Prediction, grading and early warning method for cyanobacterial bloom in large-scale reservoir

Abstract

The invention discloses a method for predicting, grading and early warning cyanobacterial bloom in a large reservoir. The method comprises the following steps: acquiring a reservoir water change period D at a prediction moment, and assigning a value to a water change period influence factor F (D); acquiring an air temperature T at a prediction moment and assigning a value to an air temperature influence factor F (T); the rainfall R at the prediction moment is obtained, and the air temperature influence factor F (R) is assigned; acquiring a wind speed V at a prediction moment and assigning a value to a wind speed influence factor F (V); calculating the chlorophyll a concentration C at a prediction moment through an LSTM model, and assigning a value to a chlorophyll a influence factor F (C); calculating the occurrence probability F of the cyanobacterial bloom in the reservoir according to each influence factor: F = F (D) * F (C) * F (T) * 0.5 [F (R) + F (V)]; and obtaining a corresponding early warning level according to the occurrence probability F of the cyanobacterial bloom in the reservoir. The scheme provides key technical support and decision basis for accurate prevention and control of cyanobacterial bloom.

Description

Technical Field

[0001] The invention relates to the field of water quality monitoring, and in particular to a classification and early warning method for predicting blue algae blooms in large reservoirs. Background Art

[0002] Normally, reservoirs store water during the annual flood season. During this period, the water level of the reservoir will rise significantly due to the increase in rainfall during the rainy season. The rising water level of the reservoir will support the rivers entering the reservoir, causing the hydrodynamics of the rivers entering the reservoir to weaken. The high-temperature season overlaps with the high-water season, and the rainfall brings rich nutrients, forming a habitat that is very conducive to the reproduction and growth of algae at the entrance of the river. If it is not controlled and handled in time, it will easily spread under the action of wind, forming lake flooding and even blue algae blooms in the reservoir area. Large reservoirs are often important sources of drinking water, and the outbreak of blue algae blooms will produce unpleasant odor substances, and toxic and harmful algae toxins will be produced during the extinction of blue algae, all of which pose a serious threat to drinking water sources and affect the safety of water intake and supply. In order to monitor the growth of algae, relevant departments usually arrange automatic algae monitoring equipment at the confluence of the main rivers entering large reservoirs and near the dams, closely monitor the algae biomass (chlorophyll a), and judge the algae growth based on the monitoring results. When the biomass reaches a certain limit, it will be supplemented by manual inspections to determine the prevention and control measures and means to be taken based on the on-site situation. Algae, especially cyanobacteria, proliferate very quickly, and algae conditions change rapidly. Early prevention and control and accurate measures are the best way to prevent the outbreak of cyanobacteria. However, the method of monitoring plus manual inspections not only has a lag but also has a certain degree of blindness, which cannot meet the needs of intelligent prediction and graded warning.

[0003] At present, there are two main modeling strategies for predicting the dynamic changes of algae: mechanism process model and machine learning algorithm. Among them, the method based on ecological dynamics process is a more effective technology for simulating and analyzing the dynamic changes of algae, and is widely used in the analysis of long-term succession trends of aquatic ecosystems. For example, principal component analysis (PCA), extreme learning machine (ELM), recurrent neural network (RNN), long short-term memory neural network (LSTM) and other technologies have been reported in the prediction of algae in shallow lakes, ponds, rivers and other water bodies.

[0004] However, the ecological processes involved in cyanobacterial blooms are still unclear in mechanism, and the prediction model expressed purely by mathematics still has many limitations. For example, the current selection of model parameters is heavily dependent on repeated debugging by researchers, which requires a lot of manpower and computing resources. At the same time, there is a lack of consideration for the dependency relationship between a single batch of input time series data. Therefore, the model prediction performance is subject to certain limitations, and it is usually only applicable to small water bodies. There are few precedents for prediction in large reservoirs, and there is usually a disconnect between the prediction and the actual early warning and prevention. The practicality, accuracy, and operability of the technology have varying degrees of defects. Summary of the invention

[0005] The present invention mainly solves the technical problems existing in the prior art such as being unable to be applied to large reservoirs and having poor accuracy and operability, and provides a graded early warning method for predicting cyanobacteria blooms in large reservoirs with high prediction accuracy.

[0006] The present invention aims to solve the above technical problems mainly through the following technical solutions: A method for predicting and grading early warning of blue algae blooms in large reservoirs, comprising the following steps: S1. Obtain the reservoir water exchange cycle D at the predicted time, and assign a value to the water exchange cycle influencing factor F(D) according to the reservoir water exchange cycle D; S2. Obtain the temperature T at the prediction time, and assign a value to the temperature influencing factor F(T) according to the temperature T; S3, obtaining the rainfall R at the predicted time, and assigning a value to the temperature influencing factor F(R) according to the rainfall R; S4, obtaining the wind speed V at the prediction time, and assigning a value to the wind speed influencing factor F(V) according to the wind speed V; S5. Calculate the chlorophyll a concentration C at the prediction time through the LSTM model, and assign a value to the chlorophyll a influence factor F(C) according to the chlorophyll a concentration C; S6. Calculate the probability of occurrence of cyanobacteria bloom in reservoirs based on various influencing factors: F=F(D)×F(C)×F(T)×0.5[F(R)+F(V)] S7. Obtain the corresponding warning level according to the probability F of occurrence of cyanobacteria bloom in the reservoir.

[0007] Preferably, the hyperparameter optimization process of the LSTM model is: A1. Set the hyperparameter group of the LSTM model as the optimization object, and randomly initialize the values ​​of m groups of hyperparameter groups according to the pre-set range; the value of m is set manually; A2. Select the i-th group of hyperparameters to build the LSTM model, and then train the LSTM model with the data in the training set; 1≤i≤m; A3. After the training is completed, calculate the prediction error of the model on the validation set as the current fitness value of the $i$-th hyperparameter group; A4. Optimize the hyperparameter group through the following formula: V i (k + 1)=ωV i (k)+c1r1(P i (k)-X i (k))+c2r2(P g (k)-X i (k)) X i (k + 1)=X i (k)+V i (k + 1) V i (k) is the change amount of the $i$-th hyperparameter group at the $k$-th iteration; X i (k) is the value of the $i$-th hyperparameter group at the $k$-th iteration; P i (k) is the best value found so far for the $i$-th hyperparameter group; P g (k) is the best value found so far for all groups of hyperparameters; the best value is the value corresponding to the minimum prediction error; ω is the inertia weight, which controls the balance between the individual and the global for the hyperparameter group; c1 and c2 are acceleration factors, which adjust the step size of the change amount; r1 and r2 are random numbers between 0 and 1; the inertia weight and the two acceleration factors are empirical parameters, which are set manually; A5. Judge whether the end optimization condition is met. If it is met, the optimization process ends; if not, go to step A6; the end optimization condition is to reach the set maximum number of iterations or the fitness value no longer changes significantly with the number of iterations, for example, the change rate is less than 1%; A6. If $i \lt m$, increase $i$ by 1 and jump to step A2; if $i = m$, set $i$ to 1 and jump to step A2; After the optimization is completed, construct an LSTM model using the hyperparameter group at the end, and then train this LSTM model with the training data. The Adam algorithm is used to optimize the parameters during the training process. After the training is completed, the chlorophyll a concentration can be predicted through the LSTM model.

[0008] Preferably, the number of neurons in the input layer of the LSTM model is 4, the number of hidden layers is 1, the number of neurons in the output layer is 1, and the required prediction step is 40; the hyperparameters in the hyperparameter group include the sample batch size, the number of hidden layer units, the learning rate, and the number of iteration values. The input parameters of the LSTM model are the air temperature, wind speed, cross-sectional flow velocity, and rainfall at the prediction moment.

[0009] Preferably, the chlorophyll a influence factor F(C) is assigned as follows: if the chlorophyll a concentration C is greater than or equal to 60 mg / L, then F(C)=1; if the chlorophyll a concentration C is greater than 10 mg / L and less than 60 mg / L, then F(C)=0.012C+0.2; if the chlorophyll a concentration C is less than 10 mg / L, then F(C)=0.2.

[0010] Preferably, in step S1, the reservoir water replacement cycle is obtained by the following formula: D=aH / Q 入_now In the formula, a is the correction coefficient, H is the reservoir capacity, Q 入_now is the current reservoir inflow; the correction factor is obtained by the following formula: a=Q 入_总 / =Q 出_总 ; In the formula, Q 入_总 is the total inflow of the reservoir from day 1 to day n, Q 出_总 is the total outflow of the reservoir from the 1st day to the nth day, where n is the number of days for statistics, and n≥10; The water exchange cycle influencing factor F(D) is assigned as follows: if the water exchange cycle is greater than 26 days, F(D)=1.0; if the water exchange cycle is greater than 13 days and less than or equal to 26 days, F(D)=0.8; if the water exchange cycle is greater than 9 days and less than or equal to 13 days, F(D)=0.6; if the water exchange cycle is greater than or equal to 6 days and less than or equal to 9 days, F(D)=0.4; if the water exchange cycle is less than 6 days, F(D)=0.2. The inflow and outflow flow of the reservoir can be obtained by multiplying the flow velocity detected by the flow meter set at the inflow monitoring section and the outflow monitoring section by the monitoring section area. If there are multiple inlets or outlets, all flows of the same type need to be added together.

[0011] Preferably, the temperature T at the predicted time is obtained according to the weather forecast, and the temperature influence factor F(T) is assigned as follows: if the temperature T is less than or equal to 25 degrees Celsius, F(T)=0.4; if the temperature T is greater than 25 degrees Celsius and less than 33 degrees Celsius, F(T)=0.075T-1.475; if the temperature T is greater than 33 degrees Celsius, F(T)=1.

[0012] Preferably, the rainfall R at the predicted time is obtained according to the weather forecast, and the assignment method of the rainfall influencing factor F(R) is: if the rainfall R is less than or equal to 10 mm, F(R)=1; if the rainfall R is greater than 10 mm and less than 50 mm, F(R)=-0.015R+1.15; if the rainfall R is greater than or equal to 50 mm and less than or equal to 250 mm, F(R)=0.4; if the rainfall R is greater than 250 mm, F(T)=0.2.

[0013] Preferably, the wind speed V at the predicted time is obtained according to the weather forecast, and the wind speed influence factor F(V) is assigned as follows: if the wind speed V is less than or equal to 0.5 m / s, F(V)=1; if the wind speed V is greater than 0.5 m / s and less than 8 m / s, F(V)=-0.1067V+1.053; if the wind speed V is greater than or equal to 8 m / s, F(V)=0.2.

[0014] As a preferred method, the grading method for warning levels of blue algae bloom outbreaks is as follows: ① Low risk: risk probability F ≤ 20%; ②Medium risk: risk probability 20% <F ≤ 40%; ③Higher risk: risk probability 40% <F ≤ 60%; ④High risk: risk probability F>60%.

[0015] Preferably, the training data is obtained in the following manner: five data items of temperature, wind speed, cross-sectional flow velocity, rainfall and chlorophyll a concentration are observed by an automatic monitoring station; the time interval of the observation data is 6-24 hours; the observed data are normalized; the normalized data are divided into a training set, a validation set and a test set according to a ratio of 3:1:1, the data span of the training set is not less than 18 months, the data span of the validation set is not less than 6 months, and the data span of the test set is not less than 6 months; if the original data is missing, linear interpolation is used to complete it.

[0016] Preferably, in step S1, after obtaining the reservoir water replacement cycle D, the reservoir water replacement cycle is corrected in the following manner: D'=D+ΔD ΔD is the water change cycle correction value, which is obtained by the following formula: ΔD=k1*(RR avg )+k2*(TT avg )+k3*(R prev_-1 -R prev_avg )+k4*(R prev_-2 -R prev_avg ) In the formula, k1 is the rainfall influence coefficient during prediction, R is the rainfall at the prediction time, and R avg is the average rainfall, k2 is the temperature influence coefficient during prediction, T is the temperature at the time of prediction, T avg is the average temperature, k3 is the rainfall influence coefficient of the previous time step, R prev_-1 is the rainfall in the previous time step, R prev_avg is the average rainfall in the previous period, k4 is the rainfall influence coefficient in the first two time steps, R prev_-1 is the rainfall in the first two time steps; Then use D' to replace D to calculate the water change cycle influencing factor F(D).

[0017] The time step refers to the interval between discrete time points, and the specific length is determined according to the actual situation, such as 1 hour, 12 hours, 24 hours, and 48 hours. For example, when the time step is one day (24 hours), it is necessary to predict the water bloom on August 20, 2024. The rainfall in the previous time step is the rainfall on August 19, 2024, and the rainfall in the previous two time steps is the rainfall on August 18, 2024. If the current time is far from the prediction moment, the rainfall in the previous time step and the rainfall in the previous two time steps use the predicted values ​​of the meteorological forecast. When the risk is small, a longer time step can be used to reduce the calculation cost; when the risk is large, the time step should be shortened to obtain more accurate prediction results.

[0018] In this scheme, rainfall refers to the rainfall in a time step, and temperature refers to the average temperature in a time step. avg The average rainfall over a long period of time, generally not less than one year (preferably three years), is the deviation between the rainfall at the predicted time and the normal rainfall level (RR avg ) can assess the impact of precipitation at the time of prediction relative to the long-term climate background; R prev_avg is the average rainfall in a relatively short period of time, generally no more than 15 days (preferably 7 days), reflecting the trend and level of recent rainfall. The deviation between the previous time step, the previous two time steps and the recent rainfall (R prev_-1 -R prev_avg , R prev_-2 -R prev_avg ) can be used to assess the impact of recent rainfall on water exchange cycles; distinguish R avg and R prev_avg It can capture the impact of long-term climate average rainfall level and short-term rainfall trend on water exchange cycle respectively, and can also have good prediction accuracy for exceptional climate (such as high average rainfall over many years but no rain in the recent period or vice versa).

[0019] The calculation of the water exchange cycle correction value introduces the influence of the rainfall in the previous time step and the previous two time steps, and takes into account the time lag effect. After the rainfall occurs, it will cause the flow of the reservoir to change in a certain period of time. The introduction of the time lag effect can more accurately describe this influence. k1, k2, k3 and k4 are empirical parameters, which can generally be set to k1=-0.02, k2=-0.01, k3=-0.01, k4=-0.005.

[0020] The substantial effects brought by the present invention are: 1. constructing an optimized LSTM chlorophyll a prediction model, which can find the optimal parameters of the model through continuous iterative optimization, avoiding the problems of low prediction accuracy and long parameter adjustment time caused by improper parameter selection; 2. creatively introducing the influencing factor of the reservoir water exchange cycle into the probability of cyanobacteria bloom outbreak, and making the water exchange cycle convenient to obtain by monitoring the inflow rate in the form of coefficient conversion; 3. establishing a cyanobacteria bloom outbreak probability calculation formula based on fitting multiple factors according to historical data, cyanobacteria bloom outbreak rules and actual experience, and the calculation result is highly consistent with the actual situation; this scheme can help managers of relevant departments to quickly diagnose changes in algae conditions, make correct and timely prevention and control decisions, reduce the risk of cyanobacteria bloom, reduce the cost of cyanobacteria bloom management, effectively ensure the water ecological environment quality and drinking water safety of large reservoirs, and has considerable economic value and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of a method for predicting and grading early warning of blue algae blooms in large reservoirs according to the present invention; Figure 2 It is a hyperparameter optimization process diagram of an LSTM model of the present invention. DETAILED DESCRIPTION

[0022] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0023] Embodiment: A method for predicting and warning of cyanobacteria blooms in large reservoirs according to this embodiment is as follows: Figure 1 As shown, the following steps are included: S1. Obtain the reservoir water exchange cycle D at the predicted time, and assign a value to the water exchange cycle influencing factor F(D) according to the reservoir water exchange cycle D; S2. Obtain the temperature T at the prediction time, and assign a value to the temperature influencing factor F(T) according to the temperature T; S3, obtaining the rainfall R at the predicted time, and assigning a value to the temperature influencing factor F(R) according to the rainfall R; S4, obtaining the wind speed V at the prediction time, and assigning a wind speed influencing factor F(V) according to the wind speed V; S5. Calculate the chlorophyll a concentration C at the prediction time through the LSTM model, and assign a value to the chlorophyll a influence factor F(C) according to the chlorophyll a concentration C; S6. Calculate the probability of occurrence of cyanobacteria bloom in reservoirs based on various influencing factors: F=F(D)×F(C)×F(T)×0.5[F(R)+F(V)] S7. According to the probability F of occurrence of cyanobacteria bloom in the reservoir, the corresponding warning level is obtained.

[0024] like Figure 2 As shown, the hyperparameter optimization process of the LSTM model is as follows: A1. Set the hyperparameter group of the LSTM model as the object to be optimized, and randomly initialize the values of m groups of hyperparameter groups according to the preset range; the value of m is set manually; A2. Select the i-th group of hyperparameter groups to construct an LSTM model, and then use the data in the training set to train the LSTM model; 1 ≤ i ≤ m; A3. After the training is completed, calculate the prediction error of the model on the validation set as the current fitness value of the i-th hyperparameter group; A4. Optimize the hyperparameter group through the following formula: V i (k + 1)=ωV i (k)+c1r1(P i (k)-X i (k))+c2r2(P g (k)-X i (k)) X i (k + 1)=X i (k)+V i (k + 1) V i (k) is the change amount of the i-th hyperparameter group at the k-th iteration; X i (k) is the value of the i-th hyperparameter group at the k-th iteration; P i (k) is the best value found so far for the i-th hyperparameter group; P g (k) is the best value found so far for all groups of hyperparameters; the best value is the value corresponding to the minimum prediction error; ω is the inertia weight; c1 and c2 are acceleration factors; r1 and r2 are random numbers between 0 and 1; A5. Judge whether the optimization termination condition is met. If it is met, the optimization process ends; if not, go to step A6; A6. If i < m, increase i by 1 and jump to step A2; if i = m, set i to 1 and jump to step A2.

[0025] After the optimization is completed, use the global best value P g (k) as the finally determined hyperparameter group to construct an LSTM model, and then use the training data to train this LSTM model.

[0026] In this embodiment, several algae profile buoy automatic monitoring stations are set up at the confluence of the river entering the large reservoir and in the reservoir area, and several cross-sectional flow velocity automatic monitoring stations are set up in the reservoir area. The more automatic monitoring stations there are, the higher the prediction accuracy. The present invention requires at least one algae profile buoy automatic monitoring station at the confluence of the river entering the reservoir and in the reservoir area, and at least one flow velocity automatic monitoring station in the reservoir area. The station monitoring data is automatically uploaded and integrated into the data warehouse, and the system collects data for a certain period of time for model training.

[0027] The collected data are preprocessed in the following ways: ① Data completion. For high-frequency monitoring data, the sliding average method can be used to control the time interval of the observation data within 6-24 hours. Correspondingly, the time interval between two prediction results is the same as the time interval of the observation data. For the missing original data of the automatic monitoring station, the linear difference method is used to complete the original data of the station in the preprocessing stage. Assuming that the original data has a total of n points, assuming that the i-th value x i Missing (i≤n), can be completed using the following formula: i =a(x i+b -x i-a ) / (a+b)+x i-a ; In the formula, x i+b For x i The adjacent non-empty data after that, and x i The difference is b data; x i-a For x i The previous adjacent non-empty data, and x i The difference is a data.

[0028] ② Data normalization. In order to eliminate singular sample data and reduce model training time, the data is normalized and the original data is mapped to (0, 1]. The normalized data is divided into training set, validation set and test set according to 3:1:1. Among them, the data of the training set (18 months) is used to train the neural network model; the data of the validation set (6 months after the training set) is used to evaluate the prediction effect of the model; the data of the test set (6 months after the validation set) is used to test the prediction accuracy of the model.

[0029] In order to reduce the data dimension, the influencing factors with high correlation with chlorophyll a concentration were screened out, and the Pearson correlation coefficient was used for correlation analysis. The value of the Pearson correlation coefficient is between -1 and 1, where 1 represents a complete positive linear correlation, -1 represents a complete negative linear correlation, and 0 indicates no linear correlation. According to the size of the correlation, the present embodiment selects influencing factors with certain significant correlations such as temperature, wind speed, cross-sectional flow velocity, and rainfall as model input parameters.

[0030] This embodiment optimizes and transforms the long short-term memory neural network (LSTM), sets the sample batch, number of hidden layer units, learning rate, number of iterations and other parameters in the LSTM model as optimization objects and gives a value range, initializes the value of each hyperparameter according to the parameter range and establishes an initial model, and continuously updates the prediction error of the model in the validation set as the fitness value. When the maximum number of iterations is met, the update optimization is stopped, and the optimized hyperparameters are used to construct an LSTM prediction model to predict chlorophyll a. ① Long short-term memory network (LSTM), composed of an input gate, a forget gate, an output gate and a cell unit. f t is the forget gate, which determines how much information in the cells of the previous moment needs to be transferred to the current moment; i t is the input gate, which is used to control the degree to which the current unit is embedded in the cell state; o t is the output gate, which combines the cell state for output; C t It is a memory unit used to record the state of cells under different gate structures; h t is the output. The calculation process between each gate is as follows: f t =σ[W f •(C t-1 ,h t-1 ,x t )+b f ] i t =σ[W i •(C t-1 ,h t-1 ,x t )+b i ] C t ^ =tanh[W c •(h t-1 ,x t )+b c ] C t =f t •C t-1 +i t •C t ^ o t =σ[W o •(C t ,h t ,x t )+b o ] h t =ot •tanh(C t ) Where W f , W i , W c , W o is the weight of each gate, b f , b i , b c , b o is the bias term for each gate.

[0031] The LSTM model has 4 input layer neurons, 1 hidden layer neuron, 1 output layer neuron, and a step size of 40 required for prediction.

[0032] The specific model optimization and training process is as follows: Step ①, take the sample batch, number of hidden layer units, learning rate and number of iterations of LSTM as the optimization object, and initialize m groups of hyperparameter groups according to the pre-set range. Step ②, use the hyperparameter group initialized in step ① to build the LSTM model, and then use the data in the training set to train the LSTM model, and use the model prediction error as the fitness value of the hyperparameter group. Step ③, compare the fitness value of each hyperparameter group and its best position, determine the best value of the hyperparameter group, and update the change and new value of the hyperparameter group according to the optimization formula, and calculate the fitness value of the new round of hyperparameter group. Step ④, stop updating when the search process reaches the pre-set maximum number of iterations, or the fitness value of the hyperparameter group no longer changes significantly with the number of iterations, and obtain the sample batch, number of hidden layer units, learning rate, and number of iterations of the LSTM model at this time. Step ⑤, construct the LSTM model with the hyperparameter values ​​obtained in step ④, input the training data for training, input the prediction parameters after the training is completed, and output the predicted chlorophyll a predicted concentration.

[0033] In order to verify the prediction performance of the optimized LSTM model for chlorophyll a concentration, the mean absolute percentage error (MAPE), root mean square error (RMSE) and mean square error (MSE) were selected as model evaluation indicators.

[0034] MAPE represents the percentage error between the predicted value and the true value, MSE reflects the expected value of the deviation between the predicted value and the true value, and RMSE is the arithmetic square root of the mean square error. The range of MAPE, RMSE, and MSE is [0, ∞]. The smaller the calculation result, the better the model prediction effect. At the same time, a period of chlorophyll a historical data was randomly selected to evaluate the prediction accuracy of the LSTM algae prediction model, and the time period with higher prediction accuracy was selected as the prediction period of the model.

[0035] According to the correlation of cyanobacteria bloom and the availability of indicators, the temperature F(T), rainfall F(R), wind speed F(V) at the predicted time of the reservoir location and the predicted chlorophyll a concentration F(C) plus the reservoir water exchange cycle F(D) are selected to form a probability calculation system for algae bloom and lake flooding, and the probability of algae bloom occurring in the reservoir within the predicted time is calculated. The specific process is as follows: ① The data of various indicators at the prediction time, temperature, rainfall and wind speed are obtained through weather forecast; chlorophyll a is predicted by LSTM model; the reservoir water exchange cycle is obtained through the following process: The first step is to measure the depth d and width w of the reservoir flow automatic monitoring section using a rangefinder and a depth sounder. Assuming that the flow section A is a rectangle and the measured flow velocity is v, the reservoir inflow Q can be estimated. 入 , i.e. Q 入 =A×v=d×w×v, m 3 / s meter.

[0036] The second step is to determine the correction coefficient. The predicted reservoir inflow and the actual reservoir outflow are divided to obtain the correction coefficient a, that is, a=Q 入_总 / =Q 出_总 , Q 入_总 is the total inflow of the reservoir from day 1 to day n, Q 出_总 It is the total outflow of the reservoir from the 1st day to the nth day, where n is the number of days for statistics. In order to make the correction coefficient more realistic, at least 10 days must be calculated, that is, n≥10.

[0037] The third step is to use the program to automatically obtain the public storage capacity of the reservoir (H), so as to calculate the reservoir water exchange cycle (D), in days, that is, D=aH / Q 入_now , Q 入_now is the current reservoir inflow.

[0038] Step 4: After obtaining the reservoir water exchange cycle D, the reservoir water exchange cycle is corrected by the following method: D'=D+ΔD ΔD is the water change cycle correction value, which is obtained by the following formula: ΔD=k1*(RR avg )+k2*(TT avg )+k3*(R prev_-1 -R prev_avg )+k4*(R prev_-2 -R prev_avg ) In the formula, k1 is the rainfall influence coefficient during prediction, R is the rainfall at the prediction time, and R avg is the average rainfall, k2 is the temperature influence coefficient during prediction, T is the temperature at the time of prediction, T avgis the average temperature, k3 is the rainfall influence coefficient of the previous time step, R prev_-1 is the rainfall in the previous time step, R prev_avg is the average rainfall in the previous period, k4 is the rainfall influence coefficient in the first two time steps, R prev_-1 is the rainfall in the first two time steps; Subsequently, D' is used instead of D to calculate the water change cycle influencing factor F(D).

[0039] The time step refers to the interval between discrete time points, and the specific length is determined according to the actual situation, such as 1 hour, 12 hours, 24 hours, and 48 hours. For example, when the time step is one day (24 hours), it is necessary to predict the water bloom on August 20, 2024. The rainfall in the previous time step is the rainfall on August 19, 2024, and the rainfall in the previous two time steps is the rainfall on August 18, 2024. If the current time is far from the prediction moment, the rainfall in the previous time step and the rainfall in the previous two time steps use the predicted values ​​of the meteorological forecast. When the risk is small, a longer time step can be used to reduce the calculation cost; when the risk is large, the time step should be shortened to obtain more accurate prediction results.

[0040] In this embodiment, the rainfall represents the rainfall in a time step, and the temperature represents the average temperature in a time step. avg is the average rainfall over a long period of time, which is 3 years in this embodiment. The deviation (RR) between the rainfall at the predicted time and the normal rainfall level is calculated. avg ) can assess the impact of precipitation at the time of prediction relative to the long-term climate background; R prev_avg is the average rainfall in a relatively short period of time, which is 7 days in this embodiment, reflecting the trend and level of recent rainfall. The deviation between the previous time step, the previous two time steps and the recent rainfall situation (R prev_-1 -R prev_avg , R prev_-2 -R prev_avg ) can be used to assess the impact of recent rainfall on water exchange cycles; distinguish R avg and R prev_avg It can capture the impact of long-term climate average rainfall level and short-term rainfall trend on water exchange cycle respectively, and can also have good prediction accuracy for exceptional climate (such as high average rainfall over many years but no rain in the recent period or vice versa).

[0041] The calculation of the water exchange cycle correction value introduces the influence of the rainfall in the previous time step and the previous two time steps, and takes into account the time lag effect. After the rainfall occurs, it will cause the flow of the reservoir to change in a certain period of time. The introduction of the time lag effect can more accurately describe this influence. k1, k2, k3 and k4 are empirical parameters, which can generally be set to k1=-0.02, k2=-0.01, k3=-0.01, k4=-0.005.

[0042] ② Assign values ​​to F(D). Based on experience, assign corresponding probabilities to the water exchange cycle of reservoirs and the occurrence of cyanobacterial blooms during the high temperature period in summer and autumn. The specific values ​​are as follows: if the water exchange cycle is greater than 26 days, F(D)=1.0; if the water exchange cycle is greater than 13 days and less than or equal to 26 days, F(D)=0.8; if the water exchange cycle is greater than 9 days and less than or equal to 13 days, F(D)=0.6; if the water exchange cycle is greater than or equal to 6 days and less than or equal to 9 days, F(D)=0.4; if the water exchange cycle is less than 6 days, F(D)=0.2.

[0043] ③ Calculate the empirical probability value F(C) of cyanobacteria bloom caused by chlorophyll a concentration at the predicted time. The specific calculation is as follows: if the chlorophyll a concentration C is greater than or equal to 60 mg / L, then F(C)=1; if the chlorophyll a concentration C is greater than 10 mg / L and less than 60 mg / L, then F(C)=0.012C+0.2; if the chlorophyll a concentration C is less than 10 mg / L, then F(C)=0.2.

[0044] ④ Calculate the empirical probability value F(T) of cyanobacteria bloom caused by air temperature at the predicted time. The specific calculation is as follows: if the air temperature T is less than or equal to 25 degrees Celsius, then F(T)=0.4; if the air temperature T is greater than 25 degrees Celsius and less than 33 degrees Celsius, then F(T)=0.075T-1.475; if the air temperature T is greater than 33 degrees Celsius, then F(T)=1.

[0045] ⑤ Calculate the empirical probability value F(R) of cyanobacteria bloom caused by rainfall at the predicted time. The specific calculation is as follows: if the rainfall R is less than or equal to 10 mm, then F(R)=1; if the rainfall R is greater than 10 mm and less than 50 mm, then F(R)=-0.015R+1.15; if the rainfall R is greater than or equal to 50 mm and less than or equal to 250 mm, then F(R)=0.4; if the rainfall R is greater than 250 mm, then F(T)=0.2.

[0046] ⑥ Calculate the empirical probability value F(V) of cyanobacteria bloom caused by wind speed at the predicted time. The specific calculation is as follows: if the wind speed V is less than or equal to 0.5m / s, then F(V)=1; if the wind speed V is greater than 0.5m / s and less than 8m / s, then F(V)=-0.1067V+1.053; if the wind speed V is greater than or equal to 8m / s, then F(V)=0.2.

[0047] ⑦ According to the correlation between the indicators of the early warning system and the probability of occurrence of cyanobacteria bloom, the above steps are assigned corresponding weights according to experience, forming an empirical formula for the probability of cyanobacteria bloom in large reservoirs based on factors such as reservoir water exchange cycle, chlorophyll a concentration, temperature, rainfall, wind speed, etc.: F=F(D)×F(C)×F(T)×0.5[F(R)+F(V)] According to the results of the monitoring and early warning system, the warning level of blue algae bloom outbreaks in large reservoirs can be divided into four levels: ① Low risk (blue): risk probability F ≤ 20%, algae density <1.0×10 7 / L, cyanobacteria become the dominant species, but no monitoring or forecast of algal blooms. ② Medium risk (yellow): risk probability 20% <F ≤ 40%,1.0×10 7 Pieces / L ≤ Algae density <5.0×10 7 / L, blue algae become the dominant species, and algae residues appear sporadically in the reservoir stagnation area. ③ Higher risk (orange): risk probability 40% <F ≤ 60%,5.0×10 7 Pieces / L ≤ Algae density <1.0×10 8 / L, cyanobacteria become the dominant species, and patchy algal blooms appear in the reservoir, with an area of ​​less than 10% of the reservoir surface area. ④ High risk (red): risk probability F>60%, algae density ≥1.0×10 8 / L, cyanobacteria became the dominant species, and sheet-like algae blooms appeared in the reservoir, accounting for more than 10% of the total water surface area of ​​the reservoir. After triggering the blue, yellow, orange, and red warnings, the probability of cyanobacteria bloom outbreaks and the corresponding levels will be displayed on the terminal (computer, mobile phone, large screen), and reminder information will be pushed in the APP. Relevant members of the registered cyanobacteria bloom prevention and control leading group will also receive text message reminders and emergency response suggestions as soon as possible.

[0048] Blue warnings generally do not initiate emergency responses to blue algae blooms, and only require tracking and monitoring of algae density. Yellow warnings recommend the initiation of a Level III emergency response to algae blooms, and the salvage and disposal of blue algae in landscape areas or living areas where people are sensitive, to ensure that blue algae gather and are cleared daily to prevent the accumulation of nearshore blue algae and affect the lives of surrounding residents and sightseeing. For drinking water sources, emergency measures should focus on blue algae capture to ensure that there is no blue algae bloom within 500 meters of the water intake to ensure water supply safety. For ecologically sensitive areas, emergency measures should focus on blue algae capture to ensure that there is no blue algae bloom in ecologically sensitive areas. Orange warnings recommend the initiation of a Level II emergency response to blue algae blooms, timely organization of blue algae capture operations, the dispatch of blue algae vehicles (ships), algae-water separation stations and other blue algae emergency salvage and treatment equipment, and increased blue algae salvage and treatment capabilities. In case of drinking water sources, emergency measures should be supplemented by in-situ control of cyanobacteria on the basis of cyanobacteria capture, to ensure that there is no cyanobacteria bloom within 500 meters outside the water intake; if black and smelly or lake overflow occurs, measures such as oxygenation and aeration can be implemented; at the same time, check the backup water source and emergency water supply facilities. In case of ecologically sensitive areas, emergency measures should be supplemented by oxygenation and aeration, in-situ control of cyanobacteria on the basis of cyanobacteria capture, to ensure that there is no cyanobacteria bloom in water ecological sensitive areas and to increase the dissolved oxygen level in the water body. For reservoirs with conditions for water diversion and diversion, water diversion and replenishment measures can be taken. The red alert recommends the initiation of a Level I emergency response to cyanobacteria blooms, the continuous organization of cyanobacteria capture operations, the implementation of oxygenation and aeration in key waters; and the extension of the operating time of cyanobacteria emergency salvage and treatment equipment such as algae vehicles (ships) and algae-water separation stations depending on the algae situation. In case of drinking water sources, timely start backup water sources or emergency water supply facilities, strengthen pollutant monitoring, and resume water intake when water quality meets water intake requirements; in case of black and smelly or large-scale lake flooding, implement measures such as oxygenation and aeration, organize emergency dredging, etc. In case of ecologically sensitive areas, increase the amount of oxygenation and aeration operations and the salvage and disposal of blue algae, and promptly salvage and dispose of dead aquatic plants, fish, and benthic animals. For reservoirs with water diversion and drainage conditions, water diversion and replenishment measures can be adopted.

[0049] The following are two verification examples: Verification Example 1: In this verification example, the reservoir capacity is selected to be 440 million m 3 The intelligent prediction method and system construction of chlorophyll a concentration in large reservoirs was carried out in Fuchunjiang Reservoir, including the following steps: 1. Data retrieval: The Fuchun River Reservoir has built an automatic algae profile monitoring station at the confluence of the main tributaries and in the main reservoir area. An automatic flow velocity monitoring station is also built in the main reservoir area. The data are integrated into the data warehouse. The monitoring data including cross-sectional flow velocity and chlorophyll a concentration from March 2021 to October 2023 are retrieved from the data warehouse. In addition, atmospheric data such as wind speed, temperature, and precipitation are collected during the same period.

[0050] 2. Data preprocessing: ① Data completion. For high-frequency monitoring data, the sliding average method can be used to control the time interval of the observation data within 6-24 hours, so that the time interval between the two prediction results is the same as the time interval of the observation data, and the missing data is completed by the linear interpolation method. ② Data normalization and division. After data completion, the data is normalized, the original data is mapped to (0, 1], and the normalized data is divided into training set, validation set and test set according to 3:1:1. Among them, the data of the training set (March 1, 2021-October 31, 2022) is used to train the neural network model; the data of the validation set (November 1, 2022-April 30, 2023) is used to evaluate the prediction effect of the model; the data of the test set (May 1, 2023-September 30, 2023) is used to test the prediction accuracy of the model.

[0051] 3. Correlation analysis: In order to reduce the data dimension and screen out the influencing factors with a high correlation with chlorophyll a concentration, the Pearson correlation coefficient was used for correlation analysis. After calculation, the Chla concentration in the Fuchunjiang Reservoir was significantly positively correlated with the temperature (r=0.27, p<0.01) and wind speed (r=0.065, p<0.01), and was significantly negatively correlated with rainfall (r=-0.01, p<0.01) and cross-sectional flow velocity (r=-0.06, p<0.01). Therefore, in this example, influencing factors with certain significant correlations such as temperature, wind speed, cross-sectional flow velocity, and rainfall were selected as model input parameters.

[0052] 4. Set LSTM model parameters and conduct prediction: This example environment management uses Anaconda, the programming language is Python3.10, and the model training framework is Tensorflow2.0 based on Keras. The number of input layer neurons of the LSTM model is 4, the number of hidden layers is 1, the number of output layer neurons is 1, and the step length required for prediction is 40. The specific value range is set as follows: the sample batch value range is [4, 128], the number of hidden layer units is [2, 20], the learning rate is [0.000 1, 0.010 0], and the number of iterations is [30, 300]. The number of hyperparameter groups is 20, the acceleration factor c1=c2=2, the inertia weight ω is set to 0.8, and the maximum number of iterations is 300. Carry out model training according to the algorithm flow of the scheme of the present invention and output the prediction result of chlorophyll a concentration.

[0053] 5. Model evaluation: The model prediction results were verified using the MAPE, RMSE, and MSE formulas. The results showed that the average MAPE, average RMSE, and average MSE values ​​of the Fuchunjiang Reservoir chlorophyll a prediction results during the model verification phase were 0.354, 0.114, and 0.013, respectively. The overall prediction performance was outstanding, with a small deviation between the predicted value and the true value and high accuracy.

[0054] A segment of data from each month in 2023 was randomly selected to evaluate the algae prediction accuracy of the LSTM algae prediction model. The results showed that the simulation accuracy of the model gradually decreased with the increase of the prediction time, and the prediction accuracy of chlorophyll a concentration gradually decreased. Among them, the chlorophyll a prediction value of the model in the first three days was in good agreement with the measured value, with a small error. Therefore, the prediction period selected by the present invention was 3 days.

[0055] Verification Example 2: In this example, the Hangzhou Ecological Intelligence Health Platform was selected to develop a cyanobacteria bloom classification early warning method and system construction. The specific steps are as follows: 1. Construct a warning indicator system for cyanobacteria bloom. The chlorophyll a concentration (C) predicted by the LSTM algae prediction model of the Fuchunjiang Reservoir is combined with the temperature (T), wind speed (V), rainfall (R) and water exchange cycle (D) of the same time series to construct a warning indicator system for cyanobacteria bloom. The probability of each indicator causing a cyanobacteria reservoir is calculated, F(D), F(C), F(T), F(R), and F(V). The probability of cyanobacteria bloom outbreak in the reservoir is calculated using the empirical formula for the probability of cyanobacteria bloom.

[0056] 2. System integration and terminal display. The prediction model control terminal automatically starts the model program at 1:00 a.m. on the same day, calculates the prediction results within the last three days (including the current day), and synchronizes the prediction results within three days (the probability of occurrence of cyanobacteria bloom) to the Hangzhou Ecological Smart Health Intermediate Database, which is automatically read by the Hangzhou Ecological Smart Health Platform and displayed on the terminal.

[0057] 3. Graded warning and classified prevention and control. According to the set graded warning standards for blue algae blooms, warning information will be issued, emergency response will be initiated as appropriate, and blue algae bloom prevention and control measures will be taken.

[0058] The following are two comparative examples: Comparative Example 1: The chlorophyll a concentration of the Fuchunjiang Reservoir with the same time series was predicted using an unoptimized LSTM prediction model, and the mean values ​​of MAPE, RMSE, and MSE were 0.662, 0.220, and 0.048, respectively. The mean values ​​of MAPE, RMSE, and MSE of Example 1 were 0.354, 0.114, and 0.013, respectively, with a difference of 0.308, 0.106, and 0.035, respectively. The prediction model constructed by the present invention has a significant effect on improving the prediction performance, and has better accuracy and stability.

[0059] Comparative Example 2: After excluding the reservoir water exchange cycle, a cyanobacteria bloom early warning index system was constructed, and the probability of cyanobacteria bloom outbreak in the Fuchun River Reservoir in the same time period was calculated. The results showed that during the winter and spring periods with lower temperatures, the calculated results were basically consistent with the actual situation, but the degree of consistency was only 70% in the summer and autumn high temperature seasons; while the early warning index system constructed by the present invention had a degree of consistency of more than 95% between the calculated results and the actual situation in summer and autumn. It can be seen that the addition of the water exchange cycle indicator in summer and autumn can significantly improve the accuracy of cyanobacteria bloom early warning.

[0060] In the 45 early warning accuracy assessments conducted from September 13, 2023 to October 28, 2023, the prediction results were all low risk, which was consistent with the risk probability shown by the measured chlorophyll a concentration, and the low risk prediction accuracy was 100%.

[0061] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

[0062] Although the terms such as reservoir water exchange cycle, influencing factor, and hyperparameter are used more frequently in this article, the possibility of using other terms is not excluded. These terms are used only to more conveniently describe and explain the essence of the present invention; interpreting them as any additional limitation is contrary to the spirit of the present invention.

Claims

1. A method for predicting and warning of cyanobacteria blooms in large reservoirs, characterized in that: It includes the following steps: S1. Obtain the reservoir water replacement period D at the prediction moment, and assign a value to the water replacement period influence factor F(D) according to the reservoir water replacement period D; S2. Obtain the temperature T at the prediction moment, and assign a value to the temperature influence factor F(T) according to the temperature T; S3. Obtain the rainfall R at the prediction moment, and assign a value to the rainfall influence factor F(R) according to the rainfall R; S4. Obtain the wind speed V at the prediction moment, and assign a value to the wind speed influence factor F(V) according to the wind speed V; S5. Calculate the chlorophyll a concentration C at the prediction moment through the LSTM model, and assign a value to the chlorophyll a influence factor F(C) according to the chlorophyll a concentration C; S6. Calculate the probability F of the occurrence of cyanobacteria blooms in the reservoir according to each influence factor: F = F(D) × F(C) × F(T) × 0.5[F(R) + F(V)] S7. Obtain the corresponding early warning level according to the probability F of the occurrence of cyanobacteria blooms in the reservoir.

2. A method for predicting and warning of cyanobacteria blooms in large reservoirs according to claim 1, characterized in that: The hyperparameter optimization process of the LSTM model is as follows: A1. Set the hyperparameter group of the LSTM model as the optimization object, and randomly initialize the values of m hyperparameter groups according to the preset range; the value of m is set manually; A2. Select the i-th hyperparameter group to construct the LSTM model, and then use the data in the training set to train the LSTM model; 1 ≤ i ≤ m; A3. After the training is completed, calculate the prediction error of the model on the validation set as the current fitness value of the i-th hyperparameter group; A4. Optimize the hyperparameter group through the following formula: V i (k+1)=ωV i (k)+c1r1(P i (k)-X i (k))+c2r2(P g (k)-X i (k)) X i (k+1)=X i (k)+V i (k+1) V i (k) is the change of the i-th hyperparameter group at the k-th iteration; X i (k) is the value of the i-th hyperparameter group at the k-th iteration; P i (k) is the best value found so far for the i-th hyperparameter group; P g (k) is the best value found so far for all groups of hyperparameters; the best value is the value corresponding to the smallest prediction error; ω is the inertia weight; c1 and c2 are acceleration factors; r1 and r2 are random numbers between 0 and 1; A5. Judge whether the end optimization condition is met. If it is met, the optimization process ends; if not, go to step A6; A6. If i < m, increase i by 1 and jump to step A2; If i = m, set i to 1 and jump to step A2; After the optimization is completed, the global optimal value P is used g (k) Build an LSTM model using the final hyperparameter group, and then train the LSTM model using the training data.

3. A method for predicting and warning of cyanobacteria blooms in large reservoirs according to claim 2, characterized in that: The number of neurons in the input layer of the LSTM model is 4, the number of hidden layers is 1, the number of neurons in the output layer is 1, and the required prediction step is 40; the hyperparameters in the hyperparameter group include the sample batch size, the number of hidden layer units, the learning rate, and the number of iteration values. The input parameters of the LSTM model are the temperature, wind speed, cross-sectional flow velocity, and rainfall at the prediction moment.

4. A method for predicting and grading blue algae blooms in large reservoirs according to claim 1 or 2, characterized in that: The method for assigning a value to the chlorophyll a influence factor F(C) is as follows: If the chlorophyll a concentration C is greater than or equal to 60 mg / L, then F(C) = 1; if the chlorophyll a concentration C is greater than 10 mg / L and less than 60 mg / L, then F(C) = 0.012C + 0.2; if the chlorophyll a concentration C is less than 10 mg / L, then F(C) = 0.

2.

5. The method for predicting and grading blue algae blooms in large reservoirs according to claim 1, characterized in that: In step S1, the reservoir water replacement period is obtained through the following formula: D=aH / Q 入_now In the formula, a is the correction coefficient, H is the reservoir capacity, Q 入_now is the current reservoir inflow; the correction factor is obtained by the following formula: a=Q 入_总 / =Q 出_总 ; In the formula, Q 入_总 is the total inflow of the reservoir from day 1 to day n, Q 出_总 is the total outflow of the reservoir from the 1st day to the nth day, where n is the number of days for statistics, and n≥10; The method for assigning a value to the water replacement period influence factor F(D) is as follows: If the water replacement period is greater than 26 days, F(D) = 1.0; if the water replacement period is greater than 13 days and less than or equal to 26 days, then F(D) = 0.8; if the water replacement period is greater than 9 days and less than or equal to 13 days, then F(D) = 0.6; if the water replacement period is greater than or equal to 6 days and less than or equal to 9 days, then F(D) = 0.4; if the water replacement period is less than 6 days, then F(D) = 0.

2.

6. A method for predicting and warning of cyanobacteria blooms in large reservoirs according to claim 1, characterized in that: The temperature T at the prediction time is obtained according to the weather forecast. The temperature influence factor F(T) is assigned as follows: if the temperature T is less than or equal to 25 degrees Celsius, then F(T)=0.4; if the temperature T is greater than 25 degrees Celsius and less than 33 degrees Celsius, then F(T)=0.075T-1.475; if the temperature T is greater than 33 degrees Celsius, then F(T)=1; The rainfall R at the predicted time is obtained according to the weather forecast, and the assignment method of the rainfall influencing factor F(R) is as follows: if the rainfall R is less than or equal to 10 mm, then F(R)=1; if the rainfall R is greater than 10 mm and less than 50 mm, then F(R)=-0.015R+1.15; if the rainfall R is greater than or equal to 50 mm and less than or equal to 250 mm, then F(R)=0.4; if the rainfall R is greater than 250 mm, then F(R)=0.

2.

7. A method for predicting and warning of cyanobacteria blooms in large reservoirs according to claim 1, characterized in that: The wind speed V at the predicted time is obtained according to the weather forecast. The wind speed influence factor F(V) is assigned as follows: if the wind speed V is less than or equal to 0.5 m / s, F(V)=1; if the wind speed V is greater than 0.5 m / s and less than 8 m / s, F(V)=-0.1067V+1.053; if the wind speed V is greater than or equal to 8 m / s, F(V)=0.

2.

8. The method for predicting and warning of cyanobacteria blooms in large reservoirs according to claim 1, characterized in that: The grading method for warning levels of blue algae bloom outbreaks is as follows: ① Low risk: risk probability F ≤ 20%; ② Medium risk: risk probability 20% < F ≤ 40%; ③Higher risk: risk probability 40% < F ≤ 60%; ④High risk: risk probability F>60%.

9. A method for predicting and warning of cyanobacteria blooms in large reservoirs according to claim 1, characterized in that: The training data is obtained in the following manner: five data items of temperature, wind speed, cross-sectional flow velocity, rainfall and chlorophyll a concentration are observed by automatic monitoring stations; the time interval of the observation data is 6-24 hours; the observed data are normalized; the normalized data are divided into a training set, a validation set and a test set according to a ratio of 3:1:1, the data span of the training set is not less than 18 months, the data span of the validation set is not less than 6 months, and the data span of the test set is not less than 6 months; if the original data is missing, linear interpolation is used to fill it in.

10. A method for predicting and warning of cyanobacteria blooms in large reservoirs according to claim 5, characterized in that: In step S1, after obtaining the reservoir water replacement cycle D, the reservoir water replacement cycle is corrected in the following manner: D'=D+ΔD ΔD is the water change cycle correction value, which is obtained by the following formula: ΔD=k1*(R-R avg )+k2*(T-T avg )+k3*(R prev_-1 -R prev_avg )+k4*(R prev_-2 -R prev_avg ) In the formula, k1 is the rainfall influence coefficient during prediction, R is the rainfall at the prediction time, and R avg is the average rainfall, k2 is the temperature influence coefficient during prediction, T is the temperature at the time of prediction, T avg is the average temperature, k3 is the rainfall influence coefficient of the previous time step, R prev_-1 is the rainfall in the previous time step, R prev_avg is the average rainfall in the previous period, k4 is the rainfall influence coefficient in the first two time steps, R prev_-1 is the rainfall in the first two time steps; Then use D' to replace D to calculate the water change cycle influencing factor F(D).

Images