A water quality prediction method based on an improved SSA algorithm optimized SVR model

Abstract

The present application relates to a kind of water quality prediction method based on improved SSA algorithm optimization SVR model.The method is first to the water quality pollutant concentration data pre-processing, then the data set is constructed into vector form input to SVR model in the way of sliding window training, while in the process of training model, improved sparrow search optimization algorithm SA-WSSA is used to optimize the parameters of SVR model, finally the trained model is used to carry out short-term prediction of river water quality.The method is improved for the deficiency of standard sparrow search algorithm, further improves the optimization ability of algorithm, so as to improve the prediction accuracy and generalization ability of SVR model.The effectiveness of the method is verified by specific experiment.

Description

Technical Field

[0001] This invention belongs to the field of optimization algorithms and machine learning technology, specifically a water quality prediction method based on an improved SSA algorithm to optimize the SVR model. Background Technology

[0002] Water pollution remains one of the most prominent environmental problems in society today. Strengthening water quality management is a crucial measure to reduce water pollution, ensure drinking water safety, and protect human health. To better analyze and manage water quality, long-term monitoring and sampling of water quality data are necessary. However, this process is undoubtedly expensive and time-consuming, and easily affected by external factors, leading to inaccurate or missing data, thus limiting the implementation of water resource protection efforts. To address this issue, research into water quality prediction models is essential. Accurate water quality prediction can effectively analyze water quality changes, thereby better addressing water pollution problems.

[0003] Traditional water quality index prediction methods mainly use time series forecasting models, statistical models, and linear models. These methods are simple and easy to implement, but they cannot fully capture the nonlinear characteristics of the data, resulting in low prediction accuracy and poor model generalization ability. Research shows that water quality prediction based on machine learning models can achieve better prediction accuracy. Commonly used machine learning models include Support Vector Regression (SVR), Random Forest Regression (RFR), Artificial Neural Networks (ANN), and Long Short-Term Memory Neural Networks (LSTM). SVR models have high prediction accuracy and good generalization ability, and are widely used in time series forecasting. However, the prediction effect of SVR models largely depends on the selection of model parameters. Therefore, it is necessary to choose appropriate methods to optimize model parameters to improve water quality prediction capabilities. Summary of the Invention

[0004] To address the shortcomings of existing river water quality prediction methods, the main technical problem solved by this invention is to use pollutant concentration as model input to train the SVR water quality prediction model and to improve the SSA algorithm. This is achieved by optimizing the SVR model parameters based on the improved Sparrow Search Algorithm (SA-WSSA), thereby improving the accuracy of water quality prediction and the generalization ability of the model.

[0005] The technical solution adopted by this invention to achieve the above objectives is: a water quality prediction method based on an improved SSA algorithm to optimize the SVR model, comprising the following steps:

[0006] 1) Obtain river pollutant concentration monitoring data and perform preprocessing;

[0007] 2) The preprocessed dataset is constructed into model input vectors using a sliding window approach, and the dataset is divided into training and test sets;

[0008] 3) Establish an SVR model, input the training set into the SVR model for model training, and optimize the parameters of the SVR model by using an improved sparrow search optimization algorithm to obtain the SA-WSSA-SVR prediction model;

[0009] 4) Test the SA-WSSA-SVR prediction model by using the test set;

[0010] 5) Perform pollutant concentration prediction on historical river pollutant concentration monitoring data through the SA-WSSA-SVR model, perform inverse normalization on the prediction results and output the results, so as to obtain the predicted pollutant concentration data within a set future time period;

[0011] 6) Determine the water quality category of the river within a certain future period of time according to the pollutant concentration predicted by the model and the river water quality category evaluation standard.

[0012] The said river pollutant concentration monitoring data include pollutant concentration and monitoring time.

[0013] The said river pollutants include at least one selected from the group consisting of chemical oxygen demand, permanganate index, ammonia nitrogen, total phosphorus and fluoride.

[0014] The said preprocessing comprises data cleaning, which is specifically as follows:

[0015] Detect outliers by using a k-nearest neighbor algorithm, and process outliers by adopting a mode of taking the mean of upper and lower adjacent values;

[0016] Fill missing values by adopting a linear difference method for missing data within a set time period, and perform smoothing processing by adopting a mode of taking the mean of data within a set distance for continuous missing values.

[0017] The step of constructing a preprocessed data set into model input vectors by adopting a sliding window method includes the following steps:

[0018] The preprocessed pollutant concentration data set data=[x1,x2,…,x n , wherein n is the size of the data set, x j represents the pollutant concentration, j is 1...n, corresponding to the monitoring time;

[0019] Construct a model input vector X=[[x1,x2,…,x m , [x2,x3,…,x m+1 , …, [x n-m ,x n-2 ,…,x n-1 , construct a model output vector Y=[x2,x3,…,x n ; wherein m<n, and X and Y are configured to be input into the SVR model for model training.

[0020] The improved sparrow search algorithm uses Tent chaotic mapping to initialize the sparrow population, as follows:

[0021] The Tent chaotic mapping expression is as follows:

[0022]

[0023] In the above formula, i represents the i-th mapping, represents the i-th sparrow, and z i Let represent the value of the i-th mapping function, indicating the initial position of the sparrow population.

[0024] The improved sparrow search algorithm updates the finder's position as follows:

[0025]

[0026] In the above formula, t is the current iteration number, iter max The maximum number of iterations, Let α represent the position information of the i-th sparrow in the d-th dimension at the current iteration number, where α∈[0,1] is a random number, R2∈[0,1] and ST∈[0.5,1] represent the warning value and the safety value respectively, Q is a random number that follows a normal distribution, and L represents a 1*d matrix where each element is 1; It is an adaptive inertia weight, and it increases with the number of iterations.

[0027] The improved sparrow search algorithm uses the following follower position update:

[0028]

[0029]

[0030] In the above formula This indicates the optimal position occupied by the discoverer at the current moment. Let A represent the current worst-case position, where A is a 1*d matrix, and each element is randomly assigned the value 1 or -1. + =A T (AA T ) -1 ; For random inertia weights, N(0,1) represents random numbers from a standard normal distribution, θ represents random numbers between (0,1), and μ min and μ max These represent the minimum and maximum values ​​of the inertia weight, respectively. σ represents the position information of the i-th sparrow in the d-th dimension at the current iteration number, and σ represents the standard deviation.

[0031] The improved sparrow search algorithm uses a fusion simulated annealing algorithm to update the global optimal position of the sparrow population, including the following steps:

[0032] First, a new sparrow population is randomly generated around the original food source location;

[0033] Then calculate the fitness value of individuals in the new population; if the fitness value of sparrows in the new population is less than the fitness value of food sources, update the location of food sources; otherwise, accept the location of sparrows in the new population with probability P.

[0034] Finally, an annealing operation is performed, T = T × 0.99, and the location of the food source with the smallest fitness value and its fitness value are recorded throughout the entire iteration process.

[0035] Where the probability p is as follows:

[0036]

[0037] In the above formula, f new (x i f represents the fitness value of the i-th sparrow in the new population generated during the simulated annealing phase. g (x i ) represents the current optimal fitness value of the population, and T represents the current temperature.

[0038] A water quality prediction system based on an improved SSA algorithm to optimize the SVR model includes:

[0039] The preprocessing module is used to acquire and preprocess river pollutant concentration monitoring data.

[0040] The model building module is used to construct the model input vector from the preprocessed dataset using a sliding window approach, and to divide the dataset into training and test sets; to build the SVR model, and to input the training set into the SVR model for model training, while using an improved sparrow search optimization algorithm to optimize the parameters of the SVR model to obtain the SA-WSSA-SVR prediction model; and to test the SA-WSSA-SVR prediction model using the test set.

[0041] The water quality prediction module is used to predict pollutant concentrations based on the SA-WSSA-SVR model, and then inversely normalizes and outputs the prediction results to obtain the pollutant concentration data for a set future time period. Based on the pollutant concentrations predicted by the model and the river water quality category evaluation criteria, the river water quality category for a certain future time period is determined.

[0042] The present invention has the following beneficial effects and advantages:

[0043] 1. This invention applies the Tent chaotic mapping to the SSA algorithm, adds adaptive inertia weights and random inertia weights, and combines the simulated annealing algorithm with the SSA algorithm to create an improved sparrow search algorithm SA-WSSA.

[0044] 2. This invention applies the improved sparrow search algorithm to optimize SVR model parameters, and establishes an SA-WSSA-SVR water quality prediction model, which effectively improves the model's prediction accuracy and generalization ability.

[0045] 3. This invention provides a practical and effective method for predicting river water quality, which can accurately predict the changing trend of river pollutants in the coming week, providing effective data support for protecting the river water environment. Attached Figure Description

[0046] Figure 1 This is a structural diagram of a water quality prediction method based on an improved SSA algorithm to optimize an SVR model, according to the present invention.

[0047] Figure 2 Fit a prediction plot to the SVR model;

[0048] Figure 3 Fit the prediction plot to the SSA-SVR model.

[0049] Figure 4 Fitted prediction plot for SA-WSSA-SVR model. Detailed Implementation

[0050] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0051] This invention proposes an SVR water quality prediction model based on the improved Sparrow Search Optimization Algorithm (SA-WSSA). The improved Sparrow Search Algorithm can better optimize the SVR model parameters, thereby improving the model's prediction accuracy and generalization ability.

[0052] like Figure 1 As shown, the water quality prediction method based on the improved SSA algorithm to optimize the SVR model includes the following steps:

[0053] Step 1: The automatic monitoring station is equipped with pollutant concentration detection devices, including chemical oxygen demand (COD) sensors, permanganate index sensors, total phosphorus sensors, fluoride sensors, and ammonia nitrogen sensors. The data detected by the detection devices is stored in the monitoring station's database. The river pollutant concentration and the corresponding time are obtained from the database of the automatic river monitoring station.

[0054] Step 2: Perform data cleaning and normalization on the acquired river monitoring data, mainly including handling missing and outlier values ​​and data normalization;

[0055] Step 3: Construct the preprocessed data into input vectors using a sliding window approach, and divide them into training and test sets;

[0056] Step 4: Input the training set into the SVR model for training, and at the same time optimize the SVR model parameters based on the improved Sparrow Search Optimization Algorithm SA-WSSA to obtain the SA-WSSA-SVR water quality prediction model.

[0057] Step 5: Input the test set into the SA-WSSA-SVR water quality prediction model for testing;

[0058] Step 6: Make predictions based on the trained model, and then inversely normalize and output the model prediction results;

[0059] Step 7: Determine the river water quality category for a given time period based on the pollutant concentrations predicted by the model and the river water quality category evaluation criteria. The river water quality category evaluation criteria primarily involve the range of pollutant concentration values. River water quality categories are divided into five levels from best to worst: I, II, III, IV, V, and worse than V. Each pollutant is classified according to its concentration range based on the predicted pollutant concentrations, and the worst category is ultimately selected as the current river water quality category for the monitoring station.

[0060] The data cleaning process specifically involves: using the k-nearest neighbor algorithm to detect outliers and averaging the data above and below to handle outliers; using linear interpolation to fill in missing data within a set time period; and averaging the data within a set distance to smooth out consecutive missing values.

[0061] The data normalization process specifically involves: applying min-max standardization to perform a linear transformation on the sample data, mapping the result values ​​to the range [0,1]; where the sample data is a pollutant concentration dataset; the transformation expression is:

[0062]

[0063] In the above formula, X′ is the normalized data, max is the maximum value of the sample data, min is the minimum value of the sample data, and X is the sample data.

[0064] The river pollutants include one of the following: chemical oxygen demand, permanganate index, ammonia nitrogen, total phosphorus, and fluoride.

[0065] The specific operation of constructing the input vector using the sliding window is as follows: Let the pollutant concentration dataset be data = [x1, x2, ..., x...].n ], where n is the dataset size. The model input vector X = [[x1, x2, ..., xm] is constructed with a sliding window size of m. m ],[x2,x3,…,x m+1 ],…,[x n-m ,x n-2 ,…,x n-1 Construct the model input vector Y = [x2, x3, ..., x...] n ].

[0066] The improved sparrow search optimization algorithm includes the following steps:

[0067] Step 1: Initialize the sparrow population using Tent chaotic mapping;

[0068] Step 2: Calculate the fitness value of each sparrow and sort them from smallest to largest;

[0069] Step 3: Select the top 10%-20% of sparrows with the highest fitness values ​​as discoverers, and the rest as followers;

[0070] Step 4: Update the discoverer's position based on the improved discoverer position update formula;

[0071] Step 5: Update the follower position based on the improved follower position update formula;

[0072] Step 6: Randomly select 10%-20% of the sparrows in the population as watchmen and update their positions using the watchmen position update formula;

[0073] Step 7: Update the global optimal position based on the simulated degradation algorithm;

[0074] Step 8: Determine if the maximum number of iterations has been reached. If it has, output the optimal value and the algorithm ends. If it has not been reached, go back to step 2 to continue iterating.

[0075] The specific improvements to the improved sparrow search algorithm SA-WSSA are as follows:

[0076] 1) The sparrow population is initialized using the Tent chaotic map. The Tent chaotic map expression is as follows:

[0077]

[0078] In the above formula, i represents the number of mappings, and z i This represents the value of the mapping function at the i-th time.

[0079] The steps for initializing a sparrow population based on the above Tent chaotic mapping expression are as follows:

[0080] (1) Randomly generate z0, z1 = z0, i = j = 1 in the interval (0, 1);

[0081] (2) Iterating according to the above formula, we get z i+1 , continue i=i-1;

[0082] (3) If z i ={0,0.25,0.5,0.75}, or z i =z i-k k = {0, 1, 2, 3, 4}, press z i =z j+1 =z j +ε updates the initial value, where ε is a random number, j = j + 1, otherwise return to step 2;

[0083] (4) The program ends when the maximum number of iterations is reached, and the sparrow population initialization sequence is obtained; otherwise, return to step (2).

[0084] 2) Improve the discoverer location update formula. The improved formula is as follows:

[0085]

[0086] In the above formula, t is the current iteration number, iter max The maximum number of iterations, Let α represent the position information of the i-th sparrow in the d-th dimension at the current iteration number, where α∈[0,1] is a random number, R2∈[0,1] and ST∈[0.5,1] represent the warning value and the safety value respectively, Q is a random number that follows a normal distribution, and L represents a 1*d matrix where each element is 1. The inertial weights are adaptive, and ω will automatically decrease as the number of iterations increases, thus enabling an adaptive range of values ​​for α.

[0087] 3) Improve the follower position update formula. The improved formula is as follows:

[0088]

[0089]

[0090] In the above formula This indicates the optimal position occupied by the discoverer at the current moment. Let A represent the current worst-case position, where A is a 1*d matrix, and each element is randomly assigned the value 1 or -1. + =A T (AA T ) -1 . For random inertia weights, N(0,1) represents random numbers from a standard normal distribution, θ represents random numbers between (0,1), and μ min and μ max Let μ and denot represent the minimum and maximum values ​​of the inertia weight, respectively, and when μ... min and μ max The algorithm converges well when the values ​​are 0.4 and 0.9 respectively. σ represents the standard deviation, which is generally taken as a number between 0.2 and 0.5.

[0091] 4) Integrate simulated annealing algorithm to update the global optimal position of the sparrow population. The simulated annealing algorithm calculates the probability of accepting inferior solutions based on the Metropolis criterion. The probability calculation formula is as follows:

[0092]

[0093] In the above formula, f new (x i f represents the fitness value of the i-th sparrow in the new population generated during the simulated annealing phase. g (x i ) represents the current optimal fitness value of the population, and T represents the current temperature.

[0094] The simulated annealing algorithm first randomly generates a new sparrow population around the original food source location; then it calculates the fitness value of the individuals in the new population. If the fitness value of a sparrow in the new population is less than the fitness value of the food source, the food source location is updated; otherwise, the probability P in the above formula is used to accept the location of the sparrow in the new population. Finally, an annealing operation is performed, T = T × 0.99, and the location of the food source with the smallest fitness value and its fitness value are recorded during the entire iteration process.

[0095] Case Study 1:

[0096] Step 1: Data Acquisition: The dataset comes from the Liaohe Zhanghuangdi Automatic Monitoring Station (Liaoning, China) in the Liaohe River Basin. This experiment selected chemical oxygen demand (COD) concentration data collected by this monitoring station from January 1, 2013 to December 31, 2020. The dataset includes monitoring time and corresponding water quality index concentration values. This data has already been sent to the database by the terminal data acquisition device for storage, so it was directly queried and exported from the database.

[0097] Step 2: Data Cleaning and Normalization: The automatic monitoring station collects data every 4 hours. To ensure the accuracy of the model, the collected data is first resampled, and the average value of the data monitored each day is taken as the historical monitoring data for that day. The resampled data is then used to detect outliers using the k-nearest neighbor algorithm. Outliers are processed by taking the upper and lower average values. Finally, the data is normalized.

[0098] Step 3: Constructing the input vector: The preprocessed data is used to construct the model input vector using a sliding window method. The sliding window size is set to 3, which means predicting the value of the 4th day based on the historical data of the previous 3 days. Then, the dataset is divided into training set and test set.

[0099] Step 4: Model Training: Establish the SA-WSSA-SVR model and initialize the parameters. Set the sparrow population size to N=100, the maximum number of iterations to M=500, and the maximum simulated annealing temperature to T=1000. Train the model based on the above training set.

[0100] Step 5: Model testing: Input the test set into the trained SA-WSSA-SVR model to test the model's predictive ability.

[0101] Step 7: Model Prediction: Use the trained model to fit historical data and predict pollutant concentrations for the next week.

[0102] After obtaining the predicted data in step 8, the trend and category of river water quality changes are determined based on the predicted data.

[0103] To fully demonstrate the improvement in the model's prediction accuracy, this experiment uses the SVR model, the SSA-SVR model, and the SA-WSSA-SVR model for training and prediction, respectively, and calculates the mean absolute percentage error (MAPE) and root mean square error (RMSE) of the model's predicted values.

[0104] The above experimental results are as follows Figure 2 , Figure 3 , Figure 4 As shown in Table 1.

[0105] Table 1

[0106]

[0107] As shown in Table 1, the SA-WSSA-SVR model of the present invention has significantly improved the prediction accuracy compared with the SVR model, and its prediction ability is superior to that of the SSA-SVR model.

Claims

1. A water quality prediction method based on an improved SSA algorithm to optimize an SVR model, characterized in that, Includes the following steps: 1) Obtain river pollutant concentration monitoring data and perform preprocessing; 2) The preprocessed dataset is constructed into model input vectors using a sliding window approach, and the dataset is divided into training and test sets; 3) Establish an SVR model and input the training set into the SVR model for model training. At the same time, use an improved sparrow search optimization algorithm to optimize the parameters of the SVR model to obtain the SA-WSSA-SVR prediction model. 4) The SA-WSSA-SVR prediction model was tested using a test set; 5) Historical river pollutant concentration monitoring data are used to predict pollutant concentrations using the SA-WSSA-SVR prediction model. The prediction results are then inversely normalized and output to obtain the predicted pollutant concentration data for a given future time period. 6) Determine the river water quality category at a future time based on the pollutant concentrations predicted by the SA-WSSA-SVR prediction model and the river water quality category evaluation criteria; The improved sparrow search algorithm updates the finder's position as follows: ； In the above formula, t is the current iteration number. The maximum number of iterations, This represents the position information of the i-th sparrow in the d-th dimension at the current iteration number. It is a random number. and Let Q and L represent the warning value and the safety value, respectively. Q is a random number that follows a normal distribution. L represents a 1*d matrix in which every element is 1. It is an adaptive inertia weight, and it increases with the number of iterations.

2. The water quality prediction method based on the improved SSA algorithm to optimize the SVR model according to claim 1, characterized in that, The river pollutant concentration monitoring data includes pollutant concentration and monitoring time.

3. The water quality prediction method based on the improved SSA algorithm to optimize the SVR model according to claim 1, characterized in that, The river pollutants include at least one of the following: chemical oxygen demand, permanganate index, ammonia nitrogen, total phosphorus, and fluoride.

4. The water quality prediction method based on the improved SSA algorithm to optimize the SVR model according to claim 1, characterized in that, The preprocessing includes data cleaning, specifically: The k-nearest neighbor algorithm is used for outlier detection, and the average of the upper and lower values ​​is used for outlier handling. For missing data within a set time period, the linear interpolation method is used to fill in the missing values. For consecutive missing values, the average of the data within a set distance is used for smoothing.

5. A water quality prediction method based on an improved SSA algorithm to optimize an SVR model according to claim 1, characterized in that, The process of constructing model input vectors from the preprocessed dataset using a sliding window method includes the following steps: Preprocessed pollutant concentration dataset data=[ , ,…, ], where n is the dataset size, x j This represents the pollutant concentration, where j is 1...n, corresponding to the monitoring time. The model input vector X is constructed with a sliding window size m as X= , ,…, , , ,…, ,… , , ,…, , and the model input vector Y is constructed as Y= , ,…, ; m<n, and X and Y are used for input into the SVR model for model training.

6. The water quality prediction method based on the improved SSA algorithm to optimize the SVR model according to claim 1, characterized in that, The improved sparrow search algorithm uses Tent chaotic mapping to initialize the sparrow population, as follows: The Tent chaotic mapping expression is as follows: ； In the above formula, i represents the i-th mapping, and represents the i-th sparrow. Let represent the value of the i-th mapping function, indicating the initial position of the sparrow population.

7. A water quality prediction method based on an improved SSA algorithm to optimize an SVR model according to claim 1, characterized in that, The improved sparrow search algorithm uses the following follower position update: ； ； In the above formula This indicates the optimal position occupied by the discoverer at the current moment. Let A represent the current worst-case position, where A is a 1*d matrix, and each element is randomly assigned a value of 1 or -1. ; For random inertia weights, Represents a random number that follows a standard normal distribution. Represents a random number between (0,1). and These represent the minimum and maximum values ​​of the inertia weight, respectively. This represents the position information of the i-th sparrow in the d-th dimension at the current iteration number, where n is the size of the dataset. It represents the standard deviation.

8. A water quality prediction method based on an improved SSA algorithm to optimize an SVR model according to claim 1, characterized in that, The improved sparrow search algorithm uses a fusion simulated annealing algorithm to update the global optimal position of the sparrow population, including the following steps: First, a new sparrow population is randomly generated around the original food source location; Then calculate the fitness value of individuals in the new population; if the fitness value of sparrows in the new population is less than the fitness value of food sources, update the location of food sources; otherwise, accept the location of sparrows in the new population with probability p. Finally, an annealing operation is performed, T = T × 0.99, and the location of the food source with the smallest fitness value and its fitness value are recorded throughout the entire iteration process. Where the probability p is as follows: ； In the above formula This represents the fitness value of the i-th sparrow in the new population generated during the simulated annealing phase. This represents the current optimal fitness value of the population, and T represents the current temperature.

9. A water quality prediction system based on an improved SSA algorithm to optimize an SVR model, the system being used to implement the water quality prediction method based on an improved SSA algorithm to optimize an SVR model as described in any one of claims 1-8, characterized in that, include: The preprocessing module is used to acquire and preprocess river pollutant concentration monitoring data. The model building module is used to construct model input vectors from the preprocessed dataset using a sliding window approach, and to divide the dataset into training and test sets. An SVR model was built, and the training set was input into the SVR model for model training. At the same time, an improved sparrow search optimization algorithm was used to optimize the parameters of the SVR model to obtain the SA-WSSA-SVR prediction model. The SA-WSSA-SVR prediction model was tested using a test set. The water quality prediction module is used to predict pollutant concentrations based on the SA-WSSA-SVR prediction model, and simultaneously inverse normalizes the prediction results and outputs them to obtain the pollutant concentration data for a set future time period. Based on the pollutant concentrations predicted by the SA-WSSA-SVR prediction model and the river water quality category evaluation criteria, the river water quality category for a certain future time period is determined.

Images