CN101625732A - Forecasting method of water level of potamic tidewater - Google Patents

Abstract

The invention relates to a method for predicting the tidal water level of rivers. The traditional tidal time forecasting methods include the propagation time method and the isolation lag time method. The above methods have weakened the influence of tides, runoff, river topography, and wind direction to a certain extent. Concrete steps of the present invention are at first with the neural network function in the toolbox function storehouse of MATLAB 6.5 and the history data of tide water level and time as the foundation to construct neural network model, utilize the neural network training function train function in the toolbox function to carry out network then Training, using the simulation function sim function for network testing, and finally using the trained and tested neural network model to predict the next high tide level or the next low tide level. The present invention uses historical tidal data to predict short-term tidal water level values, and this method can completely ignore the influence of uncertain factors such as wind direction, rainfall, water supply and drainage, and river bed changes.

Description

Prediction method of tidal water level in rivers

technical field

The invention belongs to the technical field of automation and relates to a method for predicting the tidal water level of rivers.

Background technique

The tides of some rivers and rivers are turbulent, the sound of the water is roaring and majestic, and the tide is steep, attracting hundreds of thousands of tourists to stop on both sides of the river to watch the tide every year. On the other hand, the water potential of these river tides and the turbulence of the underwater undercurrent also pose a great threat to the lives and property of the people and tourists living on both sides of the river. According to statistics, people die here every year because they don't understand the water potential of the river tide. It can be seen that it is very meaningful to analyze and predict the change law of the tidal water level and the coming time of the spring tide of the river to protect the safety of life and property of tourists and people on both sides of the strait.

The traditional tide time forecasting methods include the propagation time method and the isolation lag time method, while the tide water level prediction method is to predict the tide level height of the next day based on the tide level height of the previous day, or predict the height of the next station based on the height of the previous hydrological station. water level. The above-mentioned empirical method has to some extent weakened the relationship between tides and multiple factors such as tide, runoff, river topography, and wind direction, and weakened a certain high-dimensional nonlinear mapping relationship between them. The present invention tries to use the BP artificial neural network theory developed in recent years, and utilizes its high fault tolerance, parallel processing data and ability to approach any nonlinear function to establish a model for predicting tide level and tide time, in order to make an approximation Predictive performance of the true value.

Contents of the invention

The purpose of the present invention aims at the deficiencies of the prior art, and provides a method for predicting the tide level of rivers. The method of the present invention uses historical data to automatically identify, monitor and track the level of the tide, and realizes the forecast.

Concrete steps of the present invention are:

Step (1). Construct a neural network model based on the neural network function in the toolbox function library of MATLAB 6.5 and the historical data of tide level and time. The specific method is:

① Obtain N groups of continuous historical data of tide level and time. Tide water level and time data There are four sets of data in a normal day, which are the time and water level value of the high tide level of the daily tide, the time and water level value of the low tide level of the daily tide, the time and water level value of the high tide level of the night tide, and the low tide level of the night tide. time and water level values. Divide these N groups of data into two groups, the front data group is the training sample data group N _i group, and the latter data group is the test sample data group N _j group, where N=N _i +N _j , N _i /N _j = 2-4.

② Determine that the training model of the neural network has three layers, the input layer has 4 nodes, the middle hidden layer has 50 nodes, and the output layer has 1 node. Use the newff function to create a BP neural network function, where the network parameters are set as follows: select the tansig function (hyperbolic tangent S-shaped transfer function) for the neuron transfer function, select the purelin function (linear transfer function) for the output layer transfer function, and select traindx for the network training method function, the number of training steps is 5000 times, and the training accuracy is 0.1.

Step (2). Utilize the neural network training function train function in the MATLAB 6.5 toolbox function to carry out network training, then utilize the simulation function sim function in the MATLAB 6.5 toolbox function to carry out network testing. The specific method is:

③The method of using the neural network training function train function in the MATLAB 6.5 toolbox function for network training is: convert the training sample data group into a 4×(N _i -3) matrix as the input parameter of the train function, and run the train function to perform Training, the training process will stop when the accuracy requirement is reached or the number of training steps is reached. After the training, a neural network model will be automatically generated by the system, and the weight coefficients are hidden, which can be viewed in the MATLAB 6.5 system.

The specific method of converting the training sample data set N _i into a matrix of 4×(N _i −3) is: the first data to the fourth data of the training sample data set N _i form the first column, and the training sample data set N i The second data to the fifth data form the second column of the matrix, the third data to the sixth data of the training sample data set form the third column of the matrix, and so on until the composition of 4×(N _i -3) matrix.

④The method of using the simulation function sim function in the MATLAB 6.5 toolbox function for network testing is: convert the test sample data group into a 4×(N _j -3) matrix as the input parameter of the sim function for network testing.

The specific method for converting the test sample data set into a 4×(N _j -3) matrix is: the first data to the fourth data of the test sample data set form the first column, and the second data of the test sample data set The second column of the matrix is composed of the fifth data, and the third column of the matrix is composed of the third data to the sixth data of the test sample data group, and so on until a 4×(N _j −3) matrix is formed.

Step (3). Use the sim function of the neural network model and the MATLAB 6.5 toolbox function to predict the next high tide level water level value or the next low tide level water level value through network training and network testing. The specific method is: input the last four sets of data in the sim function as the input parameters of the sim function, and get the predicted value of the next set of data after calculation by the sim function, and this predicted value is the next high tide level or the next low tide Water level value.

The present invention adopts the historical data of river tidal water to train the neural network model according to the rules to predict, and its biggest advantage is that it can predict the tidal water level value in a relatively accurate short-term only by using the historical tidal data. Compared with the current traditional forecasting method, this method can completely ignore the influence of uncertain factors such as wind direction, rainfall, water supply and drainage, and river bed changes, so as to realize the prediction of tidal water level only relying on historical data. Neglecting the influence of other tidal water level factors and only relying on historical data to predict tidal water level at low cost is the main purpose and greatest advantage of the present invention.

Detailed ways

For the tidal water level prediction of a hydrological station in the Qiantang River, the specific implementation steps are as follows:

Step (1). Construct a neural network model based on the neural network function in the toolbox function library of MATLAB 6.5 and the historical data of the Qiantang River tide water level and time. The specific method is:

① Obtain the hydrological data of 1410 groups of tidal water level in 1990 at a hydrological station on the Qiantang River. Tide water level and time data There are four sets of data in a normal day, which are the time and water level value of the high tide level of the daily tide, the time and water level value of the low tide level of the daily tide, the time and water level value of the high tide level of the night tide, and the low tide level of the night tide. time and water level values. Divide these N groups of data into two groups before and after, the front data group is 1000 groups of training sample data groups, and the latter data group is 410 groups of test sample data groups;

② Determine that the training model of the neural network has three layers, the input layer has 4 nodes, the middle hidden layer has 50 nodes, and the output layer has 1 node. Use the newff function to create a BP neural network function, where the network parameters are set as follows: select the tansig function (hyperbolic tangent S-shaped transfer function) for the neuron transfer function, select the purelin function (linear transfer function) for the output layer transfer function, and select traindx for the network training method function, the number of training steps is 5000 times, and the training accuracy is 0.1. details as follows:

net = newff(minmax(p), [50 1], {'tansig', 'purelin'}, 'trainingdx', 'learndm', 'mse');

Among them, the network constructed by the above parameters is represented by the network name net.

The number of training termination steps is 5000, and the training error precision is 0.1. The specific settings are as follows:

net.trainParam.epochs = 5000;

net.trainParam.goal = 0.1;

Step (2). Utilize the neural network training function train function in the MATLAB 6.5 toolbox function to carry out network training, then utilize the simulation function sim function in the MATLAB 6.5 toolbox function to carry out network testing. The specific method is:

③The method of using the neural network training function train function in the MATLAB 6.5 toolbox function for network training is: convert the training sample data group into a 4×997 matrix as the input parameter of the train function, and run the train function for training, specifically: net=train(net, p, t); wherein, net in the train function parameter table represents the original network constructed, p represents the matrix of the input training sample group, and t represents the real sample used for reverse correction of weights. The training process will stop when the accuracy requirement is reached or the number of training steps is reached. After the training, the system will automatically generate a neural network model, each weight coefficient is implicit, and can be viewed in the MATLAB 6.5 system.

④The method of using the simulation function sim function in the MATLAB 6.5 toolbox function for network testing is: transform the test sample data set N _j into a 4×407 matrix as the input parameter of the sim function for network testing. Specifically: a=sim(net, p); wherein, net is the trained network, p is the matrix of the test sample group, and a is the simulated value output by the network.

Step (3). Use the sim function of the neural network model and the MATLAB 6.5 toolbox function to predict the next high tide level water level value or the next low tide level water level value through network training and network testing. The specific method is: input the last four sets of data in the sim function as the input parameters of the sim function, and get the predicted value of the next set of data after calculation by the sim function, and this predicted value is the next high tide level or the next low tide Water level value.

Adopting the tidal prediction method using the neural network of the present invention, by using 1410 groups of data experiments at the Cangqian Hydrological Station in 1990, it shows that the accuracy of water level prediction can reach 0.1 meters, and its relative error rate is controlled below 5%.

Claims (1)

1. the Forecasting Methodology of water level of potamic tidewater is characterized in that this method comprises the steps:

Step (1) is the fundamental construction neural network model with neural network function in the tool box function library of MATLAB 6.5 and the historical data of tidal level and time, and concrete grammar is:

1. obtain the continuous historical data N group of tidal level and time; Tidal level and time data normal condition had four groups of data in one day, were respectively the time and the water level value of time of time of day time of tidewater high tide level and water level value, day tidewater low tide and water level value, Evening Tide water high tide level and water level value, Evening Tide water low tide; These N group data are divided into two groups of front and back, and the data set of front is training sample data group N _iGroup, the data set of back is test sample book data set N _jGroup, wherein N=N _i+ N _j, N _i/ N _j=2～4;

2. the training pattern of determining neural network is three layers, and input layer is 4 nodes, and middle hidden layer is 50 nodes, and output layer is 1 node; With newff function creation BP neural network function, wherein network parameter is set to: the neuron transition function is selected the tansig function for use, and the output layer transition function is selected the purelin function for use, and the network training method is selected the traingdx function for use, training step number 5000 times, training precision is 0.1;

Step (2) utilizes the neural metwork training function train function in the MATLAB 6.5 tool box functions to carry out network training, utilizes the emulation function sim function in the MATLAB 6.5 tool box functions to carry out network test then, and concrete grammar is:

3. the method for utilizing neural metwork training function train function in the MATLAB 6.5 tool box functions to carry out network training is: training sample data group is converted into 4 * (N _i-3) matrix is as the input parameter of train function, and operation train function is trained, and training process will be a stop condition to reach accuracy requirement or to reach the training step number; To generate a neural network model automatically by system after training finishes, each weights coefficient is implicit, can check in MATLAB 6.5 systems;

Training sample data group N _iBe converted into 4 * (N _iThe concrete grammar of matrix-3) is: with training sample data group N _iFirst data to the four data form first row, the secondary series of second data to the five data of training sample data group being formed matrix, the 3rd data to the six data of training sample data group are formed the 3rd of matrix be listed as, and the like, up to forming 4 * (N _i-3) matrix;

4. the method for utilizing emulation function sim function in the MATLAB 6.5 tool box functions to carry out network test is: the test sample book data set is converted into 4 * (N _j-3) matrix carries out network test as the input parameter of sim function;

The test sample book data set is converted into 4 * (N _jThe concrete grammar of matrix-3) is: first data to the four data of test sample book data set are formed first row, the secondary series of second data to the five data of test sample book data set being formed matrix, the 3rd row of the 3rd data to the six data of test sample book data set being formed matrix, and the like, up to forming 4 * (N _j-3) matrix;

Step (3) uses the neural network model behind network training and network test and the sim function of MATLAB 6.5 tool box functions that next high tide level water level value or next low tide water level value are predicted, concrete grammar is: import the input parameter of last four groups data as the sim function in the sim function, obtain the predicted value of next group data after the sim function calculation, this predicted value is next high tide level water level value or next low tide water level value.