Nearshore seawater quality prediction and early warning method based on machine learning

Abstract

The invention relates to the technical field of marine environment monitoring, in particular to a coastal seawater quality prediction and early warning method based on machine learning, which comprises the following steps of: acquiring water quality data of an automatic monitoring station in real time, constructing a machine learning model input structure taking a time sliding matrix as a core, and predicting future water quality parameters by adopting a supervised learning algorithm; and a water quality risk quantitative evaluation system is innovatively provided. The defects of a traditional method in the aspects of business operation, real-time early warning and ecological risk pre-judgment are overcome, full-process automation from data collection, model training, prediction analysis to graded early warning is achieved, and intelligent decision support is provided for offshore ecological protection and disaster prevention and control.

Description

Technical Field

[0001] The present invention relates to the field of marine environmental monitoring technology, and in particular to a nearshore seawater quality prediction and early warning method based on machine learning. Background Art

[0002] The ocean is Earth's largest ecosystem, and changes in water quality have a significant impact on the ecological environment and the economic and tourism development of coastal areas. Scientific prediction of seawater quality is not only crucial to the stability of marine ecosystems but also a crucial guarantee for sustainable development.

[0003] Supervised learning is an artificial intelligence method that effectively leverages big data to predict future information. While a growing number of prediction methods have been widely applied to seawater quality forecasting, they have yet to be effectively adopted for regular operational use. Furthermore, while future water quality parameters are known, there are significant gaps in the corresponding water quality risk warning and management. Summary of the Invention

[0004] The purpose of this invention is to provide a nearshore seawater quality prediction and early warning method based on machine learning, aiming to solve the problem that existing methods are unable to accurately predict and warn of ecological risks.

[0005] To achieve the above objectives, the present invention provides a method for predicting and warning nearshore seawater quality based on machine learning, comprising the following steps:

[0006] Obtain historical seawater quality data from the seawater quality online monitoring database;

[0007] Train a machine learning model based on historical seawater quality data to obtain a prediction model;

[0008] Predicting water quality parameters using the prediction model to obtain prediction result 1;

[0009] Calculate the water quality grade according to the prediction result 1 to obtain the water quality grade;

[0010] Perform water quality risk assessment and early warning based on the water quality level, obtain prediction result 1, and transmit the prediction result 1 to the database;

[0011] Based on the prediction results, calculate whether algae proliferation is likely to occur.

[0012] Among them, in "training a machine learning model based on historical seawater quality data to obtain a prediction model", the following steps are included:

[0013] Construct a time sliding data matrix based on historical seawater quality data;

[0014] Copy the time-sliding data matrix as the input matrix for the machine learning model;

[0015] Filter the parameter columns that need to be predicted and delete other columns to form the output matrix of the machine learning model;

[0016] The machine learning model is trained using the input matrix and the output matrix to obtain a prediction model.

[0017] The step of “predicting the water quality parameters by using the prediction model to obtain a prediction result 1” includes the following steps:

[0018] Input water quality data into the prediction model to generate a sliding prediction result-a data vector;

[0019] The prediction result vector is formatted into a multi-index single data sequence format and uploaded to the prediction result receiving database.

[0020] The step of "calculating the water quality grade according to the prediction result 1 to obtain the water quality grade" includes the following steps:

[0021] Read prediction result 1 from prediction result 1 database;

[0022] According to the seawater quality standards, the predicted water quality parameters are compared with the grade classification limits in the standards, and the future water quality grade is calculated.

[0023] Among them, in "performing water quality risk assessment and early warning based on water quality level, obtaining prediction result 1, and transmitting prediction result 1 to the database", the following steps are included:

[0024] Calculate water quality risk index based on water quality level;

[0025] The water quality risk index is divided into early warning levels according to the AWI value;

[0026] Obtain specific information leading to high water quality risks and upload it to the early warning results database.

[0027] The step of "calculating whether algae proliferation is likely to occur based on the prediction result" includes the following steps:

[0028] Based on prediction result one, pH, dissolved oxygen concentration, chlorophyll a concentration and water temperature are predicted, and prediction result two is obtained;

[0029] According to the second prediction result, it is judged whether algae proliferation may occur at present and the algae proliferation situation is obtained;

[0030] According to the algae proliferation situation and warning level classification, the warning level and the specific information causing the high water quality risk are obtained to issue a warning, and the results are uploaded to the warning result database.

[0031] The present invention provides a nearshore seawater quality prediction and early warning method based on machine learning, comprising the following steps: obtaining historical seawater quality data from an online seawater quality monitoring database; training a machine learning model based on the historical seawater quality data to obtain a prediction model; predicting water quality parameters using the prediction model to obtain a prediction result (I); calculating a water quality grade based on the prediction result (I) to obtain a water quality grade; conducting a water quality risk assessment and early warning based on the water quality grade to obtain a prediction result (I), which is transmitted to a database; and calculating the likelihood of algae proliferation based on the prediction result (I). This method acquires water quality data from automatic monitoring stations in real time, constructs a machine learning model input structure centered on a time-sliding matrix, uses a supervised learning algorithm to predict future water quality parameters, and innovatively proposes a water quality risk quantitative assessment system. This method addresses the shortcomings of traditional methods in operational operation, real-time early warning, and ecological risk prediction, achieving full automation of the entire process from data acquisition, model training, prediction analysis, to graded early warning, providing intelligent decision-making support for nearshore ecological protection and disaster prevention. This method addresses the inability of existing methods to accurately predict and early warning ecological risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0033] Figure 1 This is a workflow diagram of a nearshore seawater quality prediction and early warning method based on machine learning provided by the present invention.

[0034] Figure 2 is the sliding of the matrix

[0035] Figure 3 Primary water quality parameter event code example: 32-bit integer structure

[0036] Figure 4 Water quality level event code example: 16-bit integer structure

[0037] Figure 5 This is a flow chart of a nearshore seawater quality prediction and early warning method based on machine learning provided by the present invention.

[0038] Figure 6 This is a flowchart of training a machine learning model based on historical seawater quality data to obtain a prediction model.

[0039] Figure 7This is a flow chart for predicting water quality parameters using the prediction model to obtain prediction result one.

[0040] Figure 8 It is a flow chart for calculating the water quality grade based on the prediction result 1 to obtain the water quality grade.

[0041] Figure 9 It is a flow chart for conducting water quality risk assessment and early warning based on water quality levels, obtaining prediction result one, and transmitting prediction result one to a database.

[0042] Figure 10 It is a flow chart for calculating whether algae proliferation is likely to occur based on the prediction results. DETAILED DESCRIPTION

[0043] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0044] See also Figures 1 to 10 The present invention provides a method for predicting and warning the quality of coastal seawater based on machine learning, comprising the following steps:

[0045] S1 obtains historical seawater quality data from the seawater quality online monitoring database;

[0046] Specifically, historical seawater quality data is obtained for each seawater quality monitoring site from the seawater quality online monitoring database. Corresponding module M1: The seawater quality online monitoring database data is captured to the corresponding memory of the running program. In this example, it is implemented by querying the database through Python's pyodbc, sqlalchemy and pandas. Next, the multi-index single sequence needs to be converted into an online monitoring data matrix, which is achieved by disassembling the monitoring factor index of the sequence as a column name. For missing data, due to the characteristics of online monitoring data, time linear interpolation is used. The data acquisition step may end in other cases when the online monitoring data matrix is directly stored in the database, or the online monitoring data is provided by the corresponding server interface rather than by accessing the database, or the operating platform environment is different, and there are differences in interpolation requirements, and there may be certain differences in the specific details of the method in the present invention.

[0047] S2 trains a machine learning model based on historical seawater quality data to obtain a prediction model;

[0048] S21 constructs a time sliding data matrix based on historical seawater quality data;

[0049] Specifically, through the shift function of pandas, the time index of the unslid matrix is reduced by a series of time differences to generate a series of time sliding matrices, and these matrices are aligned and merged, and rows with empty values are deleted. At this time, a time sliding matrix is obtained, and its last row is the input for the machine learning model to predict future water quality. If the column names of each matrix are not processed, there will be duplicate column names in the merged matrix, which are difficult to distinguish. Therefore, in this example, the sliding days and the station to which they belong are added before the column name (this is useful when merging matrices of multiple automatic monitoring stations), and connected with a given string delimiter (such as "_"), so that the sliding and station to which the same factor belongs can be distinguished.

[0050] S22 copies the time sliding data matrix as the input matrix of the machine learning model;

[0051] Specifically, the data matrix is copied as the output of the machine learning model. The columns containing the ocean water quality monitoring parameters that need to be predicted are selected and retained, while the remaining columns are deleted to reduce computational overhead. Some rows with early time indices in the matrix are deleted because their corresponding sliding data do not have sufficient earlier sliding data to serve as the training set input for the machine learning model.

[0052] S23 filters the parameter columns that need to be predicted and deletes other columns to form the output matrix of the machine learning model;

[0053] Specifically, the behavior of deleting rows has been implemented in the second step of deleting null values, because the rows with null values come from the fact that these positions do not have earlier or later data to fill in accordingly. The sliding of the data matrix and the source of null values are as follows: Figure 2 The subsequent behavior of removing columns for the model output depends on the factors that do not need to be predicted.

[0054] S24 uses the input matrix and the output matrix to train the machine learning model to obtain a prediction model.

[0055] Specifically, we use the third data matrix as input and the second data matrix as output to train a specific machine learning prediction model. In this example, we use Scikit-learn's machine learning regressor for prediction. Alternatively, we can use deep learning and other methods for regression prediction.

[0056] S3 predicts the water quality parameters using the prediction model to obtain a prediction result 1;

[0057] S31 inputs the water quality data into the prediction model to generate a sliding prediction result-a data vector;

[0058] Specifically, the first matrix represents the rows of the latest data, which are input into the machine learning model to generate a sliding prediction result data vector. Module M2: The captured historical data is formatted into a time-sliding data matrix to form the training set and input matrix for the machine learning model.

[0059] S32 formats the prediction result vector into a multi-index single data sequence format and uploads it to the prediction result receiving database.

[0060] Specifically, the structure of the vector is identical to each row of the third data matrix. By using the column names that distinguish each column, the column names can be separated and converted into a multi-indexed single data sequence by removing the delimiters. This also satisfies the requirement in this example for formatting the sliding prediction result data vector and uploading it to the prediction result receiving database. The specific method for uploading to the database is the reverse process provided by the same tool, so the methods are actually similar. The above content is implemented by the following modules: Module M3: Formats the machine learning prediction result matrix into the corresponding format of the prediction result receiving database and uploads it.

[0061] S4 calculates the water quality grade according to the prediction result 1 to obtain the water quality grade;

[0062] S41 reads prediction result 1 from the prediction result 1 database;

[0063] Specifically, the prediction results are directly used or read from the prediction result database to calculate the water quality level within a certain period of time in the future.

[0064] S42 compares the predicted water quality parameters with the grade classification limits in the seawater quality standards and calculates the future water quality grade.

[0065] Specifically, the future seawater quality is predicted, and the prediction output is the hourly value or daily average value and other results. Generally speaking, the water quality grade is calculated by the prediction results, following the standard document such as the National Standard of the People's Republic of China "Seawater Quality Standard" (GB 3097-1997). Each monitoring item in the prediction result is compared with the grade division limit given in such standards to obtain the water quality grade value. Among them, the first to fourth categories of water quality are expressed as water quality grade values 1 to 4 according to the requirements of this method, and the fourth category of water quality can be expressed as 5. This calculation method can be used especially for sea areas where the water quality grade is the third category or better for 70% of the time period in the past year.

[0066] If a finer division of water quality is required, the water quality grade result can be expressed as a numerical value with a decimal part to be more precise than the integer water quality grade, and the division is based on the water quality grade standard that needs to be followed.

[0067] If the water quality of an area remains poor for an extended period, with the water quality grade being Class IV or worse for 70% of the time in the past year, a value greater than 5 is needed to indicate the severity of the water quality. The water quality grade of each water quality factor can be calculated using the following formulas 1-1 to 4. The final water quality grade for the area is the maximum value of each water quality factor. Depending on the seawater quality monitoring project, these formulas can be expanded to meet the required water quality standards to ensure that the calculated results are consistent with the original standard grades.

[0068] L DO =7-[DO] [DO]≤3 Formula 1-1

[0069]

[0070] L IN =10[IN]-1[IN]>0.5 Formula 1-3

[0071]

[0072] Where LE represents the water quality level of the predicted item E, which can be dissolved oxygen (DO), pH, inorganic nitrogen (N, IN) or active phosphate (P);

[0073] [E] represents the concentration of predicted item E, mg / L;

[0074] pH is the predicted pH.

[0075] In this example, the surveyed sites were classified according to the Seawater Quality Standard (GB 3097-1997) for seawater quality, resulting in water quality grades ranging from 1 to 5. Because the prediction results are water quality parameter values rather than direct water quality grade predictions, even when the actual value is at the grade boundary, a limited error may still cause the gradient-changing water quality grade prediction result to differ from the actual result, reducing the accuracy of the water quality grade prediction. To address the issue of water quality grade gradients, non-integer water quality grades can be introduced, converting the water quality grade judgment into a continuous function that is identical to the original grade classification result at the limit value, so that the water quality grade changes continuously with the water quality parameter rather than a gradient.

[0076] S5 performs water quality risk assessment and early warning based on the water quality level, obtains prediction result 1, and transmits the prediction result 1 to the database;

[0077] S51 calculates the water quality risk index based on the water quality level;

[0078] Specifically, the water quality risk index (Aquaworsening index, AWI) is calculated according to the following formula, and water quality warning and forecast levels are determined (the AWImax limits of each level of warning can be adjusted as needed).

[0079]

[0080] AWI max =max(AWI1 AWI2 … AWI n ) Formula 2-2

[0081] Wherein, AWI represents water quality risk index;

[0082] L0 is the benchmark water quality level, dimensionless, indicating that when the water quality level is equal to or better than the benchmark water quality level, the water quality is good and no warning is needed;

[0083] Lt is the water quality classification at the prediction time t, dimensionless;

[0084] n is the total number of predictions, dimensionless. For example, if the daily average water quality value for 7 days is predicted, then n is 7, which is actually equivalent to the number of prediction results after the time point of the water quality risk index;

[0085] x=1,2,……,n;

[0086] AWIx is the water quality risk index of the xth predicted value, dimensionless;

[0087] f is the dimensionless number of predictions in a day, e.g., 1 for daily averages and 24 for hourly averages. This parameter unifies the different forecast frequencies, making one forecast per day for daily averages and 24 forecasts per day for hourly averages uniform. In practice, 1 / f represents the dimensionless time period represented by the forecast. If the duration of each forecast in a series of forecasts varies, 1 / f can be converted to the dimensionless time period in the original method using days as the unit.

[0088] S52 divides the water quality risk index into early warning levels according to the AWI value;

[0089] Specifically, AWImax is the final water quality risk index result. The recommended AWImax warning limits are: when AWImax ≥ 7, a Level I warning, the lowest level; then, when AWImax ≥ 11, a Level II warning, the intermediate level; and when AWImax ≥ 15, a Level III warning, the highest level. The warning limits at each level can be adjusted based on needs such as controlling the frequency of warnings and are not required to be integers. The adjustment method involves selecting a long period of water quality observation data, such as the last three years of water quality data in this example. The AWI value for each day of these three years is then calculated. The 94th, 97th, and 99th percentile values of these AWI values are then selected and set as the limits for the three warning levels. At this point, warnings at the three levels are generated approximately 6%, 3%, and 1% of the time, respectively.

[0090] By assigning a value to L0, the 70th percentile of the long-term observational water quality grade for the site under investigation can be set as the baseline water quality level. In this example, the 70th percentile of the water quality grade for the site under investigation is 2, so L0 is set to 2. When the water quality category at a given point in time is greater than this value, it indicates that the water quality is deteriorating and requires attention.

[0091] S53 obtains the specific information that causes the high water quality risk and uploads it to the early warning result database.

[0092] Specifically, when any warning is generated, the forecast results must be reviewed to obtain detailed warning information. First, the corresponding AWImax value, x, must be obtained. Then, from the first to the xth forecast, the water quality level must be listed. The monitoring factors that caused the water quality level to reach the current value within these forecasts are then listed. By summarizing these forecast time points and the corresponding monitoring factors, the detailed water quality risk warning information is generated.

[0093] Assume that for the sea area in this example, seven daily average forecasts for the next seven days yield water quality categories of 5, 5, 1, 1, 1, 1, and 5, respectively. According to the AWI calculation formula, seven AWI values need to be calculated. The first seven values are 3, 6, 5, 4, 3, 2, and 5, respectively. In this case, the final calculated result, AWImax, is the maximum value, AWI2 = 6. To obtain detailed warning information, we need to analyze the various factors in the forecast results for the first two days to determine which factors contributed to the final water quality category value of 5.

[0094] If the warning information receiving end cannot accept string type data, the specific information of the water quality risk warning can be converted into integer event code data to upload to the database. In this example, the water quality parameters included in the water quality risk warning are pH, dissolved oxygen concentration (DO), inorganic nitrogen concentration (TN) and active phosphate (PO4) concentration. The situation where other parameters lead to low water quality level is not considered for the time being because it has not been observed. In order to indicate which factors in a day have high water quality category values that lead to the same water quality category on that day, a 4-bit binary system is used, such as Figure 3 As shown, each bit indicates the four factors in turn, with 1 indicating that the factor is the cause of the water quality category on that day, and 0 indicating that it is not. If the water quality category on a certain day is not higher than the baseline water quality category, all four bits are set to 0, and a total of 7 days of information are calculated in the AWI calculation, so the event code occupies 7×4=28 bits; at this time, the first 4 bits of the 32-bit integer are still free, and they can all be set to 0 to occupy the place, or to store other situations where the number of information is less than 16. In addition, if you also need to store the water quality level for each of the 7 days, combined with the baseline water quality level of 2 in this example, you can consider the following Figure 4Using two binary digits to store the water quality level for each day, seven days occupy a total of 7 × 2 = 14 bits. The remaining two bits serve as placeholders for storing additional information for four scenarios, such as the number of days between the AWI warning date and the warning calculation initiation date. Consider storing this information along with the primary water quality parameter event code in a 64-bit integer. Uploading the warning results and details is similar to uploading the predicted water quality parameter values. This describes module M4: calculating the predicted water quality level, calculating the water quality risk warning level, and uploading it to the corresponding database.

[0095] S6 calculates whether algae proliferation is likely to occur based on the prediction result 1.

[0096] S61 predicts pH, dissolved oxygen concentration, chlorophyll a concentration and water temperature based on prediction result one, and predicts result two;

[0097] Specifically, the pH, dissolved oxygen concentration (DO, mg / L), chlorophyll a concentration (Chl-a, μg / L) and water temperature (Temp, ℃) are predicted.

[0098] S62 determines whether algae proliferation is likely to occur currently based on the second prediction result, and obtains the algae proliferation situation;

[0099] Specifically, if the prediction result satisfies any of the following conditions in the empirical formula, it is determined that algae proliferation may occur in the time period:

[0100] pH ≥ 8.25 and DO ≥ 8.5 mg / L and Chl-a ≥ 15 μg / L Formula 3-1

[0101] -0.166632Temp+4.194055DO-3.8966pH≥10 Formula 3-2

[0102] S63 obtains the warning level and the specific information causing the high water quality risk according to the algae proliferation situation and the warning level classification, issues a warning, and uploads the results to the warning result database.

[0103] Specifically, the algae proliferation duration, T, is calculated according to Formula 4 (the T limit for each level of warning can be adjusted as needed and is not required to be an integer). Clearly, this result indicates the predicted duration of algae proliferation in the surveyed sea area at that point in time within the forecast range.

[0104]

[0105] Where,

[0106] T algae is the duration of algae proliferation, days;

[0107] n is the total number of predictions, dimensionless. For example, if the daily average water quality value for 7 days is predicted, then n is 7;

[0108] boolt is 1 day when algae are proliferating, otherwise it is 0;

[0109] f represents the number of predictions in a day and is dimensionless. For example, it is 1 when predicting the daily average value and 24 when predicting the hourly value. This is the same as the calculation of AWI.

[0110] In the recommended warning limits for algae proliferation duration, from light to heavy, Level I, Level II, and Level III warnings are issued when T algae is greater than or equal to 1, 2, and 3, respectively. Since the judgment of algae proliferation is achieved by the combined action of multiple factors, the description of the warning details only needs to indicate the time when algae proliferation occurs. Similar to water quality warnings, if the warning information receiver cannot accept string type data, the specific information of the algae proliferation risk warning can be converted into integer event code data for uploading to the database. In this example, only one byte of 8-bit integer is needed to indicate whether there will be algae proliferation in the next 7 days (1) or not (0), and the sum of each bit of this byte is also the algae proliferation duration result. The above content is module M5: calculate the algae proliferation duration, calculate the algae proliferation risk warning level and upload it to the corresponding database.

[0111] The above disclosure is merely a preferred embodiment of the method for predicting and warning nearshore seawater quality based on machine learning of the present invention. It is certainly not intended to limit the scope of the present invention. A person skilled in the art will understand that all or part of the processes of the above embodiment and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for predicting and warning nearshore seawater quality based on machine learning, characterized in that: The following steps are involved: Obtain historical seawater quality data from the seawater quality online monitoring database; Train a machine learning model based on historical seawater quality data to obtain a prediction model; Predicting water quality parameters using the prediction model to obtain prediction result 1; Calculate the water quality grade according to the prediction result 1 to obtain the water quality grade; Perform water quality risk assessment and early warning based on the water quality level, obtain prediction result 1, and transmit the prediction result 1 to the database; Based on the prediction results, calculate whether algae proliferation is likely to occur.

2. The method for predicting and warning nearshore seawater quality based on machine learning according to claim 1, characterized in that: In "Training a machine learning model based on historical seawater quality data to obtain a prediction model," the following steps are included: Construct a time sliding data matrix based on historical seawater quality data; Copy the time-sliding data matrix as the input matrix for the machine learning model; Filter the parameter columns that need to be predicted and delete other columns to form the output matrix of the machine learning model; The machine learning model is trained using the input matrix and the output matrix to obtain a prediction model.

3. The method for predicting and warning nearshore seawater quality based on machine learning according to claim 1, characterized in that: In "predicting the water quality parameters using the prediction model to obtain prediction result 1", the following steps are included: Input water quality data into the prediction model to generate a sliding prediction result-a data vector; The prediction result vector is formatted into a multi-index single data sequence format and uploaded to the prediction result receiving database.

4. The method for predicting and warning nearshore seawater quality based on machine learning according to claim 1, wherein: In "calculating the water quality grade according to the prediction result 1 to obtain the water quality grade", the following steps are included: Read prediction result 1 from prediction result 1 database; According to the seawater quality standards, the predicted water quality parameters are compared with the grade classification limits in the standards, and the future water quality grade is calculated.

5. The method for predicting and warning nearshore seawater quality based on machine learning according to claim 1, wherein: The process of "performing water quality risk assessment and early warning based on water quality levels, obtaining prediction result 1, and transmitting prediction result 1 to a database" includes the following steps: Calculate water quality risk index based on water quality level; The water quality risk index is divided into early warning levels according to the AWI value; Obtain specific information leading to high water quality risks and upload it to the early warning results database.

6. The method for predicting and warning nearshore seawater quality based on machine learning according to claim 1, wherein: In "calculating whether algae proliferation is likely to occur based on the prediction result", the following steps are included: Based on prediction result one, pH, dissolved oxygen concentration, chlorophyll a concentration and water temperature are predicted, and prediction result two is obtained; According to the second prediction result, it is judged whether algae proliferation may occur at present and the algae proliferation situation is obtained; According to the algae proliferation situation and warning level classification, the warning level and the specific information causing the high water quality risk are obtained to issue a warning, and the results are uploaded to the warning result database.

Images