the sequence model of the monitoring time, construct the water quality time mapping relationship model for this period as: C t = F(C t-1 , weather data), where C t-1 is the data of the central monitoring station at the previous monitoring time of the central monitoring station, and the water quality time mapping relationship model outputs the data of the central monitoring station at the current moment of the central monitoring station.

[0023] Preferably, the specific steps of Step S6 are as follows:

[0024] Step S61: Obtain the data of the central monitoring station and the cruise monitoring data, check the data integrity, complete the missing periods of the message, and mark the monitoring values of the completed periods as NULL values. Exclude the missing values / NULL values, values that remain unchanged, negative values, out-of-range values, minimum values below the 0.3% quantile, and maximum values exceeding the 99.7% quantile, which are significant outliers not in line with the on-site monitoring situation;

[0025] Step S62: Exclude the data during the failure period, maintenance period, out-of-date calibration period, out-of-control period, planned maintenance period, and abnormal monitoring periods of calibration verification of the monitoring instrument, where the monitoring instrument is a tool for obtaining the data of the central monitoring station;

[0026] Step S63: Use the mean imputation method to supplement the excluded data and missing data;

[0027] Step S64: Construct a pollution mapping relationship model. The data of the central monitoring station at a certain monitoring time of the central monitoring station and the longitude, latitude, and depth of the monitoring point to be predicted are used as the characteristic indicators of this monitoring time and input into the pollution mapping relationship model. Through the LSTM model, construct the water quality spatial mapping relationship model for this period as: H t = F(C t , x, y, z), where C tFor the central monitoring station data at the current moment of the central monitoring station, where x, y, and z represent the longitude, latitude, and depth of the monitoring point respectively, the pollution mapping relationship model outputs the water quality distribution data of the monitoring object at a certain coordinate;

[0028] Step S65: Use the collected central monitoring station data and the corresponding meteorological factors as the first training data, and use the water quality distribution data, the coordinates of the monitoring points, and the central monitoring station data as the second training data;

[0029] The first training data and the second training data are randomly split into a training set and a test set in a ratio of 3:1. Use the keras of tensorflow to construct an LSTM model training framework, use Bayesian optimization for model tuning, and store the optimal parameter space prediction model; Finally, use two metrics, MAE and RMSE, to evaluate the prediction results, where where y i is the real data, is the predicted data obtained through the model.

[0030] Preferably, before performing step S64 and step S55, it is also necessary to perform standardized normalization processing on the water quality distribution data, the central monitoring station data, the monitoring time, and the meteorological factors.

[0031] Preferably, the time confirmation method for the inspection equipment to obtain the water quality distribution in step S4 is:

[0032] Obtain the time point when the central monitoring station monitors the pollutants in the river channel and use this time point as the monitoring time.

[0033] A river channel water quality spatio-temporal prediction system based on multi-source data, using the above-mentioned river channel water quality spatio-temporal prediction method based on multi-source data includes:

[0034] An object selection module, a spatial coordinate establishment module, a data acquisition module, and a prediction model module;

[0035] The object selection module is used to obtain the river channel sewage discharge type and select the river channel with the sewage discharge type of domestic sewage as the monitoring object;

[0036] The spatial coordinate establishment module is used to establish a spatial coordinate system for the monitoring object and select multiple monitoring points in the spatial coordinate system;

[0037] The data acquisition module is used to obtain the water quality distribution data of the monitoring points and the meteorological factors at that time within the specified monitoring time, select the existing fixed monitoring station in the monitoring object as the central monitoring station, and obtain the central monitoring station data of the central monitoring station;

[0038] The prediction model module is used to train based on the data obtained by the data acquisition module to obtain a water quality time series prediction model and a pollution mapping relationship model.

[0039] Preferably, it further includes a monitoring time determination module, which is used to obtain the time points when the central monitoring station monitors the pollutants in the river channel and use these time points as the time for the inspection equipment to obtain the water quality distribution.

[0040] Preferably, the prediction model module includes: a water quality distribution data integrity detection module, a central monitoring station data integrity detection module, a data completion module, a water quality time series prediction model establishment module, a pollution mapping relationship model establishment module, and a test module;

[0041] The water quality distribution data integrity detection module is used to check the integrity and effectiveness of the water quality distribution data, monitoring time, and meteorological factors, and eliminate the null values and data that do not meet the threshold requirements within the monitoring time points;

[0042] The central monitoring station data integrity detection module is used to eliminate the data during the period when the monitoring instrument is faulty, under maintenance, overdue for calibration, out of control, during planned maintenance, and during abnormal monitoring periods of calibration verification;

[0043] The data completion module is used to supplement the eliminated data and missing data using the mean imputation method;

[0044] The water quality time series prediction model establishment module is used to construct a water quality time series prediction model, and use the central monitoring station data and meteorological factors of the previous monitoring time of the central monitoring station as the characteristic indicators of the current monitoring time to train the water quality time series prediction model;

[0045] The pollution mapping relationship model establishment module is used to construct a pollution mapping relationship model, and use the central monitoring station data of the central monitoring station at a certain monitoring time, the longitude, latitude, and depth of the monitoring point to be predicted as the characteristic indicators of this monitoring time and input them into the pollution mapping relationship model to train the pollution mapping relationship model;

[0046] The test module is used to evaluate the accuracy of the water quality time series prediction model and the pollution mapping relationship model for the prediction results.

[0047] Preferably, it further includes a data normalization processing module, which is used to perform standardized normalization processing on the data processed by the data completion module.

[0048] One of the technical solutions in the above technical solutions has the following advantages or beneficial effects: This application establishes a correlation model between meteorological factors and the data of the central monitoring station in terms of time. Since the data of the central monitoring station is the result of the central monitoring station analyzing the overall concentration of pollutants in the river channel, there is a certain correlation between the water quality distribution data and the data of the central monitoring station. However, the data source of the central monitoring station is relatively single. By training with the data of the central monitoring station and the water quality distribution, the model has a high accuracy, and thus the water quality distribution data can be accurately predicted through the data of the central monitoring station. Similarly, by training with the data of the central monitoring station and meteorological factors, the model has a high accuracy, and the time series prediction values of the data of the central monitoring station at different time periods can be predicted corresponding to the meteorological factors. That is, the data of the central monitoring station in the next monitoring time is predicted through the current data of the central monitoring station and meteorological factors, and then the water quality distribution data in the next monitoring time is accurately predicted through the correlation between the data of the central monitoring station in the next monitoring time and the water quality distribution data. The accuracy of the prediction result is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic flowchart of an embodiment of the present invention.

[0050] Figure 2 It is a 24-hour pollutant change diagram of the water quality of a certain river channel;

[0051] Figure 3 It is a comparison diagram of the predicted values and the true values of different spatial coordinates of LSTM. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0053] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "axial", "radial", "circumferential", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0054] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, unless otherwise specified, the meaning of "a plurality" is two or more.

[0055] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected to" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be directly connected, or indirectly connected through an intermediate medium, and it may be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0056] As Figures 1 to 3 shown, a method for spatio-temporal prediction of river water quality based on multi-source data includes the following steps:

[0057] Step S1: The located river type is a river dominated by domestic sewage;

[0058] Since the water quality and discharge of industrial sewage are affected by factory orders and product types, the fluctuations in the discharge amount and the impact of pollutants on water quality will be relatively large. However, the discharge of domestic sewage is relatively stable and shows periodic changes in the morning, noon, and evening. The river type located in the present invention is a river dominated by domestic sewage as the monitoring object, so as to improve the accuracy of the water quality time series prediction model and the pollution mapping relationship model, and better conduct spatio-temporal prediction of the river water quality.

[0059] Step S2: Select the existing fixed monitoring station of the river as the central monitoring station;

[0060] Step S3: Establish a spatial coordinate system of the river, and establish a spatial coordinate system through longitude, latitude, and depth to reflect any monitoring point of the river;

[0061] Under the relatively static river hydrological characteristics, the mixing of river pollutants is not uniform, and the pollutant concentration shows non-linear changes with space. Generally, affected by factors such as water quality diffusion and degradation, there are certain differences in water quality at different longitudes and latitudes. The closer to the main discharge outlet, the higher the pollutant concentration usually is, and vice versa. In addition, in the vertical direction, the distribution of water quality also varies greatly. For example, in the case of the vertical distribution of dissolved oxygen, the dissolved oxygen concentration at the surface layer is the largest, and the saturation can be as high as over 200%; the dissolved oxygen at the bottom layer is less, and the saturation is only 40%-80%; the dissolved oxygen in the middle layer decreases sharply with the increase of water depth. Therefore, the present invention establishes a spatial coordinate system (x, y, z) through longitude (x), latitude (y), and depth (z) to reflect any monitoring point of the river.

[0062] Step S4: Select the inspection equipment for cruise monitoring to obtain the water quality distribution of any monitoring point in each spatial coordinate system;

[0063] During the monitoring time, a cruise device can be used to collect data on the water quality distribution at the corresponding monitoring points to obtain the water quality distribution at each monitoring time and each spatial coordinate point. And the results of the real-time monitoring by the cruise device are returned to the database with the monitoring time, the longitude of the monitoring point, the latitude of the monitoring point, the depth of the monitoring point from the water surface, and the water quality distribution data as the table headers, serving as the training data for the pollution mapping relationship model. The meteorological factors are the temperature, wind speed, sunlight, air pressure, humidity, rainfall, etc. during the collection of the water quality distribution data, and the meteorological factors are collected and returned to the database with the data of the central monitoring station as the table headers, serving as the training data for the water quality time series prediction model.

[0064] Step S5: Construct a water quality time series prediction model for the central monitoring station through deep learning methods with the temperature, wind speed, sunlight, air pressure, humidity, and rainfall meteorological factors as characteristic indicators;

[0065] Such as Figure 2 , for a river channel dominated by domestic sewage, affected by meteorological factors such as the diurnal temperature difference and solar radiation, and also affected by the peak discharge periods of residents in the morning, at noon, and in the evening, its water quality shows a 24-hour periodic change. According to the diffusion compound Fick's diffusion law of pollutants, the diffusion ability of pollutants is proportional to their concentration, that is, due to the large difference in water quality at different discharge times, the spatial mapping relationship of water quality is not the same. Therefore, in one embodiment, the monitoring periods selected for the cruise device are: 0:00, 4:00, 8:00, 12:00, 16:00, 20:00. At the same time, try to avoid cruising in heavy rain or stormy weather, on the one hand, to protect the monitoring instruments on the unmanned ship, and on the other hand, to avoid the possible short-term impact of heavy rain or storms on the water quality of the river channel.

[0066] Then, select the location for the unmanned ship to cruise and monitor, and install 3 sensors vertically for the monitored pollutants. Taking a river channel with an average depth of 2 meters as an example, the sensors are 0.5m, 1m, and 1.5m from the water surface respectively. Considering that the main discharge outlets of the river channel are generally distributed on both banks of the river channel, therefore, when cruising and monitoring in the horizontal direction, it can cruise and monitor along both banks of the river channel.

[0067] Step S6: According to the cruise monitoring data of the inspection equipment, use the LSTM model to construct a pollution mapping relationship model between the central monitoring station and any spatial coordinate point at different monitoring times;

[0068] Step S7: Combine the water quality time series prediction model and the pollution mapping relationship model to perform spatio-temporal prediction on the water quality of any monitoring point in the river channel.

[0069] If the water quality distribution is predicted using only the prediction time and meteorological factors, the relationship between the prediction time, meteorological factors and the water quality distribution data at a single monitoring point is a nonlinear relationship. The water quality distribution data at different longitudes and latitudes and depths in the monitoring point are different, and the meteorological factors include temperature, wind speed, sunshine, air pressure, humidity, rainfall and other meteorological factor data. When multiple data are associated with multiple data, the amount of data processed by the model will increase, and the relationship between the two will weaken. The accuracy of the prediction of the trained model results will decrease, and it is impossible to effectively and accurately predict the water quality distribution data of the monitoring point. This application establishes a temporal correlation model between meteorological factors and central monitoring station data. Since the central monitoring station data is the result of the central monitoring station analyzing the overall concentration of pollutants in the river, there is a certain correlation between the water quality distribution data and the central monitoring station data, and the data source of the central monitoring station is relatively single. The model accuracy obtained by training the central monitoring station data and the water quality distribution data is high, and the water quality distribution data can be accurately predicted through the central monitoring station data. Similarly, the model obtained through the training of central monitoring station data and meteorological factors has a high accuracy, and can predict the time series prediction values ​​of central monitoring station data in different time periods according to meteorological factors. That is, the central monitoring station data in the next monitoring time is predicted through the current central monitoring station data and meteorological factors, and then the water quality distribution data at the next monitoring time is accurately predicted through the association between the central monitoring station data and water quality distribution data at the next monitoring time. The accuracy of the prediction results is greatly improved, and the central monitoring station is the official unit for monitoring water quality. The accuracy and credibility of the central monitoring station data are high, and the prediction of water quality distribution data can greatly improve the accuracy of the prediction results.

[0070] Preferably, the details of step S5 are as follows:

[0071] Step S51: Obtain the data of the central monitoring station and the meteorological factor data, check the data integrity, complete the missing time period of the message, and mark the monitoring value of the completed time period as a NULL value, and eliminate the missing value / NULL value, unchanged value, negative value, over-range value, minimum value below 0.3% quantile, and maximum value exceeding 99.7% quantile, which are significant abnormal values ​​that do not conform to the on-site monitoring situation;

[0072] When the inspection equipment is acquiring water quality distribution data, it may be affected by the external environment, causing the collected data to fluctuate or the inspection equipment to fail to collect water quality distribution data. At this time, it cannot correspond to the central monitoring station data in the central monitoring station, which ultimately affects the accuracy of the model. Therefore, it is necessary to complete the missing period of the message and mark the monitoring value of the completed period as a NULL value.

[0073] Step S52: Exclude the data of the monitoring instrument during the failure period, maintenance period, uncalibrated period beyond the due date, out-of-control period, planned maintenance, and abnormal monitoring periods during calibration verification. The monitoring instrument is a tool for obtaining data from the central monitoring station;

[0074] Since the monitoring instruments of the central monitoring station will be regularly maintained by someone within the cycle to ensure the accuracy of the data measured by the monitoring instruments. Similarly, the missing or abnormal data of the central monitoring station also need to be excluded and filled.

[0075] Step S53: Use the mean imputation method to supplement the excluded data and missing data;

[0076] Step S54: Construct a water quality time series prediction model. Take the central monitoring station data of the previous monitoring time of the central monitoring station and meteorological factors as the characteristic indicators of the current monitoring time, and input them into the water quality time series prediction model. Through the time series model of the monitoring time, construct the water quality time mapping relationship model for this period as: C t = F(C t-1 , weather data), where C t-1 is the central monitoring station data of the previous monitoring time of the central monitoring station. The water quality time mapping relationship model outputs the central monitoring station data of the central monitoring station at the current moment.

[0077] Preferably, the specific steps of Step S6 are as follows:

[0078] Step S61: Obtain the central monitoring station data and cruise monitoring data, check the data integrity, complete the missing periods of the message, and mark the monitoring values of the completed periods as NULL values. Exclude the missing values / NULL values, unchanged values, negative values, out-of-range values, minimum values below the 0.3% quantile, and maximum values exceeding the 99.7% quantile, which are significantly abnormal values not in line with the on-site monitoring situation;

[0079] Since when the inspection equipment obtains the water quality distribution data, it may be affected by the external environment, causing the collected data to fluctuate or the inspection equipment fails to collect the water quality distribution data. At this time, it cannot correspond to the central monitoring station data in the central monitoring station at the time point, ultimately affecting the accuracy of the model. Therefore, it is necessary to complete the missing periods of the message and mark the monitoring values of the completed periods as NULL values.

[0080] Step S62: Exclude the data of the monitoring instrument during the failure period, maintenance period, uncalibrated period beyond the due date, out-of-control period, planned maintenance, and abnormal monitoring periods during calibration verification. The monitoring instrument is a tool for obtaining data from the central monitoring station;

[0081] Since the monitoring instruments at the central monitoring station will be regularly maintained within a period to ensure the accuracy of the data measured by the monitoring instruments. Similarly, the missing or abnormal data of the central monitoring station also needs to be removed and filled.

[0082] Step S63: Use the mean imputation method to supplement the removed data and the missing data;

[0083] The mean imputation method is to obtain the average value of the previous data and the next data of the data to be supplemented, and use the average value to supplement the data to be supplemented. When the meteorological factors change little (no heavy rain scouring occurs), the water quality distribution data and the central monitoring station data will change linearly with time. Therefore, this application uses the mean imputation method to supplement the removed and null data to ensure the integrity of the data over the entire time period.

[0084] Step S64: Construct a pollution mapping relationship model. The central monitoring station data of the central monitoring station at a certain monitoring time and the longitude, latitude, and depth of the monitoring point to be predicted are used as the characteristic indicators at this monitoring time and input into the pollution mapping relationship model. The water quality space mapping relationship model for this period is constructed through the LSTM model as: H t =F(C t , x,y,z), where C t is the central monitoring station data of the central monitoring station at the current moment, x, y, and z respectively represent the longitude, latitude, and depth of the monitoring point, and the output of the pollution mapping relationship model is the water quality distribution data of the monitoring object at a certain coordinate;

[0085] In an embodiment, the monitoring times selected for the cruise equipment are: 0:00, 4:00, 8:00, 12:00, 16:00, 20:00. Therefore, there are 6 water quality time series prediction models constructed according to the monitoring times, which are C 0：00 =F(C 22：00 , weather data), C 4：00 =F(C 0：00 , weather data), C 8：00 =F(C 4：00 , weather data), C 12：00 =F(C 8：00 , weather data), C 16：00 =F(C 12：00 , weather data), C 20：00 =F(C 16：00 , weather data). And there are also 6 corresponding pollution mapping relationship models, which are H 0：00 =F(C 0：00 , x,y,z), H 4：00 =F(C 4：00, x, y, z), H 8：00 = F(C 8：00 , x, y, z), H 12：00 = F(C 12：00 , x, y, z), H 16：00 = F(C 16：00 , x, y, z), H 20：00 = F(C 20：00 , x, y, z).

[0086] In one embodiment, it is necessary to predict the water quality distribution data of a certain monitoring point at 12:00, and the time at this moment is 4:00. At this time, it is necessary to pass through C 8：00 = F(C 4：00 , weather data) model to predict the data of the central monitoring station at 8:00. By substituting the predicted data of the central monitoring station at 8:00 into C 12：00 = F(C 8：00 , weather data), the predicted data of the central monitoring station at 12:00 is obtained. Finally, substituting the coordinates of the monitoring point and the predicted data of the central monitoring station at 12:00 into H 12：00 = F(C 12：00 , x, y, z) model, the water quality distribution data of a certain monitoring point at 12:00 can be obtained.

[0087] Step S65: Use the collected central monitoring station data and the corresponding meteorological factors as the first training data, and use the water quality distribution data, the coordinates of the monitoring point, and the central monitoring station data as the second training data;

[0088] Randomly split the first training data and the second training data into a training set and a test set at a ratio of 3:1. Use the keras of tensorflow to construct an LSTM model training framework, use Bayesian optimization for model tuning, and store the optimal parameter space prediction model; finally, use these two metrics of MAE and RMSE to evaluate the prediction results, where where y i is the real data, is the predicted data obtained through the model.

[0089] When evaluating the water quality time series prediction model, the yi is the real data of the central monitoring station data, is the predicted data of the central monitoring station data obtained through the water quality time series prediction model.

[0090] When evaluating the pollution mapping relationship model, the yi is the real data of the water quality distribution, Predicted data of the water quality distribution data obtained through the water quality time series prediction model.

[0091] Such as Figure 3 , it is tested on the real monitoring data of a certain river channel. A comparison chart of the LSTM predicted values and the real values of NH3 at 160 different spatial coordinate points is randomly selected. The root mean square error (RMSE) of the prediction is 0.338, and the mean absolute error rate (MAE) is 8.68%. The error rates of each evaluation index are relatively low, indicating that this method can accurately describe the spatial mapping relationship of water quality.

[0092] Preferably, before performing step S64 and step S55, it is also necessary to perform standardized normalization processing on the water quality distribution data, the central monitoring station data, the monitoring time, and the meteorological factors.

[0093] In the field of machine learning modeling, different evaluation indexes (that is, different features in the feature vector are the different evaluation indexes) often have different dimensions and dimension units. Such a situation will affect the results of data analysis. In order to eliminate the influence of dimensions between indexes, it is necessary to perform data standardization processing to solve the comparability between data indexes. After the original data is processed by data standardization, each index is at the same order of magnitude and is suitable for comprehensive comparative evaluation.

[0094] Preferably, the time confirmation method for the inspection equipment to obtain the water quality distribution in step S4 is:

[0095] Obtain the time point when the central monitoring station monitors the pollutants in the river channel and use this time point as the monitoring time.

[0096] In order to ensure the time correlation between the water quality distribution data and the central monitoring station data, the time for the inspection equipment to collect data is consistent with the monitoring time point of the monitoring station, and at the same time, it covers different time periods of the whole day to ensure that the water quality distribution in the monitoring points can be predicted at any time period.

[0097] A river channel water quality spatio-temporal prediction system based on multi-source data, using the above-mentioned river channel water quality spatio-temporal prediction method based on multi-source data includes:

[0098] An object selection module, a spatial coordinate establishment module, a data acquisition module, and a prediction model module;

[0099] The object selection module is used to obtain the sewage discharge type of the river channel and select the river channel with the sewage discharge type of domestic sewage as the monitoring object;

[0100] The spatial coordinate establishment module is used to establish the spatial coordinate system of the monitoring object and select multiple monitoring points in the spatial coordinate system;

[0101] The data acquisition module is used to obtain the water quality distribution data of the monitoring points and the meteorological factors at that time within the specified monitoring time, select the existing fixed monitoring station in the monitoring objects as the central monitoring station, and obtain the central monitoring station data of the central monitoring station;

[0102] The prediction model module is used to train according to the data obtained by the data acquisition module to obtain a water quality time series prediction model and a pollution mapping relationship model.

[0103] Preferably, it further includes a monitoring time determination module, which is used to obtain the time points when the central monitoring station monitors the pollutants in the river channel and use the time points as the time for the inspection equipment to obtain the water quality distribution.

[0104] Preferably, the prediction model module includes: a water quality distribution data integrity detection module, a central monitoring station data integrity detection module, a data completion module, a water quality time series prediction model establishment module, a pollution mapping relationship model establishment module, and a test module;

[0105] The water quality distribution data integrity detection module is used to check the integrity and effectiveness of the water quality distribution data, monitoring time, and meteorological factors, and eliminate the null values and data that do not meet the threshold requirements within the monitoring time points;

[0106] The central monitoring station data integrity detection module is used to eliminate the data during the period when the monitoring instrument is faulty, under maintenance, overdue for calibration, out of control, during planned maintenance, and during abnormal monitoring periods of calibration verification;

[0107] The data completion module is used to supplement the eliminated data and missing data by the mean imputation method;

[0108] The water quality time series prediction model establishment module is used to construct a water quality time series prediction model, and use the central monitoring station data and meteorological factors of the previous monitoring time of the central monitoring station as the characteristic indicators of the current monitoring time to train the water quality time series prediction model;

[0109] The pollution mapping relationship model establishment module is used to construct a pollution mapping relationship model. The central monitoring station data of the central monitoring station at a certain monitoring time, the longitude, latitude, and depth of the monitoring point to be predicted are used as the characteristic indicators of this monitoring time and input into the pollution mapping relationship model to train the pollution mapping relationship model;

[0110] The test module is used to evaluate the accuracy of the water quality time series prediction model and the pollution mapping relationship model for the prediction results.

[0111] Preferably, it further includes a data normalization processing module, which is used to perform standardized normalization processing on the data processed by the data completion module.

[0112] In the description of this specification, the descriptions with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0113] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the claims and their equivalents.

Claims

1. A spatiotemporal prediction method for river water quality based on multi-source data. It is characterized in that The following steps are involved: Step S1: The located river type is a river dominated by domestic sewage; Step S2: Select an existing fixed monitoring station of a river or creek as a central monitoring station; Step S3: Establishing a spatial coordinate system of the river, and establishing a spatial coordinate system through longitude, latitude, and depth to reflect any monitoring point of the river; Step S4: Select patrol monitoring of patrol equipment to obtain the water quality distribution of any monitoring point in each spatial coordinate system; Step S5: Using temperature, wind speed, sunshine, air pressure, humidity and rainfall meteorological factors as characteristic indicators, a water quality time series prediction model of the central monitoring station is constructed through a deep learning method; Step S6: Based on the patrol monitoring data of the inspection equipment, a pollution mapping relationship model between the central monitoring station in different monitoring periods and any spatial coordinate point is constructed using the LSTM model; Step S7: Combining the water quality time series prediction model with the pollution mapping relationship model, the water quality of any monitoring point of the river is predicted in time and space; The details of step S6 are as follows: Step S61: Obtain the central monitoring station data and cruise monitoring data, check the data integrity, complete the missing period of the message, and mark the monitoring value of the completed period as a NULL value, and eliminate missing values / NULL values, unchanged values, negative values, over-range values, minimum values ​​below the 0.3% quantile, and maximum values ​​exceeding the 99.7% quantile that do not conform to the significant abnormal values ​​of the on-site monitoring situation; Step S62: Eliminate the data of the monitoring instrument during the failure period, maintenance period, overdue uncalibrated period, out-of-control period, planned maintenance, calibration and non-normal monitoring period, where the monitoring instrument is a tool for obtaining data from the central monitoring station; Step S63: Supplement the deleted data and missing data using mean interpolation method; Step S64: Construct a pollution mapping relationship model. The central monitoring station data of the central monitoring station at a certain monitoring time and the longitude, latitude, and depth of the monitoring point to be predicted are used as the characteristic indicators at this monitoring time and input into the pollution mapping relationship model. The water quality spatial mapping relationship model for this period is constructed through the LSTM model as follows: , where is the central monitoring station data of the central monitoring station at the current moment, x, y, and z respectively represent the longitude, latitude, and depth of the monitoring point, and the output of the pollution mapping relationship model is the water quality distribution data of the monitoring object at a certain coordinate; Step S65: using the collected central monitoring station data and corresponding meteorological factors as first training data, and using the water quality distribution data, the coordinates of the monitoring points, and the central monitoring station data as second training data; The first training data and the second training data are randomly split into a training set and a test set in a ratio of 3:

1. Use the keras of tensorflow to construct an LSTM model training framework, adopt Bayesian optimization for model parameter tuning, and store the optimal parameter space prediction model; finally, use these two metrics, MAE and RMSE, to evaluate the prediction results, where , , where is the real data, is the predicted data obtained through the model.

2. A spatiotemporal prediction method for river water quality based on multi-source data according to claim 1, It is characterized in that The details of step S5 are as follows: Step S51: Obtain the data of the central monitoring station and the meteorological factor data, check the data integrity, complete the missing time period of the message, and mark the monitoring value of the completed time period as a NULL value, and eliminate the missing value / NULL value, unchanged value, negative value, over-range value, minimum value below 0.3% quantile, and maximum value exceeding 99.7% quantile, which are significant abnormal values ​​that do not conform to the on-site monitoring situation; Step S52: Eliminate the data of the monitoring instrument during the failure period, maintenance period, overdue uncalibrated period, out-of-control period, planned maintenance, calibration and non-normal monitoring period, where the monitoring instrument is a tool for obtaining data from the central monitoring station; Step S53: Supplement the deleted data and missing data using mean interpolation method; Step S54: Construct a water quality time series prediction model. Use the data of the central monitoring station at the previous monitoring time and meteorological factors of the central monitoring station as the characteristic indicators at the current monitoring time, and input them into the water quality time series prediction model. Construct a water quality time mapping relationship model for this period through the sequence model of the monitoring time as follows: , where is the data of the central monitoring station at the previous monitoring time of the central monitoring station, and the water quality time mapping relationship model outputs the data of the central monitoring station at the current moment.

3. The spatiotemporal prediction method of river water quality based on multi-source data according to claim 1, It is characterized in that Before performing step S64 and step S55, it is also necessary to perform standard normalization processing on the water quality distribution data, central monitoring station data, monitoring time, and meteorological factors.

4. A method for predicting the temporal and spatial water quality of a river channel based on multi-source data according to claim 1, wherein, the time confirmation method for the inspection equipment to obtain the water quality distribution in step S4 is: Obtain the time point when the central monitoring station monitors the pollutants in the river channel and use this time point as the monitoring time.

5. A system for predicting the temporal and spatial water quality of a river channel based on multi-source data, using the method for predicting the temporal and spatial water quality of a river channel based on multi-source data according to any one of claims 1 to 4, wherein, it includes: an object selection module, a spatial coordinate establishment module, a data acquisition module, and a prediction model module; the object selection module is used to obtain the sewage discharge types of the river channels and select the river channels with the sewage discharge type of domestic sewage as the monitoring objects; the spatial coordinate establishment module is used to establish the spatial coordinate system of the monitoring objects and select multiple monitoring points in the spatial coordinate system; the data acquisition module is used to obtain the water quality distribution data of the monitoring points and the meteorological factors at that time within the specified monitoring time, select the existing fixed monitoring station in the monitoring object as the central monitoring station, and obtain the central monitoring station data of the central monitoring station; the prediction model module is used to train according to the data obtained by the data acquisition module to obtain a water quality time series prediction model and a pollution mapping relationship model.

6. A system for predicting the temporal and spatial water quality of a river channel based on multi-source data according to claim 5, wherein, it further includes a monitoring time determination module, and the monitoring time determination module is used to obtain the time point when the central monitoring station monitors the pollutants in the river channel and use this time point as the time for the inspection equipment to obtain the water quality distribution.

7. A system for predicting the temporal and spatial water quality of a river channel based on multi-source data according to claim 5, wherein, the prediction model module includes: a water quality distribution data integrity detection module, a central monitoring station data integrity detection module, a data complementation module, a water quality time series prediction model establishment module, a pollution mapping relationship model establishment module, and a test module; the water quality distribution data integrity detection module is used to check the integrity and effectiveness of the water quality distribution data, monitoring time, and meteorological factors, and eliminate the null values and data that do not meet the threshold requirements within the monitoring time point; the central monitoring station data integrity detection module is used to eliminate the data during the period when the monitoring instrument is faulty, under maintenance, overdue for calibration, out of control, during planned maintenance, and during abnormal monitoring periods of calibration verification; the data complementation module is used to supplement the eliminated data and missing data using the mean imputation method; the water quality time series prediction model establishment module is used to construct a water quality time series prediction model, and use the central monitoring station data of the previous monitoring time of the central monitoring station and the meteorological factors as the characteristic indicators of the current monitoring time to train the water quality time series prediction model; The pollution mapping relationship model establishment module is used to construct a pollution mapping relationship model. The data of the central monitoring station at a certain monitoring time and the longitude, latitude, and depth of the monitoring point to be predicted are used as the characteristic indicators at this monitoring time and input into the pollution mapping relationship model for training. The testing module is used to evaluate the accuracy of the water quality time series prediction model and the pollution mapping relationship model for the prediction results.

8. A river channel water quality spatio-temporal prediction system based on multi-source data according to claim 7, characterized in that, it further includes a data normalization processing module, and the normalization processing module is used to perform standard normalization processing on the data processed by the data completion module.

Images