WO2022257243A1 - Water quality early-warning method and system - Google Patents

Abstract

A water quality early-warning method and system. The method comprises: acquiring current biological motion feature data of a water area to be tested (101); and inputting the current biological motion feature data of said water area into a water quality early-warning model, so as to obtain an early-warning result (102), wherein the water quality early-warning model is obtained by means of training a support vector machine model by using a differential evolution algorithm and a gray wolf optimization algorithm. By means of the method and the system, a support vector machine model is trained by means of a differential evolution algorithm and a gray wolf optimization algorithm, so as to obtain a water quality early-warning model, and water quality early-warning is performed, by using the water quality early-warning model, on a water area to be tested, such that the accuracy of water quality early-warning can be improved. The method and the system have the advantage of a high global search capability.

Description

A water quality early warning method and system

This application claims the priority of the Chinese patent application submitted to the China Patent Office on June 07, 2021, the application number is 202110629371.6, and the invention title is "A Water Quality Early Warning Method and System", the entire content of which is incorporated by reference in this application .

technical field

The invention belongs to the field of water quality early warning, and in particular relates to a water quality early warning method and system.

Background technique

Water quality is the abbreviation of water quality, which is generally measured by the physical (such as color, turbidity, odor, etc.), chemical (inorganic and organic content) and biological (bacteria, microorganisms, plankton, benthic organisms) of the water body. . At present, with the development of science and technology and the improvement of human living standards, human beings pay more and more attention to water quality. Water quality early warning has become a hot topic for researchers.

At present, water quality early warning has achieved certain results. According to the characteristic parameters of biological (mainly fish) trajectory, parameter weighted summation or machine learning algorithm is used for biological water quality early warning. However, the method of parameter weighted summation requires a large number of experiments to obtain the optimal weight value; and most machine learning methods are based on the SVM (Support Vector Machine, Support Vector Machine) model for binary classification of biological feature data to obtain early warning results. However, although the current SVM model and the improved SVM model have good local search capabilities, they have problems of poor stability and low accuracy in global search.

Therefore, there is an urgent need for a water quality early warning technology, which has the advantages of high accuracy and strong global search ability.

Contents of the invention

The object of the present invention is to provide a water quality early warning method and system, which can improve the accuracy of water quality early warning and has the advantage of strong global search capability.

To achieve said object, the present invention provides following scheme:

A water quality early warning method, comprising:

Obtain the current biological movement characteristic data of the water area to be tested;

The current biological movement characteristic data of the water area to be measured is input into the water quality early warning model to obtain the early warning result of the water area to be measured; the water quality early warning model uses the differential evolution algorithm and the gray wolf optimization algorithm to train the support vector machine model owned.

Optionally, before inputting the current biological movement characteristic data of the water area to be tested into the water quality early warning model to obtain the early warning result of the water area to be tested, it also includes:

Build a support vector machine model;

Obtain the historical data of biological movement characteristics in the water area to be tested;

The historical data of biological movement characteristics in the water area to be tested is used as a training set, and the support vector machine model is trained by using the differential evolution algorithm and the gray wolf optimization algorithm to obtain a water quality early warning model.

Optionally, the historical data of biological movement characteristics of the water area to be tested is used as a training set, and the support vector machine model is trained using a differential evolution algorithm and a gray wolf optimization algorithm to obtain a water quality early warning model, which specifically includes:

Build an initial pack of wolves;

Utilize differential evolution algorithm to process described initial wolf pack, obtain optimized initial wolf pack, use optimized initial wolf pack as training wolf pack and initialize optimal wolf pack objective function value;

Both the training wolves and the training set are input into the support vector machine model, and the individual objective function value of each individual in the training wolves is calculated;

According to the individual objective function value, the individuals in the training wolf group are sorted in descending order, and the first three individuals are respectively regarded as α wolf, β wolf and δ wolf;

According to the individual objective function value of the alpha wolf, the individual objective function value of the beta wolf and the individual objective function value of the delta wolf, update the optimal wolf group objective function value;

Update the position of each individual in the training wolves, increase the value of the first iteration number by 1, and return to the step "input the training wolves and the training set into the support vector machine model, calculate the Train the individual objective function value of each individual in the wolf pack" until the first iteration number reaches the first iteration number threshold, and use the position of the individual corresponding to the optimal wolf pack objective function value as the support vector The parameters of the machine model are obtained to obtain the water quality early warning model; the parameters of the support vector machine model include: penalty factors and kernel function coefficients.

Optionally, the process of using the differential evolution algorithm to process the initial wolf group to obtain the optimal initial wolf group specifically includes:

Construct parent generation population according to described initial wolf group;

Both the parent population and the training set are input into the support vector machine model, and the population objective function value of the parent population is calculated;

Using a differential evolution algorithm, performing crossover and mutation processing on the parent population to obtain the offspring population;

Both the offspring population and the training set are input into the support vector machine model, and the population objective function value of the offspring population is calculated;

judging whether the population objective function value of the offspring population is greater than the population objective function value of the parent population to obtain a first judgment result;

If the first judgment result is yes, updating the parent population to the child population;

If the first judgment result is no, then retain the parent population;

Increase the value of the second iteration number by 1, and return to the step "Use the differential evolution algorithm to perform crossover and mutation processing on the parent population to obtain the offspring population" until the second iteration number reaches the second iteration number threshold, And determine the parent population as the optimized initial wolf population.

Optionally, updating the optimal wolf group objective function value according to the individual objective function value of the alpha wolf, the individual objective function value of the beta wolf, and the individual objective function value of the delta wolf, specifically includes :

judging whether the optimal wolf pack objective function value is less than the individual objective function value of α wolf, and obtaining a second judgment result;

If the second judgment result is yes, then update the optimal wolf group objective function value to the individual objective function value of α wolf;

If the second judgment result is no, then judge whether the optimal wolf group objective function value is less than the individual objective function value of the β wolf, and obtain the third judgment result;

If the third judgment result is yes, then update the optimal wolf group objective function value to the individual objective function value of β wolf;

If the third judgment result is no, the optimal wolf group objective function value is updated to the individual objective function value of δ wolf.

Optionally, the updating the position of each individual in the training wolf group specifically includes:

According to the position of each individual in the training wolves, use the formula X _p (t+1)=X _p (t)-AD to update each individual in the training wolves to obtain the One update position of each individual; and one update position of each individual except α wolf in the training wolf group as the position of each individual other than α wolf during the next iteration;

According to the updated position of α wolf, the updated position of β wolf and the updated position of δ wolf, using the formula

The alpha wolf is updated again to obtain the second update position of the alpha wolf; and the second update position of the alpha wolf is used as the position of the alpha wolf in the next iteration;

In the formula, X _p (t+1) is the position of individual p at the t+1th iteration; X _p (t) is the position of individual p at the tth iteration; A is the convergence factor; D is the distance from the individual to the prey Distance, D=|CX _p (t)-X(t)|, X(t) is the prey position at the tth iteration, C=2r ₁ , C is the coefficient constant, r ₁ is the distance between [0,1] Random number; X _α (t+1) is the position of α-wolf at the t+1th iteration; X ₁ , X ₂ and X ₃ are the updated positions of α-wolf, β-wolf and δ-wolf at the t-th iteration respectively .

A water quality early warning system comprising:

The current data acquisition module is used to acquire the current biological movement characteristic data of the water area to be measured;

The water quality early warning module is used to input the current biological movement characteristic data of the water area to be measured into the water quality early warning model to obtain the early warning result of the water area to be measured; the water quality early warning model uses differential evolution algorithm and gray wolf optimization algorithm to The support vector machine model is trained.

Optionally, the system also includes:

A support vector machine model building module, used to set up a support vector machine model;

The historical data acquisition module is used to acquire the historical data of biological movement characteristics in the water area to be measured;

The water quality early warning model determination module is used to use the historical data of biological movement characteristics of the water area to be tested as a training set, and use the differential evolution algorithm and the gray wolf optimization algorithm to train the support vector machine model to obtain a water quality early warning model.

Optionally, the water quality early warning model determination module specifically includes:

The initial wolf pack construction unit is used to build the initial wolf pack;

The optimal initial wolf pack determination unit is used to process the initial wolf pack using a differential evolution algorithm to obtain an optimized initial wolf pack, use the optimized initial wolf pack as a training wolf pack and initialize the optimal wolf pack objective function value;

An individual objective function value calculation unit, used to input the training wolves and the training set into the support vector machine model, and calculate the individual objective function value of each individual in the training wolves;

A sorting unit, configured to sort the individuals in the training wolf group in descending order according to the individual objective function value, and use the first three individuals as α wolves, β wolves, and δ wolves respectively;

An optimal wolf pack objective function value updating unit, configured to update the optimal wolf pack according to the individual objective function values of the alpha wolves, the individual objective function values of the beta wolves, and the individual objective function values of the delta wolves objective function value;

The water quality early warning model determination unit is used to update the position of each individual in the training wolves, increase the value of the first iteration number by 1, and return to the step "input the training wolves and the training set into the Support vector machine model, calculate the individual objective function value " of each individual in the described training wolf group ", until the first iteration number reaches the first iteration number threshold value, and will correspond to the optimal wolf group objective function value The location of the individual is used as a parameter of the support vector machine model to obtain a water quality early warning model; the parameters of the support vector machine model include: a penalty factor and a kernel function coefficient.

Optionally, the optimal initial wolf group determination unit specifically includes:

The parent population constructs a subunit, which is used to construct the parent population according to the initial wolf group;

The first population objective function value calculation subunit is used to input both the parent population and the training set into the support vector machine model, and calculate the population objective function value of the parent population;

The offspring population constructs a subunit, which is used to perform crossover and mutation processing on the parent population by using a differential evolution algorithm to obtain the offspring population;

The second population objective function value calculation subunit is used to input both the offspring population and the training set into the support vector machine model, and calculate the population objective function value of the offspring population;

The first judging subunit is used to judge whether the population objective function value of the child population is greater than the population objective function value of the parent population, and obtain the first judgment result; if the first judgment result is yes, then execute the first The parent population update subunit; if the first judgment result is no, execute the second parent generation population update subunit;

The first parent population update subunit is configured to update the parent population to the child population;

The second parent population update subunit is used to retain the parent population;

The optimal initial wolf group determines the child unit, which is used to increase the value of the second iteration number by 1, and returns to the step "use the differential evolution algorithm to perform crossover and mutation processing on the parent population to obtain the offspring population" until the above The second iteration number reaches the second iteration number threshold, and the parent population is determined as the optimized initial wolf group.

The beneficial effects of the present invention are:

The present invention provides a water quality early warning method and system. The method includes: obtaining the current biological movement characteristic data of the water area to be measured; inputting the current biological movement characteristic data of the water area to be measured into the water quality early warning model to obtain the early warning result; the water quality early warning The model is obtained by using the differential evolution algorithm and the gray wolf optimization algorithm to train the support vector machine model. The present invention trains the support vector machine model to obtain the water quality early warning model through the differential evolution algorithm and the gray wolf optimization algorithm, and uses the water quality early warning model to carry out water quality early warning in the water area to be measured, which can improve the accuracy of the water quality early warning and has the advantage of strong global search ability.

Instructions attached

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some of the present invention. Embodiments, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without paying creative labor.

Fig. 1 is the flow chart of the water quality early warning method provided by the embodiment of the present invention;

Fig. 2 is the flow chart of the water quality early warning model training method provided by the embodiment of the present invention;

Fig. 3 is the gradation diagram of wolves provided by the embodiment of the present invention;

Fig. 4 is a schematic diagram of the hunting process of wolves provided by the embodiment of the present invention;

Fig. 5 is the GA-SVM model fitness graph that the embodiment of the present invention provides;

Fig. 6 is the GWO-SVM model fitness curve chart provided by the embodiment of the present invention;

Fig. 7 is a DEGWO-SVM model fitness curve diagram provided by the embodiment of the present invention;

Fig. 8 is a schematic structural diagram of the water quality early warning system provided by the embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

The object of the present invention is to provide a water quality early warning method and system, which can improve the accuracy of water quality early warning and has the advantage of strong global search capability.

In order to make the purpose, features and advantages of the present invention more comprehensible, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

Fig. 1 is the flow chart of the water quality early warning method provided by the embodiment of the present invention, as shown in Fig. 1, the present invention provides a kind of water quality early warning method, comprising:

Step 101: Obtain the current biological movement characteristic data of the water area to be tested;

Step 102: Input the current biological movement characteristic data of the water area to be tested into the water quality early warning model to obtain the early warning result of the water area to be tested; the water quality early warning model is obtained by using the differential evolution algorithm and the gray wolf optimization algorithm to train the support vector machine model.

The water quality early warning method provided by the present invention, before step 102, also includes:

Build a support vector machine model;

Obtain the historical data of biological movement characteristics in the water area to be tested;

The historical data of biological movement characteristics in the water area to be tested is used as the training set, and the support vector machine model is trained by using the differential evolution algorithm and the gray wolf optimization algorithm to obtain the water quality early warning model.

Specifically, the historical data of biological movement characteristics in the water area to be tested is used as a training set, and the support vector machine model is trained using the differential evolution algorithm and the gray wolf optimization algorithm to obtain a water quality early warning model, which specifically includes:

Build an initial pack of wolves;

Use the differential evolution algorithm to process the initial wolf group to obtain the optimized initial wolf group, use the optimized initial wolf group as the training wolf group and initialize the optimal wolf group objective function value;

Both the training wolf group and the training set are input into the support vector machine model, and the individual objective function value of each individual in the training wolf group is calculated; the individual objective function value takes the individual position in the training wolf group as the support vector machine model parameter to perform The fitness function obtained by training.

According to the individual objective function value, the individuals in the training wolf pack are sorted in descending order, and the first three individuals are respectively regarded as α wolf, β wolf and δ wolf;

According to the individual objective function value of α wolf, the individual objective function value of β wolf and the individual objective function value of δ wolf, update the optimal wolf group objective function value;

Update the position of each individual in the training wolf group, increase the value of the first iteration by 1, and return to the step "Input the training wolf group and the training set into the support vector machine model, and calculate the individual target of each individual in the training wolf group function value" until the first iteration number reaches the first iteration number threshold, and the position of the individual corresponding to the optimal wolf group objective function value is used as the parameter of the support vector machine model to obtain the water quality early warning model; the parameters of the support vector machine model Including: penalty factor and kernel function coefficient.

Among them, the initial wolf group is processed by using the differential evolution algorithm to obtain the optimized initial wolf group, which specifically includes:

Construct the parent population based on the initial wolf population;

Both the parent population and the training set are input into the support vector machine model, and the population objective function value of the parent population is calculated; the population objective function value is the fitness function obtained by training the training set with the population parameter as the support vector machine model parameter.

Using the differential evolution algorithm, the parent population is crossed and mutated to obtain the offspring population;

Both the offspring population and the training set are input into the support vector machine model, and the population objective function value of the offspring population is calculated;

Judging whether the population objective function value of the offspring population is greater than the population objective function value of the parent population, and obtaining the first judgment result;

If the first judgment result is yes, update the parent population to the child population;

If the first judgment result is no, the parent population is retained;

Increase the value of the second iteration number by 1, and return to the step "use the differential evolution algorithm to perform crossover and mutation processing on the parent population to obtain the offspring population" until the second iteration number reaches the second iteration number threshold, and the parent generation The population is determined as the optimized initial wolf group.

The water quality early warning method provided by the present invention updates the optimal wolf group objective function value according to the individual objective function value of α wolf, the individual objective function value of β wolf and the individual objective function value of δ wolf, specifically including:

Judging whether the optimal wolf group objective function value is less than the individual objective function value of α wolf, and obtaining the second judgment result;

If the second judgment result is yes, update the optimal wolf group objective function value to the individual objective function value of α wolf;

If the second judgment result is no, it is judged whether the optimal wolf group objective function value is less than the individual objective function value of the β wolf, and the third judgment result is obtained;

If the third judgment result is yes, update the optimal wolf group objective function value to the individual objective function value of β wolf;

If the third judgment result is no, update the optimal wolf group objective function value to the individual objective function value of δ wolf.

Update the position of each individual in the training pack, including:

According to the position of each individual in the training wolves, use the formula X _p (t+1)=X _p (t)-AD to update each individual in the training wolves, and get an update of each individual in the training wolves position; and the updated position of each individual in the training wolf group except α wolf as the position of each individual except α wolf in the next iteration;

According to the updated position of α wolf, the updated position of β wolf and the updated position of δ wolf, using the formula

Update the α wolf again to obtain the secondary update position of the α wolf; and use the secondary update position of the α wolf as the position of the α wolf in the next iteration;

In the formula, X _p (t+1) is the position of individual p at the t+1th iteration; X _p (t) is the position of individual p at the tth iteration; A is the convergence factor; D is the distance from the individual to the prey Distance, D=|CX _p (t)-X(t)|, X(t) is the prey position at the tth iteration, C=2r ₁ , C is the coefficient constant, r ₁ is the distance between [0,1] Random number; X _α (t+1) is the position of α-wolf at the t+1th iteration; X ₁ , X ₂ and X ₃ are the updated positions of α-wolf, β-wolf and δ-wolf at the t-th iteration respectively .

Fig. 2 is the flow chart of the water quality early warning model training method provided by the embodiment of the present invention, as shown in Fig. 2, the water quality early warning model training method in the present invention is as follows:

Step (1): Initialize relevant parameters

The motion characteristic behavior parameters of the obtained fish schools were normalized and standardized, and the training set and test set were constructed. The kernel function in the SVM model uses the Gaussian radial basis kernel function

Among them, x _i and x _j are any point in the space, and γ is the width parameter of the kernel function, which controls the radial range of the function. The initial value of the parameters of the SVM model (the penalty factor C, the parameter γ of the kernel function) is the coordinate of the wolf α in the gray wolf optimization algorithm, and the initial fitness function value of each gray wolf is set to an infinite amount (inf). Space dimension D _n , number of wolves N, maximum number of iterations T _max , coefficient constant C, convergence factor A in the algorithm, crossover probability CR, upper and lower search limits, maximum number of iterations T′ _max and scaling factor in the differential evolution algorithm F (its value range is between [0,2]) are all initialized, assuming that the parameter a=2, r ₁ and r ₂ are random numbers between 0 and 1.

Step (2): Use the DE algorithm (differential evolution algorithm) to initialize the gray wolf population

The evolution process of the differential evolution algorithm mainly has three operations, namely mutation, crossover and selection.

Mutation: The mutation operation refers to adding a vector different from the original vector to the original vector. That is, randomly select the position vectors of two different gray wolf individuals in the initial gray wolf population, adjust the ratio of their vectors and synthesize them with the individual vectors to be mutated. The specific formula is as follows:

D _i (T+1)＝X _r1 (T)+F*(X _r2 (T)-X _r3 (T))

Among them: T is the number of iterations, F is the scaling factor, X _n is the population individual before mutation, D _i (T+1) is the population individual after mutation.

Crossover: After the mutation is completed, the crossover operation is performed on the T generation population individual X _i,j (T) and its mutated individual D _i,j (T+1). The mathematical model formula is as follows:

Among them: j is the dimension, i is the i-th individual of this dimension, CR is the crossover probability, r is the random probability between [0,1], and D _n is the space dimension.

Selection: In order to ensure the goodness of the population, a greedy algorithm is used to select individuals to enter the next generation. When the new generation is better than the parent generation (the fitness value of the new generation is lower than the fitness value of the parent generation), the new generation is retained, otherwise the parent generation is retained. The selection formula is as follows:

Xi (T+1) is the parent population of the T+1 iteration, Xi ₍ T) is the parent population of the T iteration, U _{i (} T+1) is the child population of the T iteration , f(U _i (T+1)) is the fitness value of the offspring population of the T-th iteration, and f(X _i (T)) is the fitness value of the parent population of the T-th iteration.

Step (3): Judging whether the maximum number of iterations T _max is reached; if yes, execute step 4; if not, set T=T+1, and execute step (2).

Step (4): Construct the gray wolf social hierarchy model, calculate the fitness function value and select α wolf, β wolf, and δ wolf respectively;

Fig. 3 is a gradation diagram of wolves provided by the embodiment of the present invention. As shown in Fig. 3, α wolf is defined as the first level, which is the head wolf in the wolves and is mainly responsible for making decisions on predation activities. The wolf with the best management ability in the group; the β wolf is defined as the second level, obeying the alpha wolf, and assisting the alpha wolf to make decisions; the delta wolf is defined as the third level, obeying the alpha wolf and the beta wolf; the rest of the wolves Set to ω wolf, this is the lowest level, obeying the wolves of the first three levels. Wolves strictly abide by this hierarchy and their own responsibilities when hunting.

The initial fitness function value of each gray wolf is set to infinite quantity (inf), and the objective function values of the top three individuals are set to X _α , X _β and X _δ respectively.

Step (5): update the wolf pack position, train the SVM model and predict;

Figure 4 is a schematic diagram of the hunting process of wolves provided by the embodiment of the present invention. As shown in Figure 4, α wolves, β wolves, and δ wolves approach and surround prey constantly, and then command candidate wolves ω wolves to capture prey, and the hunting of gray wolves Behaviors include steps such as searching, approaching, surrounding, hunting, and attacking, as follows:

For the optimization problem, the final position of the prey is its global optimal solution, so the distance between the individual gray wolf and the prey is very important, and the distance formula between the two is as follows:

D＝|CX _p (t)-X(t)|

t is the number of iterations, X _p (t) is the position of the wolf individual at the t-th iteration, X(t) is the position of the prey at the t-th iteration, C is the coefficient constant, C=2r ₁ , r ₁ It is a random number between [0,1].

The distances between α-wolves, β-wolves, δ-wolves and their prey are represented by D _α , D _β , and D _δ respectively, and the formulas are as follows:

D _α ＝|C ₁ .X _α (t)-X(t)|

D _β ＝|C ₂ .X _β (t)-X(t)|

D _δ ＝|C ₃ .X _δ (t)-X(t)|

X _α (t), X _β (t) and X _δ (t) are respectively the positions of α wolf, β wolf, and δ wolf in the t-th iteration, then the mathematical model of gray wolf individual position update is as follows:

X(t+1) _＝ Xp(t)-AD

Among them: X _p (t+1) is the position of individual p at the t+1th iteration; X _p (t) is the position of individual p at the tth iteration; t is the number of first iterations, and D is the distance from the individual to the prey. The distance A is the convergence factor, the initial value of A can be obtained by the formula A=2ar ₂ -a, r ₂ is a random number between [0, 1], a=2, and linearly decreases to 0 with the increase of the number of iterations . For the convergence factor A, the larger A is, the stronger the global search ability is, and the weaker the local search ability is; the smaller A is, the weaker the global search ability is, and the stronger the local search ability is. The specific positions of α wolves, β wolves, and δ wolves are shown by the following formula:

X ₁ =X _α (t)-A ₁ .D _α

X ₂ ＝X _β (t)-A ₂ .D _β

X ₃ ＝X _δ (t)-A ₃ .D _δ

A ₁ , A ₂ and A ₃ are the values of the convergence factors of α wolf, β wolf, and δ wolf in iterations respectively. These three values will change with the number of iterations. X _α (t+1) is the t-th +1 optimal solution vector, the formula is:

The iterative optimal solution vector is used as the parameter combination of the SVM prediction model and substituted into the SVM model for training.

Step (6): Recalculate the fitness function, update the positions of α wolf, β wolf, and δ wolf; and judge whether the maximum number of iterations T′ _max is reached; each wolf is determined by X(t+1)=X _p (t )-AD is updated, but the alpha wolf position will be replaced by

In this iteration, the prediction accuracy of the SVM model will be used as a new fitness function value, with the goal of minimizing the error rate, and the difference between the final fitness function value and 100 will be taken as the absolute value. If the fitness function value is less than the fitness function value of α wolf, update the fitness function value of α wolf to the fitness function value, if the fitness function value is between the fitness function value of α wolf and β wolf between, update the fitness function value of β wolf to the fitness function value, if the fitness function value is between the fitness function value of β wolf and δ wolf, then update the fitness function value of δ wolf Update to the fitness function value. Update the position of the wolves, and judge whether the maximum number of iterations T′ _max is reached. If so, the final position of the α wolf is the value of the parameter combination (penalty factor C, kernel function parameter γ) in the support vector machine. If not, T=T+1, parallel step (4).

Step (8): The final result is used as the parameter combination of the SVM prediction model.

The final position of α wolf is taken as the value of the SVM model parameter combination (C, γ), and the test data set is substituted into the trained SVM model for prediction. Because the value of the parameter combination (C, γ) is optimized by the gray wolf optimization algorithm, the coordinates of α wolves in the initial population of gray wolves are random, and this coordinate is used as the value of (C, γ), and the accurate value after SVM training is After the rate is used as the fitness function of each wolf, the coordinates of the α wolf are getting closer and closer to the optimal (C, γ) value, and the final position of the α wolf is the optimal value obtained by the gray wolf optimization algorithm.

The water quality early warning method proposed by the present invention uses the DE algorithm and the gray wolf optimization algorithm to improve the parameter combination value of the SVM model, and improves the predictive performance of the SVM algorithm in biological water quality early warning, which improves the defects of premature maturity and local optimum, and With global search capability. Aiming at the defect of the classic gray wolf algorithm's slow optimization speed in the early stage, the DE differential evolution algorithm is introduced to generate the optimal value of the initial population of the gray wolf optimization algorithm. The initial population is an important factor affecting the convergence speed of continuous iterations. And the quality of the initial population also affects the optimal value of the prediction model. Therefore, directly optimizing the initial population can improve the ability of the gray wolf optimization algorithm to find the optimal value, overcome the limitations of the random generation of the initial population of the gray wolf optimization algorithm, make the gray wolf optimization algorithm have a better search ability, and improve the SVM biological The classification accuracy of the water quality early warning model. In order to verify the effect of the present invention, technicians have done the following work:

(1) Improve accuracy

Carry out a simulation experiment, after tracking the fish body trajectory under normal water quality and abnormal water quality conditions in the early stage, obtain the five movement characteristic behavior parameters of the fish school (speed, acceleration, average swimming speed, average swimming distance and fish dispersion ), mark the data of these 1000 sample points, including 500 sample points with normal water quality and 500 sample points with abnormal water quality. From these 1000 sample points, 800 sample data points are selected as the training set, and the remaining 200 sample data points are used as the test set. The initial gray wolf population N is set to 30, the search space dimension D _n is set to 2, the scale factor range is [0.01,100], the crossover probability CR is 0.2, and the maximum number of iterations T is 1000.

Since the training set and test set were selected randomly, 10 experiments were carried out in the same environment. The experimental results are shown in Table 1. The values of the model parameters γ and C and the test error are recorded in the table.

Table 1 DEGWO-SVM algorithm test accuracy

As can be seen from Table 1, the water quality early warning model obtained by using the DEGWO-SVM algorithm (that is, the present invention uses differential evolution algorithm and gray wolf optimization algorithm to train the support vector machine model to obtain the water quality early warning model) has better classification accuracy , the average classification accuracy rate of the test set is 94.4%, and the generalization performance of the model is good, so the model is suitable for the biological water quality early warning system. In addition, it can be seen that the model parameter γ changes greatly after each operation, which shows that the fixed model parameters are not suitable for all classification situations. The values of the model parameters are also reserved to 6 decimal places, which shows that the value range of the model parameters is wider and more accurate. The introduction of the differential evolution algorithm and the gray wolf optimization algorithm can find the best parameter combination of the model in a larger range. Find the global optimal solution for the model parameters.

In order to verify the accuracy and convergence of the water quality early warning model of the DEGWO-SVM algorithm, two other optimization algorithms of the improved SVM model were used to compare with the improved and optimized SVM model of the invention. Table 2 is the genetic algorithm improved support vector machine model (Genetic Algorithm-Support Vector Machines, GA-SVM), gray wolf optimization algorithm improved support vector machine model (Grey Wolf Optimization-Support Vector Machines, GWO-SVM) and in the present invention After the DEGWO-SVM algorithm model of DEGWO-SVM runs the algorithm 100 times, the average accuracy rate of the prediction results is compared, and the comparison results are shown in Table 2.

Table 2 Average accuracy comparison table

It can be seen from Table 2 that the average accuracy of DEGWO-SVM is higher than the average accuracy of the other two classification algorithms. The experiment uses mean absolute error (MAE), mean square error (MSE), and mean absolute percentage error (MAPE) as the evaluation criteria of the prediction method. The three evaluation criteria are defined as follows:

In the formula, n is the number of output samples, f _i and y _i are the predicted output value and actual value respectively. Run each algorithm 10 times and take the mean value as the prediction result of the model. Table 3 shows the comparison results of the prediction error of the three algorithms: GA-SVM, GWO-SVM and DEGWO-SVM.

Table 3 Prediction result error comparison table

According to Table 3, compared with the other two algorithms, DEGWO-SVM has higher advantages in the three evaluation criteria, indicating that the DEGWO-SVM algorithm has the smallest test error and the highest prediction accuracy.

(2) Improve premature convergence and speed up convergence

The fitness curves of GA-SVM, GWO-SVM, and DEGWO-SVM, that is, the relationship between the average fitness and the number of iterations are shown in Figure 5-7. As the number of iterations increases, the optimal fitness value of DEGWO-SVM is smaller than that of GA-SVM and GWO-SVM algorithms, and the optimal fitness value of GA-SVM algorithm drops sharply at the 50th iteration, and then at the 100th iteration It basically remains stable when iterating. Compared with the other two algorithms, it is premature and easy to fall into local optimum. The GWO-SVM algorithm fluctuates violently as the number of iterations increases, indicating that the algorithm is unstable. However, the DEGWO-SVM algorithm curve shows a downward trend before 200 iterations, after which the change is small and tends to be stable. Through experimental comparison and analysis, the improved algorithm of the present invention has higher classification accuracy, the lowest three error evaluation criteria, relatively stable performance, high global search ability, and great significance for biological water quality early warning.

Fig. 8 is a schematic structural diagram of the water quality early warning system provided by the embodiment of the present invention. As shown in Fig. 8, the present invention provides a water quality early warning system, including:

The current data acquisition module 901 is used to acquire the current biological movement characteristic data of the water area to be measured;

The water quality early warning module 902 is used to input the current biological movement characteristic data of the water area to be measured into the water quality early warning model to obtain the early warning result of the water area to be measured; the water quality early warning model is to use the differential evolution algorithm and the gray wolf optimization algorithm to carry out obtained by training.

The water quality early warning system provided by the present invention also includes:

A support vector machine model building module, used to set up a support vector machine model;

The historical data acquisition module is used to acquire the historical data of biological movement characteristics in the water area to be measured;

The water quality early warning model determination module is used to use the historical data of biological movement characteristics in the water area to be tested as a training set, and use the differential evolution algorithm and the gray wolf optimization algorithm to train the support vector machine model to obtain the water quality early warning model.

Water quality early warning model determination module, including:

The initial wolf pack construction unit is used to build the initial wolf pack;

The optimal initial wolf group determination unit is used to process the initial wolf group using the differential evolution algorithm to obtain an optimized initial wolf group, use the optimized initial wolf group as a training wolf group and initialize the optimal wolf group objective function value;

The individual objective function value calculation unit is used to input the training wolves and the training set into the support vector machine model, and calculates the individual objective function value of each individual in the training wolves;

The sorting unit is used to sort the individuals in the training wolf group in descending order according to the individual objective function value, and use the first three individuals as α wolf, β wolf and δ wolf respectively;

The optimal wolf group objective function value updating unit is used to update the optimal wolf group objective function value according to the individual objective function value of α wolf, the individual objective function value of β wolf and the individual objective function value of δ wolf;

The determination unit of the water quality early warning model is used to update the position of each individual in the training wolf group, increase the value of the first iteration number by 1, and return to the step "input the training wolf group and the training set into the support vector machine model, and calculate the training wolf The individual objective function value of each individual in the group" until the first iteration reaches the first iteration threshold, and the position of the individual corresponding to the optimal wolf pack objective function value is used as a parameter of the support vector machine model to obtain a water quality warning Model; the parameters of the support vector machine model include: penalty factor and kernel function coefficient.

The optimal initial wolf pack determination unit specifically includes:

The parent population constructs a subunit, which is used to construct the parent population according to the initial wolf group;

The first population objective function value calculation subunit is used to input both the parent population and the training set into the support vector machine model to calculate the population objective function value of the parent population;

The offspring population constructs a subunit, which is used to perform crossover and mutation processing on the parent population by using the differential evolution algorithm to obtain the offspring population;

The second population objective function value calculation subunit is used to input both the offspring population and the training set into the support vector machine model to calculate the population objective function value of the offspring population;

The first judging subunit is used to judge whether the population objective function value of the child population is greater than the population objective function value of the parent population to obtain the first judgment result; if the first judgment result is yes, then perform the first parent population update A subunit; if the first judgment result is no, then execute the second parent generation population update subunit;

The first parent population update subunit is used to update the parent population to the child population;

The second parent generation population updates the sub-unit for retaining the parent generation population;

The optimal initial wolf group determines the child unit, which is used to increase the value of the second iteration number by 1, and returns to the step "use the differential evolution algorithm to perform crossover and mutation processing on the parent population to obtain the offspring population" until the second iteration number The second iteration threshold is reached, and the parent population is determined as the optimized initial wolf population.

The optimal wolf pack objective function value updating unit, specifically includes:

The second judging subunit is used to judge whether the optimal wolf group objective function value is less than the individual objective function value of α wolf, and obtain the second judgment result; if the second judgment result is yes, then execute the first optimal wolf group objective function A value update subunit; if the second judgment result is no, execute the third judgment subunit;

The first optimal wolf group objective function value update subunit is used to update the optimal wolf group objective function value to the individual objective function value of α wolf;

The third judging subunit is used to judge whether the optimal wolf group objective function value is less than the individual objective function value of the β wolf, and obtain the third judgment result; if the third judgment result is yes, then execute the second optimal wolf group objective function A value update subunit; if the third judgment result is no, execute the third optimal wolf pack objective function value update subunit;

The second optimal wolf group objective function value update subunit is used to update the optimal wolf group objective function value to the individual objective function value of the β wolf;

The third optimal wolf pack objective function value updating subunit is used to update the optimal wolf pack objective function value to the individual objective function value of δ wolf;

The determination unit of the water quality early warning model includes:

The first position update subunit is used to update each individual in the training wolves by using the formula X _p (t+1)=X _p (t)-AD according to the position of each individual in the training wolves to obtain the training The updated position of each individual in the wolf pack; and the updated position of each individual in the training wolf pack except α wolf as the position of each individual except α wolf in the next iteration;

The second position update subunit is used to use the formula

Update the α wolf again to obtain the secondary update position of the α wolf; and use the secondary update position of the α wolf as the position of the α wolf in the next iteration;

In the formula, X _p (t+1) is the position of individual p at the t+1th iteration; X _p (t) is the position of individual p at the tth iteration; A is the convergence factor; D is the distance from the individual to the prey Distance, D=|CX _p (t)-X(t)|, X(t) is the prey position at the tth iteration, C=2r ₁ , C is the coefficient constant, r ₁ is the distance between [0,1] Random number; X _α (t+1) is the position of α-wolf at the t+1th iteration; X ₁ , X ₂ and X ₃ are the updated positions of α-wolf, β-wolf and δ-wolf at the t-th iteration respectively .

The embodiments of the present invention have been described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific embodiments, which are only illustrative, rather than restrictive, and those of ordinary skill in the art will Under the enlightenment of the present invention, many forms can also be made without departing from the gist of the present invention and the protection scope of the claims, and these all belong to the protection of the present invention.

Claims (10)

1. A water quality early warning method is characterized in that said method comprises:

   Obtain the current biological movement characteristic data of the water area to be tested;

   The current biological movement characteristic data of the water area to be measured is input into the water quality early warning model to obtain the early warning result of the water area to be measured; the water quality early warning model uses the differential evolution algorithm and the gray wolf optimization algorithm to train the support vector machine model owned.
2. The water quality early warning method according to claim 1, wherein, before inputting the current biological movement characteristic data of the water area to be measured into the water quality early warning model to obtain the early warning result of the water area to be measured, it also includes:

   Build a support vector machine model;

   Obtain the historical data of biological movement characteristics in the water area to be tested;

   The historical data of biological movement characteristics in the water area to be tested is used as a training set, and the support vector machine model is trained by using the differential evolution algorithm and the gray wolf optimization algorithm to obtain a water quality early warning model.
3. The water quality early warning method according to claim 2, wherein, the historical data of biological movement characteristics of the water area to be measured is used as a training set, and the support vector machine model is carried out by using a differential evolution algorithm and a gray wolf optimization algorithm. Training to get the water quality early warning model, including:

   Build an initial pack of wolves;

   Utilize differential evolution algorithm to process described initial wolf pack, obtain optimized initial wolf pack, use optimized initial wolf pack as training wolf pack and initialize optimal wolf pack objective function value;

   Both the training wolves and the training set are input into the support vector machine model, and the individual objective function value of each individual in the training wolves is calculated;

   According to the individual objective function value, the individuals in the training wolf group are sorted in descending order, and the first three individuals are respectively regarded as α wolf, β wolf and δ wolf;

   According to the individual objective function value of the alpha wolf, the individual objective function value of the beta wolf and the individual objective function value of the delta wolf, update the optimal wolf group objective function value;

   Update the position of each individual in the training wolves, increase the value of the first iteration number by 1, and return to the step "input the training wolves and the training set into the support vector machine model, calculate the Train the individual objective function value of each individual in the wolf pack" until the first iteration number reaches the first iteration number threshold, and use the position of the individual corresponding to the optimal wolf pack objective function value as the support vector The parameters of the machine model are obtained to obtain the water quality early warning model; the parameters of the support vector machine model include: penalty factors and kernel function coefficients.
4. The water quality early warning method according to claim 3, wherein the initial wolf pack is processed using a differential evolution algorithm to obtain an optimized initial wolf pack, specifically comprising:

   Construct parent generation population according to described initial wolf group;

   Both the parent population and the training set are input into the support vector machine model, and the population objective function value of the parent population is calculated;

   Using a differential evolution algorithm, performing crossover and mutation processing on the parent population to obtain the offspring population;

   Both the offspring population and the training set are input into the support vector machine model, and the population objective function value of the offspring population is calculated;

   judging whether the population objective function value of the offspring population is greater than the population objective function value of the parent population to obtain a first judgment result;

   If the first judgment result is yes, updating the parent population to the child population;

   If the first judgment result is no, then retain the parent population;

   Increase the value of the second iteration number by 1, and return to the step "Use the differential evolution algorithm to perform crossover and mutation processing on the parent population to obtain the offspring population" until the second iteration number reaches the second iteration number threshold, And determine the parent population as the optimized initial wolf population.
5. The water quality early warning method according to claim 3, characterized in that, according to the individual objective function value of the alpha wolf, the individual objective function value of the beta wolf and the individual objective function value of the delta wolf, update the Describe the optimal wolf pack objective function value, including:

   judging whether the optimal wolf pack objective function value is less than the individual objective function value of α wolf, and obtaining a second judgment result;

   If the second judgment result is yes, then update the optimal wolf group objective function value to the individual objective function value of α wolf;

   If the second judgment result is no, then judge whether the optimal wolf group objective function value is less than the individual objective function value of the β wolf, and obtain the third judgment result;

   If the third judgment result is yes, then update the optimal wolf group objective function value to the individual objective function value of β wolf;

   If the third judgment result is no, the optimal wolf group objective function value is updated to the individual objective function value of δ wolf.
6. The water quality early warning method according to claim 3, wherein the updating the position of each individual in the training wolf pack specifically includes:

   According to the position of each individual in the training wolves, use the formula X _p (t+1)=X _p (t)-AD to update each individual in the training wolves to obtain the One update position of each individual; and one update position of each individual except α wolf in the training wolf group as the position of each individual other than α wolf during the next iteration;

   According to the updated position of α wolf, the updated position of β wolf and the updated position of δ wolf, using the formula

   

   The alpha wolf is updated again to obtain the second update position of the alpha wolf; and the second update position of the alpha wolf is used as the position of the alpha wolf in the next iteration;

   In the formula, X _p (t+1) is the position of individual p at the t+1th iteration; X _p (t) is the position of individual p at the tth iteration; A is the convergence factor; D is the distance from the individual to the prey Distance, D=|CX _p (t)-X(t)|, X(t) is the prey position at the tth iteration, C=2r ₁ , C is the coefficient constant, r ₁ is the distance between [0,1] Random number; X _α (t+1) is the position of α-wolf at the t+1th iteration; X ₁ , X ₂ and X ₃ are the updated positions of α-wolf, β-wolf and δ-wolf at the t-th iteration respectively .
7. A water quality early warning system is characterized in that said system includes:

   The current data acquisition module is used to acquire the current biological movement characteristic data of the water area to be measured;

   The water quality early warning module is used to input the current biological movement characteristic data of the water area to be measured into the water quality early warning model to obtain the early warning result of the water area to be measured; the water quality early warning model uses differential evolution algorithm and gray wolf optimization algorithm to The support vector machine model is trained.
8. The water quality early warning system according to claim 7, wherein said system also includes:

   A support vector machine model building module, used to set up a support vector machine model;

   The historical data acquisition module is used to acquire the historical data of biological movement characteristics in the water area to be measured;

   The water quality early warning model determination module is used to use the historical data of biological movement characteristics of the water area to be tested as a training set, and use the differential evolution algorithm and the gray wolf optimization algorithm to train the support vector machine model to obtain a water quality early warning model.
9. The water quality early warning system according to claim 8, wherein the water quality early warning model determination module specifically includes:

   The initial wolf pack construction unit is used to build the initial wolf pack;

   The optimal initial wolf pack determination unit is used to process the initial wolf pack using a differential evolution algorithm to obtain an optimized initial wolf pack, use the optimized initial wolf pack as a training wolf pack and initialize the optimal wolf pack objective function value;

   An individual objective function value calculation unit, configured to input both the training wolves and the training set into the support vector machine model, and calculate the individual objective function value of each individual in the training wolves;

   A sorting unit, configured to sort the individuals in the training wolf group in descending order according to the individual objective function value, and use the first three individuals as α wolves, β wolves, and δ wolves respectively;

   An optimal wolf pack objective function value updating unit, configured to update the optimal wolf pack according to the individual objective function values of the alpha wolves, the individual objective function values of the beta wolves, and the individual objective function values of the delta wolves objective function value;

   The water quality early warning model determination unit is used to update the position of each individual in the training wolves, increase the value of the first iteration number by 1, and return to the step "input the training wolves and the training set into the Support vector machine model, calculate the individual objective function value " of each individual in the described training wolf group ", until the first iteration number reaches the first iteration number threshold, and will be corresponding to the optimal wolf group objective function value The location of the individual is used as a parameter of the support vector machine model to obtain a water quality early warning model; the parameters of the support vector machine model include: a penalty factor and a kernel function coefficient.
10. The water quality early warning system according to claim 9, wherein the optimal initial wolf group determination unit specifically includes:

    The parent population constructs a subunit, which is used to construct the parent population according to the initial wolf group;

    The first population objective function value calculation subunit is used to input both the parent population and the training set into the support vector machine model, and calculate the population objective function value of the parent population;

    The offspring population constructs a subunit, which is used to perform crossover and mutation processing on the parent population by using a differential evolution algorithm to obtain the offspring population;

    The second population objective function value calculation subunit is used to input both the offspring population and the training set into the support vector machine model, and calculate the population objective function value of the offspring population;

    The first judging subunit is used to judge whether the population objective function value of the child population is greater than the population objective function value of the parent population to obtain a first judgment result; if the first judgment result is yes, then execute the first The parent population update subunit; if the first judgment result is no, execute the second parent generation population update subunit;

    The first parent population update subunit is configured to update the parent population to the child population;

    The second parent population update subunit is used to retain the parent population;

    The optimal initial wolf group determines the child unit, which is used to increase the value of the second iteration number by 1, and returns to the step "use the differential evolution algorithm to perform crossover and mutation processing on the parent population to obtain the offspring population" until the The second iteration number reaches the second iteration number threshold, and the parent population is determined as the optimized initial wolf group.

Images