A method for water bloom prevention and control based on machine learning

Abstract

The present invention discloses a method for scheduling water bloom prevention and control based on machine learning, comprising the following steps: (1) constructing a water bloom prevention and control scheduling model including economic and ecological objective functions; (2) establishing an ecological objective function using an LSTM model of a water body's comprehensive nutrient index and phytoplankton density; and (3) employing a cuckoo optimization algorithm to solve the multi-objective problem, obtaining a non-inferior solution, and selecting a satisfactory ecological scheduling scheme from the solution. The present invention can reduce the phytoplankton density downstream of a reservoir without compromising the economic benefits of the reservoir, thereby reducing the frequency and severity of water blooms and improving the ecological environment in the downstream river channel. The raw data required by the present invention is simple and readily available, and the resulting optimized scheme has a good water bloom prevention and control effect.

Description

Technical Field

[0001] The present invention belongs to the field of reservoir ecological regulation and relates to an algal bloom prevention and control regulation method based on machine learning. Background Art

[0002] Against the backdrop of the growing importance of ecological progress, the concepts of river ecological progress and environmental governance have profoundly impacted water conservancy, and the theory of reservoir ecological regulation is becoming increasingly sophisticated. Currently, reservoir regulation, such as tidal regulation, is commonly used to prevent and control algal blooms. However, the complex mechanisms of algal bloom formation make accurate simulation difficult, resulting in inefficient control. To address this issue, the present invention proposes a method for algal bloom prevention and control that uses machine learning to improve the accuracy of algal bloom simulation and couples it with reservoir regulation. Summary of the Invention

[0003] To address the current problems of complex algal bloom formation mechanisms, which make accurate simulation difficult and the low efficiency of algal bloom prevention and control scheduling, the present invention proposes an algal bloom prevention and control scheduling method that uses machine learning to improve the accuracy of algal bloom simulation and couples it with reservoir scheduling. The ecological scheduling scheme described in this method can reduce the harm of algal blooms in rivers and has guiding value for algal bloom prevention and control.

[0004] The technical solutions provided by the present invention are as follows:

[0005] A method for controlling water blooms based on machine learning, comprising the following steps:

[0006] Step 1: Collect and organize the required flow data and ecological data;

[0007] Step 2: Establish an algal bloom simulation model using the LSTM long-short-term memory model. Time and flow rates, which have a significant impact on algal blooms, were selected as the model inputs, and phytoplankton density and the water body's comprehensive nutrient index were used as the model outputs. Model hyperparameters were adjusted, and the algal bloom simulation model was established. Random time and flow rate data were used to test the model's accuracy and rationality.

[0008] Step 3: Determine the objective function of the multi-objective optimization scheduling. Select the water body nutrient status destruction rate and the multi-year average density of phytoplankton as the ecological objectives of the scheduling, and select the multi-year average water supply and multi-year average power generation as the economic objectives of the scheduling. Then, use the multi-objective optimization method based on the cuckoo algorithm to solve the algal bloom simulation model to obtain a non-inferior solution set for the problem.

[0009] Step 4: Select a satisfactory solution for ecological scheduling from the non-inferior solution set according to the optimal solution, draw a reservoir scheduling diagram, and obtain the algal bloom prevention and control scheduling plan.

[0010] Furthermore, in step 1, the flow data includes a flow sequence corresponding to time, and the data interval should be 10 days or less. The ecological data includes total nitrogen, total phosphorus, chlorophyll a concentration, water transparency, chemical oxygen demand, and phytoplankton density in the river channel, and the data interval should be 30 days or less. The flow data and ecological data correspond in both time and space.

[0011] Furthermore, in step 2, the time scale of the model input is ten days, and the flow unit is m 3 / s; the comprehensive trophic status index is a dimensionless quantity, and the unit of phytoplankton density is cells / L. When using machine learning models for algal bloom simulation, a certain amount of data must be maintained so that the machine learning model can detect the inherent patterns and trends in the data.

[0012] Furthermore, the model hyperparameters include HiddenUnits, MaxEpochs, GradientThreshold, InitialLearnRate, LearnRateDropPeriod, and LearnRateDropFactor; the hyperparameters used in training the model are adjusted based on the characteristics of the input data. Preferably, the model input data is split into 80% training set and 20% test set.

[0013] Furthermore, in step 2, when calculating the comprehensive nutritional status index, multiple sub-indicators need to be calculated, including five indicators: chlorophyll a concentration in the water body, total nitrogen, total phosphorus, permanganate index and transparency.

[0014] Furthermore, the comprehensive nutritional status index is calculated by assigning different weights to each sub-indicator;

[0015] The calculation formula is as follows:

[0016]

[0017]

[0018]

[0019] Where: TLI(Σ) is the comprehensive nutritional index, Wj is the relevant weight of the nutritional status index of the jth parameter, TLI(j) is the nutritional status index of the jth parameter, r ij is the ratio coefficient of the jth parameter to the benchmark parameter Chl-a, m is the number of evaluation indicators; TLI(*) is the nutritional status index of each sub-indicator, TLI(Chl a)The water body nutrient index for chlorophyll a concentration is TLI(Chl-a), the water body nutrient index for total phosphorus concentration is TLI(TP), the water body nutrient index for total nitrogen concentration is TLI(TN), the water body nutrient index for water transparency is TLI(SD), and the water body nutrient index for permanganate index is TLI(COD). All of the above indicators are sub-indices of the comprehensive water body nutrient index;

[0020] Classify the nutrient status of the water body according to the value of the comprehensive nutrient index:

[0021] 0 < TLI(Σ) ≤ 30, oligotrophic; 30 < TLI(Σ) ≤ 50, mesotrophic; 50 < TLI(Σ) ≤ 60, slightly eutrophic; 60 < TLI(Σ) ≤ 70, moderately eutrophic; 70 < TLI(Σ) ≤ 100, severely eutrophic; Evaluate the nutrient status of the water body to determine the nutrient status of the middle and lower reaches of the Han River, and compare the improvement effect after ecological regulation compared with before ecological regulation.

[0022] Furthermore, in step 3, determine the objective function of multi-objective optimization. The calculation formula of the objective function is as follows:

[0023]

[0024]

[0025] Min TLI σ =S σ / S (3)

[0026]

[0027] In the formula: WS is the average annual water supply, S is the total length of the regulation period, t is the regulation time period, Wst is the water supply in each time period; HG is the average annual power generation, Ht is the power generation in each time period; TLI σ is the water body nutrient status damage rate, S σ is the length of the time period when damage occurs during the regulation period; MPD is the average annual phytoplankton target, PD t is the phytoplankton density in each time period.

[0028] Furthermore, in step 3, the boundary conditions in the multi-objective optimization process are as follows:

[0029] (1) Water volume balance constraint

[0030] V i+1 =V i +(Q Ii -Q Oi )Δt (5)

[0031] In the formula: Vi represents the reservoir capacity in the i-th time period, Q IiIndicates the inflow flow in each period, Q Oi represents the outbound flow in each period, and Δt is the time interval;

[0032] (2) Reservoir capacity constraints

[0033]

[0034] Where: V i 、 Respectively represent the upper and lower bounds of storage capacity in period i;

[0035] (3) Outbound flow constraints

[0036]

[0037] Where: Q Oi 、 They represent the upper and lower bounds of the outbound flow in period i respectively;

[0038] (4) Initial state of the reservoir

[0039] V1=V s (8)

[0040] Where: V1 represents the initial storage capacity during the scheduling period, V s represents the storage capacity at the end of the scheduling period;

[0041] (5) Minimum flow limit of river

[0042] Q e ≤Q Oi (9)

[0043] Where: Q e Indicates the minimum ecological flow in the downstream river channel.

[0044] Furthermore, in step 2, the Nash efficiency coefficient NSE and mean absolute percentage error MAPE are used to evaluate the LSTM simulation effect, and the calculation formula is as follows:

[0045]

[0046]

[0047] Where: y k is the original sequence, The model outputs a sequence, where n is the total number of data sets and k represents the kth data in the data set.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The method provided by the present invention is based on an algal bloom simulation model of a machine learning model. Machine learning is used to improve the accuracy of algal bloom simulation, and the method of algal bloom prevention and control scheduling of reservoir scheduling is coupled to obtain an ecological scheduling plan that can reduce the harm of algal blooms in rivers.

[0050] 2. The present invention provides a water bloom simulation model based on a machine learning model and verifies the accuracy and rationality of the model.

[0051] 3. The use of multi-objective optimization has effectively balanced the economic and ecological benefits of the reservoir. The proposed scheduling scheme has improved the ecological benefits of the reservoir without damaging the economic benefits of the reservoir. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Flowchart of the method of the present invention.

[0053] Figure 2 This is the annual distribution of the comprehensive trophic index and phytoplankton density simulation model.

[0054] Figure 3 The algal bloom prevention and control scheduling diagram corresponding to the satisfactory solution. DETAILED DESCRIPTION

[0055] In order to make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0056] Figure 1 The process of the method of the present invention is shown, and the steps are as follows:

[0057] Step 1: Collect and organize the required flow data and ecological data according to research needs.

[0058] Step 2: Establish an algal bloom simulation model. Using the LSTM (Long Short-Term Memory) model, we selected time and flow rates, which are most influential in algal blooms, as inputs, and phytoplankton density and the water body's comprehensive nutrient index as outputs. We adjusted the model parameters to establish a reasonable algal bloom simulation model and used randomized time and flow rate data to verify the model's accuracy and rationality.

[0059] Step 3: A multi-objective optimization method based on the cuckoo algorithm is used to obtain a set of non-inferior solutions. The water nutrient destruction rate and the multi-year average phytoplankton density are selected as the ecological objectives for scheduling, and the multi-year average water supply and multi-year average power generation are selected as the economic objectives for scheduling.

[0060] Step 4: Select a satisfactory solution for ecological scheduling from the non-inferior solution set according to the optimal solution, draw a reservoir scheduling diagram, and obtain the algal bloom prevention and control scheduling plan.

[0061] Example

[0062] Taking the Danjiangkou Reservoir and the middle and lower reaches of the Han River as an example, the Danjiangkou Reservoir inflow data from 1979 to 2020 were selected for routine scheduling and multi-objective optimization.

[0063] Data from multiple stations along the middle and lower reaches of the Han River were compiled as input for the training model. Streamflow data included flow at each station, and ecological data included chlorophyll-a concentration, phytoplankton density, total nitrogen, total phosphorus, and the permanganate index at the corresponding station. Where individual data were missing or incomplete, they were supplemented using linear interpolation to extend the data to the ten-day timescale.

[0064] The LSTM long short-term memory network was used to construct an algal bloom simulation model. The flow rate and decade number of each measuring station in the study area were used as the input of the model, and the corresponding phytoplankton density and comprehensive water nutrition index were used as the output of the model.

[0065] The hyperparameters used to build the model are shown in Table 1.

[0066] Table 1 Hyperparameters of LSTM model

[0067]

[0068] The simulation results of the algal bloom simulation model are shown in Table 2. From Table 2, it can be seen that the Nash efficiency coefficient and mean absolute percentage error of the LSTM model are within the acceptable range. The simulated sequence and the measured sequence are correlated at the 0.01 level. It can be considered that the simulation results of each LSTM model are good. Figure 2 As shown by Figure 2 (a) It can be seen that phytoplankton density is significantly affected by season and flow. In terms of season, phytoplankton density is highest in spring and gradually decreases with the passage of the season. In terms of flow, phytoplankton density is inversely proportional to river flow, with a decrease in flow leading to a significant increase in phytoplankton density. Figure 2 (b) reflects the intra-annual variation of the comprehensive trophic index, with high values mainly concentrated in spring and low flow areas, which is similar to the intra-annual distribution of phytoplankton density.

[0069] Table 2 LSTM model simulation results

[0070]

[0071] During multi-objective optimization, key points of the reservoir operation lines were optimized, with four inflection points set for each operation line. The optimization population size was set to 100, the number of iterations was set to 200, and the total number of optimization variables was set to 27. The lower bound of the optimization variables was the dead water level, and the upper bound was the flood control level. The optimized reservoir operation lines should not intersect. The multi-objective optimization process yielded a set of non-inferior solutions, and a satisfactory solution for this operation was selected based on a solution that maximized the reservoir's ecological benefits without compromising its economic benefits.

[0072] The satisfactory solution of multi-objective optimization and the improvement rate compared with conventional scheduling are shown in Table 3. The algal bloom prevention and control scheduling diagram obtained according to the optimal solution is shown in Figure 3 shown.

[0073] Table 3 Pareto satisfactory solution results

[0074]

[0075] To address the algal bloom problem, which is common in the middle and lower reaches of the Han River, this study coupled the LSTM (Long Short-Term Memory) model with multi-objective optimization to identify a satisfactory scheduling solution. The study used the middle and lower reaches of the Han River as the research object and concluded that the LSTM model, used to establish a comprehensive nutrient index and phytoplankton density model, demonstrated good simulation results. The measured and simulated series showed significant correlation at the 0.01 level, indicating its potential application in guiding ecological scheduling. The satisfactory multi-objective optimization solution increased the multi-year average water supply by 0.45%, increased the multi-year average power generation by 1.06%, and reduced the multi-year average phytoplankton density by 1.57% compared to conventional scheduling. This achieved excellent results while maintaining economic benefits while enhancing the ecological benefits of scheduling, validating the feasibility of the proposed scheduling solution. The proposed ecological scheduling solution can, to a certain extent, mitigate the harm posed by algal blooms in the middle and lower reaches of the Han River.

[0076] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent replacements and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention should be included in the scope of protection of the invention.

Claims

1. A method for controlling water blooms based on machine learning, characterized in that: The steps include: Step 1: Collect and organize the required flow data and ecological data; Step 2: Build an algal bloom simulation model using the LSTM (Long Short-Term Memory) model. Select time and flow rates, which are most influential in algal blooms, as the model inputs, and phytoplankton density and a comprehensive water nutrient index as the outputs. Adjust model hyperparameters, establish an algal bloom simulation model, and use randomized time and flow data to verify the model's accuracy and rationality. The calculation of the comprehensive nutrient index of water bodies requires the calculation of multiple sub-indicators, including chlorophyll a concentration, total nitrogen, total phosphorus, permanganate index, and transparency in the water body. The comprehensive nutrient index of water bodies is calculated by assigning different weights to each sub-indicator. The calculation formula is as follows: Where: TLI ( Σ ) is the comprehensive nutrient index of water bodies, Wj For the j The relevant weights of the nutritional status index of each parameter, TLI (j) For the j Nutritional status index of various parameters, r ij For the j Parameters and benchmark parameters Chl-a The ratio coefficient, m is the number of evaluation indicators; TLI(*) is the nutritional status index of each sub-indicator, TLI (Chla) Chlorophyll a Concentration of water nutrient index, TLI (TP) is the water nutrient index of total phosphorus concentration, TLI (TN ) is the water nutrient index of total nitrogen concentration, TLI (SD) is the water nutrient index of water transparency, TLI (COD) The permanganate index is the water nutrient index. The above indicators are all sub-indices of the comprehensive water nutrient index. The nutritional status of water bodies is graded according to the value of the comprehensive nutritional index of water bodies: 0 < TLI(Σ) ≤ 30, oligotrophic; 30 < TLI(Σ) ≤ 50, mesotrophic; 50 < TLI(Σ) ≤ 60, slightly eutrophic; 60 < TLI(Σ) ≤ 70, moderately eutrophic; 70 < TLI(Σ) ≤ 100, severe eutrophication; the nutrient status of the water body was rated to determine the nutrient status of the middle and lower reaches of the Han River, and the improvement effect after ecological regulation was compared with that before ecological regulation; Step 3: Determine the objective function of the multi-objective optimization scheduling. Select the water body nutrient status destruction rate and the multi-year average density of phytoplankton as the ecological objectives of the scheduling. Select the multi-year average water supply and multi-year average power generation as the economic objectives of the scheduling. Then, use the multi-objective optimization method based on the cuckoo algorithm to solve the algal bloom simulation model and obtain the non-inferior solution set of the model. Step 4: Select a satisfactory solution for ecological scheduling from the non-inferior solution set according to the optimal solution, draw a reservoir scheduling diagram, and obtain the algal bloom prevention and control scheduling plan.

2. The method for controlling water blooms according to claim 1, wherein: In step 1, the flow data includes a flow sequence corresponding to time, and the data interval should be 10 days or less. The ecological data includes total nitrogen, total phosphorus, chlorophyll a concentration, water transparency, chemical oxygen demand, and phytoplankton density in the river, and the data interval should be 30 days or less. The flow data and ecological data correspond in both time and space.

3. The method for controlling water blooms according to claim 1, wherein: In step 2, the time scale of the model input is ten days, and the flow unit is m 3 / s; The comprehensive trophic status index is a dimensionless quantity, and the unit of phytoplankton density is cells / L.

4. The method for controlling water blooms according to claim 1, wherein: The model hyperparameters include HiddenUnits, MaxEpochs, GradientThreshold, InitialLearnRate, LearnRateDropPeriod, and LearnRateDropFactor; the hyperparameters used in the training model are adjusted according to the characteristics of the input data.

5. The method for controlling water blooms according to claim 1, wherein: In step 3, the objective function of the multi-objective optimization is determined, and the calculation formula of the objective function is as follows: (1) (2) (3) (4) Where: WS is the average water supply over many years, S is the total length of the scheduling period, t For the scheduling period, Wst is the water supply in each period; HG is the average power generation over many years, Ht is the power generation in each period; is the destruction rate of water body nutrient status, is the length of the period during which the disruption occurs during the scheduling period; MPD is the multi-year average phytoplankton density, is the phytoplankton density at each time period.

6. The method for controlling water blooms according to claim 1, wherein: In step 3, the boundary conditions in the multi-objective optimization process are as follows: (1) Water balance constraints （5） Where: Vi Indicates the i Storage capacity during the period, Q Ii Indicates the inflow flow in each period, Q Oi Indicates the outbound flow in each period, Δ t is the time interval; (2) Reservoir capacity constraints （6） Where: 、 Respectively represent i The upper and lower bounds of storage capacity for the time period; (3) Outbound flow constraints （7） Where: 、 Respectively represent i The upper and lower bounds of outbound flow during the time period; (4) Initial state of the reservoir （8） Where: represents the initial storage capacity during the scheduling period, represents the storage capacity at the end of the scheduling period; (5) Minimum flow limit of river （9） Where: Indicates the minimum ecological flow in the downstream river channel.

7. The method for controlling water blooms according to claim 1, wherein: In step 2, the Nash efficiency coefficient NSE and mean absolute percentage error MAPE are used to evaluate the LSTM simulation effect. The calculation formula is as follows: （10） （11） Where: y k is the original sequence, Output sequence for the model, n is the total number of data sets, k Indicates the first k data.

Images