Hyperspectral short-term prediction and early warning method and system for water quality and algal bloom of near-sensing lake and reservoir

Abstract

The application discloses a hyperspectral near-sensing lake and reservoir water quality and water bloom short-term prediction and early warning method and system, and belongs to the technical and method field of water quality prediction and early warning. The water body near-sensing hyperspectral reflectivity after pretreatment is combined with space-time attribute information to generate a water body near-sensing hyperspectral reflectivity-time sequence dataset; the water quality parameters measured synchronously are combined, a preset water quality inversion model set is constructed according to a space-time matching principle; a multi-water quality parameter-time sequence dataset is generated by using the preset water quality inversion model set; a prediction and early warning model is built based on the multi-water quality parameter-time sequence dataset, and water quality and water bloom prediction and early warning results are output. The application can realize real-time transmission of measurement data to a cloud server, conversion of hyperspectral reflectivity data into water quality parameters which are easy for users to understand by constructing a water quality inversion model set, integration of the data on a software platform based on multi-parameter water quality, and real-time and on-the-spot checking of current water quality information, long-term historical changes and future short-term water bloom prediction and early warning information.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of water environment monitoring and early warning, and particularly relates to a high-spectral near-sensing lake and reservoir water quality and water bloom short-term prediction and early warning method and system. BACKGROUND

[0002] Water quality and water bloom high-frequency monitoring and accurate prediction are important basis and cornerstone for mastering lake and reservoir water environment dynamic characteristics, analyzing change trend, and carrying out blue-green algae bloom cause mechanism analysis and scientific prevention and control. However, traditional manual patrol measurement lags behind the needs of water environment management and decision-making departments in data acquisition frequency, timeliness and representativeness; satellite and unmanned aerial vehicle remote sensing are limited by spatial and spectral resolution, cloud and rain weather and atmospheric correction, and cannot meet the needs of continuous water quality and water bloom monitoring; high-frequency underwater probes are limited by high cost, easy pollution, low precision and difficult calibration, and cannot guarantee monitoring accuracy and continuity. Therefore, the existing water quality and water bloom monitoring methods face challenges in capturing sudden water pollution events and rapid water ecological disturbance processes caused by extreme weather. In addition, blue-green algae bloom is the result of the coupling of water physics, chemistry and biology and meteorological and hydrological factors, and the mechanism is complex and the influencing factors are numerous. Traditional blue-green algae bloom prediction models based on ecological dynamics have problems such as complex form, numerous parameters and unclear mechanism; statistical models based on water environment data have problems such as simple form, numerous parameters, and poor continuity of input water environment data, mainly relying on daily manual and satellite inversion data, resulting in unsatisfactory prediction model accuracy and low automation and intelligence level.

[0003] In view of the problems in lake and reservoir water quality and water bloom monitoring and prediction under the dual influence of current high-intensity human activities and rapid climate change, an original, minute-level high-spectral near-sensing (non-contact near-surface sensing) water quality instrument is developed and utilized, multi-scene measured water quality data are coupled, a high-precision inversion model of key water quality parameters of lakes and reservoirs is upgraded, high-frequency monitoring of water quality and water bloom under complex weather and water conditions is realized, a deep learning algorithm with strong learning and nonlinear approximation ability is introduced, a real-time prediction and early warning model of water quality and water bloom driven by big data is constructed, and the intelligent level of ecological environment monitoring and the disaster emergency handling capability are improved. SUMMARY

[0004] The application provides a high-spectral near-sensing lake and reservoir water quality and water bloom short-term prediction and early warning method and system to solve the technical problems in the background.

[0005] The application adopts the following technical scheme: a high-spectral near-sensing lake and reservoir water quality and water bloom short-term prediction and early warning method, comprising the following steps:

[0006] The water body proximal hyperspectral reflectance is collected in real time according to the spatio-temporal attribute information, and preprocessed to obtain preprocessed water body proximal hyperspectral reflectance; the preprocessed water body proximal hyperspectral reflectance is combined with the spatio-temporal attribute information to generate a water body proximal hyperspectral reflectance-time series data set; and the spatio-temporal attribute information at least includes spatial information, time information and water body type of a monitoring point;

[0007] According to the water body type of the monitoring point, corresponding water body proximal hyperspectral reflectance is screened out, and a preset water quality inversion model set related to the water body type of the monitoring point is constructed according to the spatio-temporal matching principle, combined with the simultaneously measured water quality parameters; the water body proximal hyperspectral reflectance-time series data set is taken as an input variable, and the measured water quality parameters are taken as an output dependent variable, and the preset water quality inversion model set is used to generate a multi-water quality parameter-time series data set;

[0008] Based on the multi-water quality parameter-time series data set, statistical analysis of multiple time scales is performed, and a key multi-water quality parameter-time series data set is screened out to build a prediction and early warning model; the preprocessed water body proximal hyperspectral reflectance is taken as an input feature, and a water quality and water bloom prediction and early warning result is output; the key multi-water quality parameter-time series data set is updated according to the water quality and water bloom prediction and early warning result, and the prediction and early warning model is optimized;

[0009] The user's query for prediction and early warning demand of different time scales is received at a web terminal, and the above steps are repeated to update and display the output result in real time.

[0010] In further embodiments, the collection process of the water body proximal hyperspectral reflectance is as follows:

[0011] The water surface downward irradiance is obtained by using a transmittance plate with a transmittance of : ; wherein, is the irradiance of the transmittance plate;

[0012] The collected water body proximal hyperspectral reflectance is obtained by using the following formula: ; wherein, is the water-off irradiance of the transmittance plate.

[0013] In further embodiments, the preprocessing steps of the preprocessed water body proximal hyperspectral reflectance are as follows:

[0014] A spectral curve is drawn according to the water body proximal hyperspectral reflectance, the spectral curve is taken as a vector, and the length of each vector is calculated :

[0015] ; wherein, ,​​ Real-time collection of water near-sensing hyperspectral reflectance based on spatiotemporal attribute information;

[0016] The normalized near-sensing hyperspectral reflectance is calculated by using the following formula to calculate the ratio of each wavelength to the mode length

[0017] ;

[0018] The normalized near-sensing hyperspectral reflectance is linearly corrected by using the following formula

[0019] ; wherein, is the near-sensing hyperspectral reflectance after radiation correction, represents a radiation correction coefficient, is a radiation correction slope.

[0020] In further embodiments, the monitoring point water body type at least includes: clean reservoir, natural river, urban river and lake;

[0021] The synchronous measured water quality parameters at least include: total nitrogen, total phosphorus, chlorophyll a, transparency, water temperature, dissolved oxygen, permanganate index, turbidity and total suspended solids concentration;

[0022] The preset water quality inversion model set at least includes: near-sensing hyperspectral total phosphorus inversion model, near-sensing hyperspectral total nitrogen inversion model, near-sensing hyperspectral transparency inversion model, near-sensing hyperspectral total suspended solids concentration inversion model, near-sensing hyperspectral water temperature inversion model and near-sensing hyperspectral dissolved oxygen inversion model.

[0023] In further embodiments, the spatiotemporal matching principle at least includes: time matching principle and space matching principle;

[0024] The time matching principle is represented as follows: the collection time point of the water near-sensing hyperspectral reflectance is defined as , the detection time point of the synchronous water quality parameter is , the relationship between the collection time point and the detection time point should be as follows: ;

[0025] The space matching principle includes: longitude and latitude matching principle and distance matching principle; the longitude and latitude matching principle is represented as follows: , wherein, is the longitude and latitude of the collection point of the water near-sensing hyperspectral reflectance, represents the reduced dimension of the detection point of the synchronous water quality parameter;

[0026] The distance matching principle is represented as follows:​ , wherein, is a standard observation range radius, represents the horizontal distance from the water sample collection point to the monitoring center of the proximal hyperspectral instrument; wherein, , is the height of the proximal hyperspectral instrument placed on the water surface, represents the field of view angle.

[0027] In further embodiments, the multi-time scale at least includes: hourly scale and daily scale, and the key multi-water quality parameter-time series dataset includes: hourly water quality parameters and daily water quality parameters; the screening process of the key multi-water quality parameter-time series dataset is as follows: the correlation between the hourly water quality parameters and the synchronous chlorophyll a concentration, and the correlation between the daily water quality parameters and the synchronous chlorophyll a concentration are introduced , and whether the corresponding water quality parameter is a key parameter is judged by the assignment of the correlation

[0028] ; wherein, is the Pearson correlation coefficient, is the significance level;

[0029] If the assignment of the correlation is 1, it indicates that the multi-water quality parameter and the synchronous chlorophyll a concentration are significantly correlated, which is a key water quality parameter and is updated to the key multi-water quality parameter-time series dataset; if the assignment of the correlation is 0, it indicates that the multi-water quality parameter and the synchronous chlorophyll a concentration are weakly correlated, which is not a key water quality parameter.

[0030] In further embodiments, the analysis process of the prediction and early warning model is as follows:

[0031] The sliding window is defined as , and the sliding step is The mean , standard deviation and median of the multi-water quality parameters in the best sliding window are calculated, and the water quality parameter at the center position of the best sliding window is analyzed :

[0032] If , the water quality and water bloom prediction and early warning result indicates that the water quality is normal, and no warning processing is performed;

[0033] If , the water quality and water bloom prediction and early warning result indicates that the water quality is abnormal, and an alarm information is issued,

[0034] In further embodiments, the matching logic relationship of the space-time matching principle is as follows:​

[0035] assigning the logical judgment result of the spatio-temporal matching principle to When the time matching principle and the space matching principle are both satisfied, then corresponding synchronous measured water quality parameters and water body near-sensing hyperspectral reflectance are added to the data for model construction;

[0036] On the contrary, when at least one of the time matching principle and the space matching principle is not satisfied, then 0 is assigned to , corresponding synchronous measured water quality parameters and water body near-sensing hyperspectral reflectance are deleted;

[0037] .

[0038] In further embodiments, further comprising: corresponding water quality parameters are removed from the key multi-water quality parameter-time series data set and are replaced with the median value;

[0039] The multi-water quality parameter-time series data set is smoothed by using a Savitzky-Golay filter to reduce noise in the collection process, a 5th order polynomial curve fitting is used, and a nearest neighbor interpolation method is used for filling, thereby obtaining an updated key multi-water quality parameter-time series data set.

[0040] The hyperspectral near-sensing lake and reservoir water quality and algal bloom short-term prediction and early warning system is used to implement the water quality and algal bloom short-term prediction and early warning method as described above, and comprises:

[0041] A first module is configured to collect water body near-sensing hyperspectral reflectance in real time according to spatio-temporal attribute information and obtain preprocessed water body near-sensing hyperspectral reflectance after preprocessing; the preprocessed water body near-sensing hyperspectral reflectance is combined with spatio-temporal attribute information to generate a water body near-sensing hyperspectral reflectance-time series data set; the spatio-temporal attribute information at least includes spatial information, time information, and water body type of a monitoring point;

[0042] A second module is configured to filter out corresponding water body near-sensing hyperspectral reflectance according to the water body type of the monitoring point, combine synchronous measured water quality parameters, construct a preset water quality inversion model set related to the water body type of the monitoring point according to a spatio-temporal matching principle, use the water body near-sensing hyperspectral reflectance-time series data set as an input variable, use the measured water quality parameters as an output dependent variable, and use the preset water quality inversion model set to generate a multi-water quality parameter-time series data set;

[0043] ​The third module is configured to perform statistical analysis of multiple time scales based on the multi-water quality parameter-time sequence data set, filter out key multi-water quality parameter-time sequence data set, and build a prediction and early warning model; the preprocessed water body near-sensing hyperspectral reflectance is taken as an input feature, and a water quality and water bloom prediction and early warning result is output; the key multi-water quality parameter-time sequence data set is updated according to the water quality and water bloom prediction and early warning result, and the prediction and early warning model is optimized;

[0044] The fourth module is configured to receive user query of prediction and early warning demand of different time scales on a webpage terminal, and repeatedly update and display the output result in real time.

[0045] The present application has the following advantages: the present application uses a near-sensing hyperspectral instrument to realize real-time high-frequency collection of water body hyperspectral data, and transmits the measurement data to a cloud server for processing in real time; a water quality inversion model set is constructed to convert the hyperspectral reflectance data into water quality parameters (such as total nitrogen, chlorophyll content, dissolved oxygen, etc.) that are easy for users to understand, and these data are integrated on a software platform based on multi-parameter water quality;

[0046] The user only needs to install the software platform on the terminal to view the current water quality information, historical long-term changes and future short-term water bloom prediction and early warning information at any time and anywhere without the need to obtain information through multiple channels or complex processes; this water quality monitoring method not only improves the efficiency and accuracy of information acquisition, but also reduces the monitoring cost and expands the monitoring range, thereby providing strong technical support for water quality monitoring and environmental protection work. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is a flowchart of a short-term prediction and early warning method for lake and reservoir water quality and water bloom based on hyperspectral near-sensing. DETAILED DESCRIPTION

[0048] The present application will be further described below in combination with the drawings and examples of the specification.

[0049] Example 1

[0050] Blue-green algae bloom is the result of the coupling of physical, chemical and biological factors and meteorological and hydrological factors, and the mechanism is complex and the influencing factors are numerous. The traditional blue-green algae bloom prediction model based on ecological dynamics has problems such as complex form, numerous parameters and unclear part of mechanism; the statistical model based on water environment data has problems such as simple form, various parameters, poor continuity of input water environment data, low prediction model accuracy, low automation and low intelligence.

[0051] In order to solve this problem, the present embodiment provides a short-term prediction and early warning method for lake and reservoir water quality and water bloom based on hyperspectral near-sensing, as shown in Figure 1 The method comprises the following steps:

[0052] The near-sensing hyperspectral monitor is placed on the water surface based on a multi-carrying platform including a land-based, ship-based and tower-based platform, near-sensing hyperspectral reflectance of the water body is collected in real time according to space-time attribute information, and preprocessed near-sensing hyperspectral reflectance of the water body is obtained; the preprocessed near-sensing hyperspectral reflectance of the water body is combined with the space-time attribute information to generate a near-sensing hyperspectral reflectance-time series data set; the space-time attribute information at least includes spatial information, time information and water body type of the monitoring point. In further embodiments, the spatial information can be understood as geographic longitude and latitude information, and the time information can be year, month, day, hour, minute and second when the near-sensing hyperspectral reflectance is collected. In order to obtain stable and reliable near-sensing hyperspectral reflectance of the water body, there are strict requirements for the installation environment of the near-sensing hyperspectral reflectance: 1) there are no trees, buildings and shadow obstructions around the equipment; 2) there are no high white walls, reflective glass and other items that can cause spectral changes within a horizontal range of 2 meters; 3) the water depth of the spectral collection area is greater than the water transparency, and the bottom cannot be seen; 4) there are no long-term fixed white waves, foam, aquatic vegetation and other non-water body items in the spectral collection area; 5) the near-sensing hyperspectral lens should form an angle of 135°~180° with the plane where the sun is located; 6) the near-sensing hyperspectral monitor is placed at an observation distance of about 2~10 meters above the water surface.

[0053] According to the water body type of the monitoring point, the corresponding near-sensing hyperspectral reflectance of the water body is screened out, combined with the simultaneously measured water quality parameters, and a preset water quality inversion model set related to the water body type of the monitoring point is constructed according to the space-time matching principle; the near-sensing hyperspectral reflectance-time series data set is used as an input variable, and the measured water quality parameters are used as an output dependent variable, and a multi-water quality parameter-time series data set is generated by using the preset water quality inversion model set;

[0054] Based on the multi-water quality parameter-time series data set, statistical analysis of multiple time scales is performed, key multi-water quality parameter-time series data sets are screened out, and a prediction and early warning model is built; the preprocessed near-sensing hyperspectral reflectance of the water body is used as an input feature, and a water quality and algal bloom prediction and early warning result is output; the key multi-water quality parameter-time series data set is updated according to the water quality and algal bloom prediction and early warning result, and the prediction and early warning model is optimized;

[0055] In the webpage terminal, user inquiries of different time scales for prediction and early warning are received, and the above steps are repeated to update and display the output results in real time.

[0056] Based on the above description, the collection process of the near-sensing hyperspectral reflectance of the water body is as follows:

[0057] The downwelling irradiance of the water surface is obtained by using a transmittance plate with a transmittance of : : ; wherein, is the irradiance of the transmittance plate.

[0058] The following formula is used to obtain the collected water near-sensing hyperspectral reflectance : ; wherein, is the water-side irradiance of the transmission plate.

[0059] It is worth mentioning that the hyperspectral instrument is inclined at an angle of 45° to the horizontal direction, the azimuth angle is between 90-135°, and the lens is loaded with a polarizer to filter stray light from other directions and water quality without light, thereby improving the signal-to-noise ratio. The high-definition video camera is a circular camera ball machine that can realize 360° horizontal rotation and 180° up-down rotation, so as to capture and record the current water environment and query the historical water surface conditions, thereby assisting in confirming and judging the abnormal water environment conditions.

[0060] Considering that the apparent optical properties of the water body in the field change with the change of illumination conditions, and are affected by wind waves, solar elevation angle, illumination intensity and other factors, it is necessary to perform normalization and radiation correction on the near-sensing hyperspectral water quality reflectance data for pretreatment, so as to eliminate the influence of different weather, wind speed and illumination intensity on the amplitude of the spectral curve. The pretreatment at least includes normalization and radiation correction; the normalization adopts a ratio normalization method; the radiation correction adopts a cross-calibration method, and a linear function is used for calibration.

[0061] Further, the pretreatment steps of the water near-sensing hyperspectral reflectance after pretreatment are as follows:

[0062] According to the water near-sensing hyperspectral reflectance, a spectral curve is drawn, the spectral curve is regarded as a vector, and the length of each vector is calculated :

[0063] ; wherein, , is the water near-sensing hyperspectral reflectance collected based on the real-time spatial and temporal attribute information;

[0064] The following formula is used to calculate the ratio of each wavelength to the length to obtain the normalized near-sensing hyperspectral reflectance :

[0065] ;

[0066] The following formula is used to linearly correct the normalized near-sensing hyperspectral reflectance :

[0067] ; wherein, is the near-sensing hyperspectral reflectance after radiation correction, represents a radiation correction coefficient, is a radiation correction slope. In the embodiment, the value of the total nitrogen is 1.086, the value of the total phosphorus is 0.0054.

[0068] In another embodiment, the monitoring point water body type at least includes: clean reservoir, natural river, urban river and lake. Correspondingly, the synchronous water quality parameter at least includes: total nitrogen, total phosphorus, chlorophyll a, transparency, water temperature, dissolved oxygen, permanganate index, turbidity and total suspended substance concentration. Correspondingly, the multi-water quality parameter-time series data set is the parameter-time series data set about nitrogen, total phosphorus, chlorophyll a, transparency, water temperature, dissolved oxygen, permanganate index, turbidity and total suspended substance concentration.

[0069] The preset water quality inversion model set at least includes: near-sensing hyperspectral total phosphorus inversion model, near-sensing hyperspectral total nitrogen inversion model, near-sensing hyperspectral transparency inversion model, near-sensing hyperspectral total suspended substance concentration inversion model, near-sensing hyperspectral water temperature inversion model and near-sensing hyperspectral dissolved oxygen inversion model. Among them, the model in the preset water quality inversion model set is based on attribute information to filter out urban river, natural river, reservoir and natural lake water body near-sensing hyperspectral reflectance data in the water body near-sensing hyperspectral time series data set, and increase the water body attribute type label; according to the space-time matching principle, the synchronous measured water quality parameter is found, and the high-spectral water quality quasi-synchronous data set is generated; according to the distribution principle of 3:1, the generated hyperspectral in the modeling data set and the verification data set is input as the input feature, and the synchronous water quality parameter is input as the output feature, and the multi-parameter water quality inversion model is constructed by inputting into the machine learning algorithm. The machine learning algorithm in this embodiment can be extreme gradient boosting tree algorithm, support vector machine, random forest algorithm, deep neural network or other.

[0070] In another embodiment, the space-time matching principle at least includes: time matching principle and space matching principle;

[0071] The time matching principle is represented as follows: the collection time point of the water body near-sensing hyperspectral reflectance is , the detection time point of the synchronous water quality parameter is , the relationship between the collection time point and the detection time point should be as follows: ;

[0072] The space matching principle includes: longitude and latitude matching principle and distance matching principle; the longitude and latitude matching principle is represented as follows: , wherein, is the longitude and latitude of the collection point of the water body near-sensing hyperspectral reflectance, represents the dimension reduction of the detection point of the synchronous water quality parameter;

[0073] The distance matching principle is represented as follows: , wherein, is a standard observation range radius, represents a horizontal distance from a water sample collection point to a near-sensing hyperspectral instrument monitoring center; wherein, , is a height at which the near-sensing hyperspectral instrument is placed on the water surface, represents a field of view angle.

[0074] Based on the above description, the matching logic relationship of the space-time matching principle is as follows:

[0075] The logical judgment result of the space-time matching principle is assigned to When the time matching principle and the space matching principle are both satisfied, then The corresponding synchronous measured water quality parameter and water body near-sensing hyperspectral reflectivity are added to the data for model construction;

[0076] On the contrary, when at least one of the time matching principle and the space matching principle is not satisfied, then 0 is assigned to , The corresponding synchronous measured water quality parameter and water body near-sensing hyperspectral reflectivity are deleted;

[0077] .

[0078] In further embodiments, the multiple time scales at least include: hourly scale and daily scale, and the key multi-water quality parameter-time series data set includes: hourly water quality parameters and daily water quality parameters. Since the hourly multi-parameter water quality time series data set is continuous data, in this embodiment, the 3 standard deviation criterion is used for outlier processing, the median replacement method is selected for missing value interpolation, or the Savitzky-Golay filter is used for smoothing processing to reduce data abnormalities and noise caused by wind waves, solar height and flares.

[0079] The screening process of the key multi-water quality parameter-time series data set is as follows: the correlation between the hourly water quality parameters and the synchronous chlorophyll a concentration, and the correlation between the daily water quality parameters and the synchronous chlorophyll a concentration are introduced The assignment of the correlation is used to determine whether the corresponding water quality parameter is a key parameter:

[0080] ; wherein, is a Pearson correlation coefficient, is a significance level;

[0081] If the correlation If the value is assigned to 1, it indicates that the multi-water quality parameter is significantly, moderately or greater, correlated with the synchronous chlorophyll a concentration, and is considered a key water quality parameter, thus being updated in the key multi-water quality parameter-time series dataset; if the correlation... If the value is 0, it means that the water quality parameter is weakly correlated with the concentration of chlorophyll a and is not a key water quality parameter.

[0082] It should be noted that, since the growth and reproduction of phytoplankton are closely related to the aquatic environment, the relationship between transparency, total nitrogen, total phosphorus, permanganate index, and temperature retrieved from near-sensor hyperspectral data and chlorophyll a was analyzed. The results showed that both daily and hourly water quality parameters were highly significantly correlated with synchronous chlorophyll a, with the highest correlation to the permanganate index, followed by total nitrogen and then total phosphorus. However, the correlation between chlorophyll a concentration retrieved from both hourly and daily data and other water quality parameters was greater than 0.3, thus establishing the correlation between water quality parameters and synchronous chlorophyll a concentration. Based on the screening of the above key multi-water quality parameter time series datasets, the analysis process of the prediction and early warning model is as follows:

[0083] Define the sliding window as The sliding step size is ,like , Calculate the average values ​​of multiple water quality parameters within the optimal sliding window. Standard deviation and median Analyze the water quality parameters at the center of the optimal sliding window. :

[0084] like If the water quality algal bloom prediction and early warning result is normal, no warning action will be taken.

[0085] like If the water quality algal bloom prediction and early warning result is indicated as water quality abnormality, an alarm message will be issued.

[0086] Furthermore, Corresponding water quality parameters Removed from key multi-water quality parameter time series datasets and used as the median. Substitute;

[0087] The Savitzky-Golay filter was used to smooth the multi-water quality parameter time series dataset to reduce noise during the acquisition process. Fifth-order polynomial curve fitting was used, and nearest neighbor interpolation was used for filling to obtain the updated key multi-water quality parameter time series dataset.

[0088] It is worth mentioning that the prediction and early warning model in this embodiment is a 4-layer long short-term memory network model, i.e., an input layer, two hidden layers of memory modules, and a fully connected output layer. 90% of the input key multi-water quality parameter time series data set is used for modeling, and the remaining 10% of the input key multi-water quality parameter time series data set is used for verification. The total nitrogen, total phosphorus, chlorophyll a, transparency, suspended solids, permanganate index, turbidity, and extinction coefficient in the past time steps are input features.

[0089] The root mean square error and the mean absolute percentage error between the output chlorophyll a and the measured chlorophyll a are calculated to evaluate the accuracy and robustness of the model. In addition, the hyperparameters involved in the long short-term memory time network model are enumerated and compared using grid search method, and the model is evaluated using 5-fold cross-validation method to select the optimal hyperparameters: the number of LSTM neurons in 2 layers is 100 and 150 respectively, the optimizer is adam, the activation function is sigmoid, and the model iteration number is set to 60 times. The prediction time step is 1h, 2h, 3h, 1d, 2d, and 3d.

[0090] The past time step is between 2 and 30, and the optimal step refers to the past time step when the error of the long short-term memory network model based on the past time step of 2 to 30 is the smallest. At this time, the prediction and early warning model is also the optimal model.

[0091] The water quality data displayed to the user by the method is the water quality data collected and updated in real time at the current monitoring point, and the method also provides the rapid change of water quality at different time scales during the historical operation, ensuring that the user can comprehensively understand the current and past water quality change information.

[0092] The method supports web interaction, and the user can select query content according to his own needs and concerns; that is, the user can intuitively view the current interest point water quality condition, and can also conveniently view the historical water quality time series change and future short-term water bloom change information, greatly improving the query and monitoring efficiency and improving the auxiliary decision-making level.

[0093] The method can realize real-time high-frequency monitoring, display, and prediction and early warning of water quality information, ensuring that the user obtains the latest and most accurate information to assist the user in decision-making. Compared with the information lag problem existing in traditional manual monitoring, the method provides higher real-time and forward-looking.

[0094] The method displays the water quality information of interest to the user according to the user's needs and concerns; this means that the user can intuitively obtain effective information without being disturbed by other information.

[0095] Embodiment 2

[0096] The embodiment provides a hyperspectral near-sensing lake and reservoir water quality and water bloom short-term prediction and early warning system, which is used for realizing the water quality and water bloom short-term prediction and early warning method as described in Embodiment 1, and comprises the following modules.

[0097] A first module is configured to collect water body near-sensing hyperspectral reflectivity in real time according to space-time attribute information, and obtain preprocessed water body near-sensing hyperspectral reflectivity through preprocessing; the preprocessed water body near-sensing hyperspectral reflectivity is combined with space-time attribute information to generate a water body near-sensing hyperspectral reflectivity-time sequence dataset; the space-time attribute information at least comprises spatial information, time information and water body type of a monitoring point;

[0098] A second module is configured to screen out corresponding water body near-sensing hyperspectral reflectivity according to the water body type of the monitoring point, combine synchronous measured water quality parameters, and construct a preset water quality inversion model set related to the water body type of the monitoring point according to a space-time matching principle; the water body near-sensing hyperspectral reflectivity-time sequence dataset is used as an input variable, and the measured water quality parameters are used as an output dependent variable, and the preset water quality inversion model set is used to generate a multi-water quality parameter-time sequence dataset;

[0099] A third module is configured to perform statistical analysis of multiple time scales based on the multi-water quality parameter-time sequence dataset, screen out a key multi-water quality parameter-time sequence dataset, and build a prediction and early warning model; the preprocessed water body near-sensing hyperspectral reflectivity is used as an input feature, and a water quality and water bloom prediction and early warning result is output; the key multi-water quality parameter-time sequence dataset is updated according to the water quality and water bloom prediction and early warning result, and the prediction and early warning model is optimized;

[0100] A fourth module is configured to receive user query of prediction and early warning requirements of different time scales through a webpage terminal, and repeatedly update and display the output result in real time through the above steps.

Claims

1. A short-term prediction and early warning method for water quality and algal bloom in hyperspectral near-sensing lake and reservoirs, characterized in that, Includes the following steps: The near-sensory hyperspectral reflectance of water bodies is collected in real time based on spatiotemporal attribute information and preprocessed to obtain the preprocessed near-sensory hyperspectral reflectance of water bodies. The preprocessed water body near-sensory hyperspectral reflectance is combined with spatiotemporal attribute information to generate a water body near-sensory hyperspectral reflectance-time series dataset. The spatiotemporal attribute information includes: spatial information, temporal information, and the water body type at the monitoring point; Based on the water body type at the monitoring point, the corresponding near-sensory hyperspectral reflectance of the water body is selected. Combined with synchronously measured water quality parameters, a set of preset water quality inversion models related to the water body type at the monitoring point is constructed according to the spatiotemporal matching principle. The near-sensory hyperspectral reflectance-time series dataset of the water body is used as the input variable and the measured water quality parameters are used as the output dependent variable. The set of preset water quality inversion models is used to generate a multi-water quality parameter-time series dataset. Statistical analysis at multiple time scales is performed based on the aforementioned multi-water quality parameter-time series dataset to select key multi-water quality parameter-time series datasets and build a prediction and early warning model. The preprocessed near-sensory hyperspectral reflectance of the water body is used as the input feature to output the prediction and early warning results of water quality and algal blooms. The key multi-water quality parameter-time series dataset is updated based on the prediction and early warning results of water quality and algal blooms, and the prediction and early warning model is optimized. The multi-time scale includes a time-by-time scale and a day-by-day scale, the key multi-water quality parameter-time series dataset includes time-by-time water quality parameters and day-by-day water quality parameters, and the screening process of the key multi-water quality parameter-time series dataset is as follows: the correlation between the time-by-time water quality parameters and the synchronous chlorophyll a concentration and the correlation between the day-by-day water quality parameters and the synchronous chlorophyll a concentration are introduced , and whether the corresponding water quality parameter is a key parameter is judged through the assignment of the correlation . ; wherein, is the Pearson correlation coefficient, is the significance level; If the assignment of the correlation is 1, it indicates that the multi-water quality parameter is significantly and moderately correlated with the synchronous chlorophyll a concentration, belongs to the key water quality parameter, and is updated to the key multi-water quality parameter-time series data set; if the assignment of the correlation is 0, it indicates that the multi-water quality parameter is weakly correlated with the synchronous chlorophyll a concentration, and does not belong to the key water quality parameter; the webpage terminal receives user query of different time scale prediction warning demand, and repeats the above steps to update and display the output result in real time; The analysis process of the prediction and early warning model is as follows: The sliding window is defined as , the sliding step is , the average value of the water quality parameters in the optimal sliding window is calculated , the standard deviation and the median respectively, and the water quality parameters at the center position of the optimal sliding window are analyzed : If , the water quality and algal bloom prediction and early warning result is represented as normal water quality, and no early warning processing is performed. If , the water quality water bloom prediction and early warning result is represented as water quality anomaly, and an alarm information is issued; Also included: will be Corresponding water quality parameters Removed from the key multi-water quality parameter-time series data set and replaced with the median Substitute; The Savitzky-Golay filter was used to smooth the multi-water quality parameter time series dataset to reduce noise during the acquisition process. Fifth-order polynomial curve fitting was used, and nearest neighbor interpolation was used for filling to obtain the updated key multi-water quality parameter time series dataset.

2. The hyperspectral short-term prediction and early warning method for water quality and algal blooms in the eutrophic lake and reservoir according to claim 1, characterized in that, The process for collecting near-sensory hyperspectral reflectance of water bodies is as follows: The underwater downward irradiance is obtained by using a transmissive plate with transmittance : : ; wherein, is the irradiance of the transmissive plate; The following formula is used to obtain the collected water near-sensing hyperspectral reflectivity : ; in which, is the water-off irradiance of the transmission plate.

3. The hyperspectral short-term prediction and early warning method for water quality and algal blooms in the lake and reservoir based on the hyperspectral near-sensing according to claim 1, characterized in that, The pretreatment steps for the near-sensory hyperspectral reflectance of the pretreated water are as follows: According to the spectral curve of the water body near-sensing hyperspectral reflectivity, the spectral curve is regarded as a vector, and the length of each vector is calculated : ; wherein, , is based on real-time acquisition of water near-sensing hyperspectral reflectance based on spatio-temporal attribute information; The ratio of each wavelength to the mode length was calculated to obtain the normalized near-sensing hyperspectral reflectance using the following equation : ； The normalized near-sensed hyperspectral reflectance is linearly corrected using the following equation Linear correction: ; wherein, is the near-sensed hyperspectral reflectance after radiance correction, represents a radiance correction coefficient, is a radiance correction slope.

4. The hyperspectral short-term prediction and early warning method for water quality and algal blooms in eutrophic lakes and reservoirs according to claim 1, characterized in that, The water body types at the monitoring points include: clean reservoirs, natural rivers, urban rivers, and lakes; The synchronously measured water quality parameters include: total nitrogen, total phosphorus, chlorophyll a, transparency, water temperature, dissolved oxygen, permanganate index, turbidity, and total suspended solids concentration. The preset water quality inversion model set includes: near-sensor hyperspectral total phosphorus inversion model, near-sensor hyperspectral total nitrogen inversion model, near-sensor hyperspectral transparency inversion model, near-sensor hyperspectral total suspended solids concentration inversion model, near-sensor hyperspectral water temperature inversion model, and near-sensor hyperspectral dissolved oxygen inversion model.

5. The hyperspectral short-term prediction and early warning method for water quality and algal blooms in eutrophic lakes and reservoirs according to claim 1, characterized in that, The spatiotemporal matching principle includes: the time matching principle and the space matching principle; The time matching principle is represented as follows: defining the collection time point of water body near-sensing hyperspectral reflectivity as , the detection time point of water quality parameters as , the relationship between the collection time point and the detection time point should be as follows: ; The space matching principle comprises a longitude and latitude matching principle and a distance matching principle; the longitude and latitude matching principle is represented as follows: , wherein, is longitude and latitude of a collection point of water body near-sensing hyperspectral reflectivity, represents longitude and latitude of a detection point of a synchronous water quality parameter; The distance matching principle is expressed as follows: , wherein, is a standard observation range radius, represents a horizontal distance from a water sample collection point to a monitoring center of the near-sensing hyperspectral instrument; wherein, , is a height at which the near-sensing hyperspectral instrument is placed on the water surface, represents a field of view angle.

6. The hyperspectral short-term prediction and early warning method for water quality and algal blooms in the eutrophic lake and reservoir based on claim 5, characterized in that, The matching logic of the spatiotemporal matching principle is as follows: The logical judgment result of the space-time matching principle is assigned to When the time matching principle and the space matching principle are satisfied simultaneously, then The corresponding synchronous measured water quality parameters and water body near-sensing hyperspectral reflectivity are added to the data for model construction; Conversely, when at least one of the time matching principle and the space matching principle is not satisfied, 0 is assigned to , , the corresponding synchronous measured water quality parameters and water near-sensing hyperspectral reflectance are deleted. 。 7. The hyperspectral near-sensing lake and reservoir water quality water bloom short-term prediction and early warning system, used for realizing the water quality water bloom short-term prediction and early warning method according to any one of claims 1 to 6, characterized in that, include: The first module is set to collect near-sensory hyperspectral reflectance of water bodies in real time based on spatiotemporal attribute information and obtain pre-processed near-sensory hyperspectral reflectance of water bodies after preprocessing. The preprocessed water body near-sensory hyperspectral reflectance is combined with spatiotemporal attribute information to generate a water body near-sensory hyperspectral reflectance-time series dataset. The spatiotemporal attribute information includes: spatial information, temporal information, and the water body type at the monitoring point; The second module is configured to filter out corresponding water body proximate hyperspectral reflectance according to the water body type of the monitoring point, combine the synchronous measured water quality parameters, and construct a preset water quality inversion model set related to the water body type of the monitoring point according to a space-time matching principle; take the water body proximate hyperspectral reflectance-time series data set as an input variable, take the measured water quality parameters as an output dependent variable, and generate a multi-water quality parameter-time series data set by using the preset water quality inversion model set; The third module is configured to perform statistical analysis of multiple time scales based on the multi-water quality parameter-time series data set, filter out a key multi-water quality parameter-time series data set, and build a prediction and early warning model; take the preprocessed water body proximate hyperspectral reflectance as an input feature, output a water quality and algal bloom prediction and early warning result, update the key multi-water quality parameter-time series data set according to the water quality and algal bloom prediction and early warning result, and optimize the prediction and early warning model; The fourth module is configured to receive user inquiries about prediction and early warning requirements of different time scales on a webpage terminal, and repeatedly update and display the output result in real time.

Images